As you likely know, Perl 5 is fundamentally single threaded. While threads.pm allows for concurrency, it creates "heavy" threads that clone the entire interpreter state, leading to slow startup times and high memory consumption. For high-concurrency systems, the preferred primitive is the fiber or green threads.

In this chapter, we dissect Acme::Parataxis, a fiber system built with Affix that implements true fibers by performing low-level surgery on the Perl interpreter's internal state. My early goal for this was to demonstrate messing around with perl's internals from an affix'd shared library with the help of Affix::Build.

The Recipe

Interfacing with a fiber system requires binding to custom C code that understands the Perl C API.

use v5.40;
use Acme::Parataxis qw[:all];
$|++;
async {
    say 'Main task started';

    # 'fiber' is a shorter alias for 'spawn'
    my $f1 = fiber {
        say '  Task 1: Sleeping...';
        await_sleep(1000);
        return 42;
    };
    my $f2 = fiber {
        say '  Task 2: Performing I/O...';

        # ...
        return 'I/O Done';
    };

    # 'await' works on fibers and futures
    say 'Result 1: ' . await($f1);
    say 'Result 2: ' . await($f2);
};

How It Works

Acme::Parataxis is perhaps the most extreme example of what is possible with Affix. It doesn't just wrap a library; it redefines the execution flow of the Perl interpreter itself.

1. The Dual-Layer Switch

A fiber switch in Parataxis is a two-stage process performed in a custom C library bound via Affix.

Stage A: The OS Context Switch Affix binds to OS-specific primitives like SwitchToFiber (Windows) or swapcontext (Unix). This saves the CPU registers, including the Instruction Pointer and the Hardware Stack Pointer.

Stage B: The Perl Context Switch The C core defines a para_fiber_t struct that holds the entire state of a Perl interpreter thread. Acme::Parataxis uses Affix to bind a function that manually swaps over 20 global pointers found in the Perl core (PL_stack_sp, PL_markstack, PL_scopestack, etc.).

2. Architectural Inspiration: The Wren Model

The design of Acme::Parataxis is heavily inspired by the Wren scripting language. In Wren, fibers are the primary unit of execution and are designed to be...

• Cooperative: They only switch when explicitly yielding.
• Deterministic: Control flow is predictable, eliminating many "race condition" bugs common in preemptive threading.
• Lightweight: Each fiber has its own stack (512KB by default in Parataxis), allowing you to spawn thousands of them where you could only spawn dozens of OS threads.

Real-World Efficiency: Non-blocking HTTP

The true power of fibers is making asynchronous I/O look like simple synchronous code. By subclassing HTTP::Tiny and replacing its internal socket checks with await_read() and await_write(), we can perform multiple requests in parallel on a single thread:

async {
    my $http = My::CooperativeHTTP->new();
    my @urls = qw[http://perl.org http://metacpan.org];

    # Spawn concurrent requests
    my @futures = map { spawn { $http->get($_) } } @urls;

    # Wait for all to finish
    my @results = map { await $_ } @futures;
};

A functioning example using HTTP::Tiny in such a system comes bundled with the Acme::Parataxis dist.

Benchmarks: The Speed of Cooperation

Context switching in Acme::Parataxis happens at the nanosecond scale because the "trampoline" generated by Affix is optimized for your specific CPU architecture.

Benchmark: The Stress of 1,000 Fibers

This test measures the scheduler's ability to handle many short-lived fibers performing rapid yield/resume cycles.

use v5.40;
use Acme::Parataxis qw[:all];
use Time::HiRes     qw[time];
no warnings 'recursion';
async {
    my $total_fibers = 1000;
    my $iterations   = 5;
    say "Starting Stress Test: $total_fibers fibers, $iterations iterations...";
    my $start_time = time();
    for my $iter ( 1 .. $iterations ) {
        my $iter_start = time();
        my @futures;

        # Spawn a batch of 1,000 fibers
        for my $i ( 1 .. $total_fibers ) {
            push @futures, fiber {

                # Perform a tiny bit of CPU work
                my $val = 0;
                $val += $_ for 1 .. 100;

                # Cooperatively yield control halfway through
                yield() if $i % 2 == 0;
                return $val;
            };
        }

        # Await the completion of the entire batch
        await $_ for @futures;
        my $elapsed = time() - $iter_start;
        say sprintf '  Iteration %2d completed in %.4fs', $iter, $elapsed;
    }
    my $total_elapsed = time() - $start_time;
    say sprintf 'Total time: %.4fs', $total_elapsed;
    say 'Average time per fiber lifecycle: ' . ( $total_elapsed / ( $total_fibers * $iterations ) ) . 's';
};

In terms of throughput, fibers can outperform threads.pm for many tasks because they avoid the kernel overhead of OS-level scheduling.

Real-World Demo: Threaded I/O Offloading

The following demonstration shows how Acme::Parataxis handles blocking operations (like sleep or read) by offloading them to a native C thread pool while the main Perl thread continues to schedule other fibers.

use v5.40;
use Acme::Parataxis qw[:all];
use Time::HiRes     qw[time];
async {
    say 'Main: Initial thread pool size is ' . Acme::Parataxis::get_thread_pool_size();
    my @futures;
    for my $id ( 1 .. 5 ) {
        push @futures, fiber {
            my $sleep_time = int( rand(1000) ) + 500;
            say "  [Fiber $id] Will sleep for ${sleep_time}ms in background pool...";
            my $start = time();

            # This suspends the fiber, the C pool sleeps, and we are resumed when done.
            await_sleep($sleep_time);
            my $elapsed = int( ( time() - $start ) * 1000 );
            say "  [Fiber $id] Woke up after ${elapsed}ms!";
            return "Result $id";
        };
    }
    say 'Main: All fibers spawned. Total pool size: ' . Acme::Parataxis::get_thread_pool_size();

    # Wait for all workers to finish
    my @results = map { await $_ } @futures;
    say 'Main: All fibers finished!';
};

Kitchen Reminders

• Hybrid Power: Acme::Parataxis uses a "Hybrid" model. Perl code runs on one thread, but await_* calls offload blocking work to a native C thread pool.
• The Pointer[SV] Responsibility: When using Affix to pass raw Perl variables (SV*) to C fibers, you must manually manage refcounts. A single mistake here leads to leaks that standard Perl profilers cannot see.
• Cooperative Preemption: If a fiber runs a tight loop for too long, it can "starve" other fibers. Use maybe_yield() to allow the scheduler to breathe.

Frequent calls to malloc() and free() (or Perl's new and garbage collection) can become a significant bottleneck in performance-critical code. Each allocation is a request to the operating system or the memory manager, which involves searching for a free block, updating metadata, and potentially triggering a syscall.

A Memory Arena (also known as a Region or Zone allocator) is a technique where you allocate one large block of memory upfront and then "carve" out smaller pieces as needed. When you are done, you free the entire block at once.

The Recipe

We will build a simple Arena allocator in Perl using Affix's memory utilities. This allows us to create millions of small C-compatible structs with near-zero allocation overhead.

use v5.40;
use Affix qw[:all];

# Define a Struct for our data
typedef Point => Struct [ x => Int, y => Int ];
my $point_size = sizeof Point();

# Setup the arena
# We allocate 1MB upfront (enough for ~131,000 Points)
my $ARENA_SIZE     = 1024 * 1024;
my $arena_ptr      = malloc $ARENA_SIZE;
my $current_offset = 0;

# Allocation Helper
# Carves a block of memory from the arena
sub arena_alloc($size) {
    die 'Arena exhausted' if $current_offset + $size > $ARENA_SIZE;

    # Get the address of the current slot
    my $ptr = ptr_add( $arena_ptr, $current_offset );

    # Advance the pointer and return
    $current_offset += $size;
    return $ptr;
}

# Use the Arena
say 'Allocating 100,000 points...';
my $start_time = time();
my @points;
for ( 1 .. 100_000 ) {

    # Grab memory from the arena
    my $raw = arena_alloc($point_size);

    # Cast it to our Struct type (Zero-copy view)
    # We use Pointer[Point] so we get unified access
    my $p = cast $raw, Pointer [ Point() ];

    # Initialize
    $p->{x} = $_;
    $p->{y} = $_ * 2;

    # Store the pin (optional, we could just process and forget)
    push @points, $p;
}
say 'Finished in ' . ( time() - $start_time ) . ' seconds.';
say 'Point 500: x=' . $points[499]{x} . ', y=' . $points[499]{y};

# Cleanup
# Instead of 100,000 free() calls, we just do one!
free $arena_ptr;
undef @points;    # Pins are now invalid, but that's okay.

How It Works

• 1. Pre-allocation By calling malloc 1024 * 1024, we make a single request to the system. From that point on, every "allocation" we make via arena_alloc is just simple integer addition ($current_offset += $size). This is the fastest possible way to allocate memory.

• 2. ptr_add and cast

  my $raw = ptr_add($arena_ptr, $current_offset);
  my $p = cast $raw, Pointer[Point()];
  

  The ptr_add function returns a new unmanaged pin pointing to an offset within our big block. cast then tells Affix how to interpret the bytes at that address. No memory is copied during these steps.

• 3. Bulk Deallocation In a standard FFI script, you might rely on own($ptr, 1) to let Perl's GC free each struct individually. With an Arena, you set own($arena_ptr, 1) (which is the default for malloc) and when the arena itself goes out of scope, the entire 1MB block is reclaimed.

Kitchen Reminders

• Fragmentation Arenas are perfect for data that has the same lifetime. If you need to free specific items while keeping others, an Arena is the wrong tool (use standard malloc/free).

• Alignment In a real-world scenario, you must ensure that each "carved" piece starts on a valid alignment boundary (usually 4 or 8 bytes). You can achieve this with a simple bitwise mask:

  # Align to 8 bytes
  $current_offset = ($current_offset + 7) & ~7;
  

• Performance This technique is especially powerful when combined with direct_affix. If you pass your arena-allocated pointers to a JIT-optimized C function, the total overhead is virtually indistinguishable from native C.

Perl's threads.pm implements "heavy" threads, where each thread is a complete clone of the interpreter. While this model provides excellent isolation, it creates unique challenges when working with FFI. If you aren't careful, you can end up with clobbered memory or race conditions in the JIT engine.

This chapter teaches you how to safely manage Affix in a multithreaded environment by respecting the boundary between setup and execution.

The Recipe

We will build a simple worker pool where multiple Perl threads call into a single C mathematical function.

use v5.40;
use threads;
use Affix qw[:all];
use Affix::Build;

# PHASE 1: Initialization (Main Thread)
# All setup MUST happen before threads are spawned.
my $c = Affix::Build->new();
$c->add( \<<~'END', lang => 'c' );
    #include <math.h>
    double heavy_calc(double input) {
        // Simulate a CPU-heavy mathematical task
        return sin(input) * cos(input) + sqrt(input);
    }
END
my $lib = $c->link;

# Register the type and bind the function
affix $lib, 'heavy_calc', [Double] => Double;

# PHASE 2: Execution (Worker Threads)
sub worker ($id) {
    say "Worker $id: Starting on OS thread " . threads->tid();

    # Each thread can safely call the bound function
    for ( 1 .. 100_000 ) {
        my $res = heavy_calc( rand() );
    }
    return "Worker $id: Done";
}

# Spawn 4 worker threads
my @threads = map { threads->create( \&worker, $_ ) } 1 .. 4;

# Gather results
say $_->join() for @threads;
say 'All workers finished safely.';

How It Works

The Initialization Rule

The most important rule in Affix thread safety is: Initialize in the main thread.

Functions that modify Affix's internal global state are not thread-safe. This includes:

• load_library / locate_lib
• affix / wrap / direct_affix
• typedef

When threads->create is called, Perl clones the entire interpreter. Affix ensures that the trampolines and the registered types are available in the child threads. However, if you attempt to call affix inside a child thread, you risk corrupting the internal JIT registry.

Thread-Safe Trampolines

Once a function is bound (via affix or wrap), the resulting machine code is read-only and thread-safe. Multiple Perl threads can jump into the same C function simultaneously without locking.

Thread Context (THX)

When a C function is called, Affix handles the perl context (the aTHX macro in XS terms). Each thread has its own context and Affix ensures that when a C function returns, the data is marshalled back into the correct thread's stack.

Modern Callbacks and Foreign Threads

If a C library spawns its own native OS threads (not ithreads) and tries to trigger an Affix callback, Affix performs a context injection in an attempt to keep things flowing. It detects that the thread does not have a Perl interpreter attached and attempts to provide a temporary, safe environment for the Perl subroutine to run. This is what allows Affix to work with GUI event loops and audio drivers.

Kitchen Reminders

• C Library Safety: Just because Affix is thread-safe doesn't mean the library you are wrapping is. If the C function uses static variables or non-atomic global state, you must use Thread::Semaphore in Perl to gate access.
• Shared Memory: Pointers created with malloc are just memory addresses. You can pass them between threads, but Perl's threads.pm does not automatically synchronize access to that memory. Use memcpy or memset carefully!
• Memory Leaks: If you allocate memory in a thread using malloc, ensure it is either freed in that thread or the ownership is clearly defined. Perl's garbage collector only cleans up managed pins when their specific thread-local scalar goes out of scope.

C is often called "portable assembly," but sometimes you need the real thing.

If you need to read the CPU Time Stamp Counter (RDTSC) for nanosecond-precision benchmarking, access specific processor flags, or utilize SIMD instructions not exposed by your C compiler, you can write raw assembly and call it directly from Perl.

The Recipe

This time, we'll implement a function to read the CPU cycle count (RDTSC) on x86-64 Linux/Unix/Windows systems.

use v5.40;
use Affix qw[:all];
use Affix::Build;

# 1. Write Assembly
# We use AT&T syntax (standard for GCC/Clang on Unix).
# This file (.s) will be passed to the assembler.
my $c = Affix::Build->new( name => 'asm_lib' );
$c->add( \<<~'END', lang => 's' );
    .text
    .global get_ticks

    get_ticks:
        # rdtsc puts 32 bits in EDX (high) and 32 bits in EAX (low)
        rdtsc

        # We need to pack them into a single 64-bit register (RAX)
        # Shift RDX left by 32 bits
        shl $32, %rdx

        # Combine RDX and RAX
        or %rdx, %rax

        # Return value is in RAX
        ret
END

# 2. Link
# Note: On macOS, symbols need a leading underscore ('_get_ticks').
my $lib = $c->link;

# 3. Bind
# The function takes no args and returns a 64-bit unsigned integer.
affix $lib, 'get_ticks', [] => UInt64;

# 4. Benchmark
say 'Measuring CPU cycles for a Perl operation...';
my $start = get_ticks();

# Do some work
my $x = 0;
$x += $_ for 1 .. 1000;

# Check the clock again
my $end = get_ticks();
say 'Start: ' . $start;
say 'End:   ' . $end;
say 'Diff:  ' . ( $end - $start ) . ' cycles';

On my system, the output looks like this:

Measuring CPU cycles for a Perl operation...
Start: 4098233110110909
End:   4098233110188467
Diff:  77558 cycles

How It Works

• 1. The Assembler When you pass lang => 's', Affix::Build invokes the system assembler (usually as or gcc). It compiles the raw instructions into machine code.

• 2. The ABI You must respect the Calling Convention of your platform (System V AMD64 in this example).

  • Arguments: RDI, RSI, RDX...
  • Return Values: RAX
  • Safety: You must preserve "callee-saved" registers (RBX, RBP, R12-R15) if you use them. Since rdtsc only clobbers RAX and RDX (which are caller-saved/volatile), we don't need to push/pop anything.

Kitchen Reminders

• Architecture Specific Assembly is not portable. This code will crash on ARM64 (Apple Silicon, Raspberry Pi) or 32-bit systems. You must detect $Config{archname} and provide different assembly implementations for different CPUs.

• Syntax Flavors

  • Unix/Linux: Uses AT&T syntax (%reg, src, dest).
  • Windows: Uses Intel syntax (reg, dest, src) via NASM. If you are targeting Windows and Intel, you should use lang => 'asm' (NASM) instead of 's'.

If you are doing heavy scientific computing, you will eventually encounter Fortran (BLAS, LAPACK).

Binding Fortran can be tricky because:

1. Name Mangling: Compilers often lowercase names and append underscores (e.g., DGEMM becomes dgemm_).
2. Pass-by-Reference: Historically, Fortran passes everything by reference (pointers).
3. Strings: Legacy Fortran expects string pointers to be followed by hidden length arguments at the end of the argument list.

However, modern Fortran (2003+) offers the iso_c_binding module, which allows us to define a standard C ABI.

The Recipe

This time, we'll compile a Fortran subroutine that adds two numbers, and another that prints a string based on an explicit length argument.

use v5.40;
use Affix qw[:all];
use Affix::Build;
$|++;

# 1. Compile Fortran
# Requires a Fortran compiler be installed
my $c = Affix::Build->new( name => 'fortran_demo' );
$c->add( \<<~'END', lang => 'f90' );
        ! Numeric subroutine: Passes pointers by default
        subroutine f_add(a, b, res) bind(c, name='f_add')
            use iso_c_binding
            integer(c_int), intent(in) :: a, b
            integer(c_int), intent(out) :: res
            res = a + b
        end subroutine

        ! String subroutine: Explicit length handling
        subroutine f_hello(msg, len) bind(c, name='f_hello')
            use iso_c_binding
            character(kind=c_char, len=1), intent(in) :: msg(*)
            integer(c_int), value :: len

            ! Print the slice using the passed length
            print *, "Fortran says: ", msg(1:len)
        end subroutine
    END
my $lib = $c->link;

# 2. Bind
# Modern Fortran with bind(c) respects C names.
# However, Fortran arguments are usually pointers.
affix $lib, 'f_add', [ Pointer [Int], Pointer [Int], Pointer [Int] ] => Void;

# For strings, we pass the buffer (String) and the length (Int) explicitly.
affix $lib, 'f_hello', [ String, Int ] => Void;

# 3. Call (Math)
# We must pass references (\$) to our numbers because Fortran expects pointers.
my $a   = 10;
my $b   = 20;
my $res = 0;
f_add( \$a, \$b, \$res );
say "10 + 20 = $res";    # 30

# 4. Call (Strings)
my $msg = 'Hello World';

# We pass the string string (char*), AND the length manually.
f_hello( $msg, length($msg) );

How It Works

• 1. iso_c_binding This standard module is the key to sanity. By adding bind(c, name='f_add'), we force the Fortran compiler to generate a symbol named f_add (no trailing underscores) that uses the standard C calling convention.

• 2. Pointers vs Values In f_add, the arguments a and b do not have the value attribute. This means Fortran expects memory addresses. In Perl, we handle this by passing references (\$a). In f_hello, len is defined with value. This means Fortran expects a raw integer, so we pass it directly from Perl.

Kitchen Reminders

• Runtime Dependencies Shared libraries compiled with gfortran depend on libgfortran. On Linux, this is usually automatic. On Windows (MinGW), if you ship your compiled DLL to another machine, you must ensure libgfortran-*.dll is also in the PATH, or the library will fail to load.

• Hidden Arguments If you are binding to legacy Fortran (BLAS/LAPACK) that was not written with iso_c_binding, you must account for "Hidden Arguments." For every string argument in the signature, you must usually append a Size_t argument to the very end of the argument list containing the string length.

Perl is great for glue; Rust is great for logic.

Historically, binding Rust to Perl required complex XS modules or specific crates like perl-xs. With Affix::Build, you can treat Rust just like C. By using standard ABI hooks (extern "C"), we can compile and run Rust code dynamically without ever leaving the Perl script.

The Recipe

Let's define a simple Rust library that operates on a Struct and handles strings.

use v5.40;
use Affix qw[:all];
use Affix::Build;
$|++;

# 1. Compile Rust
# Affix::Build detects the 'rs' extension and invokes `rustc`.
my $c = Affix::Build->new( name => 'rust_demo' );

# We define a struct to share between Rust and Perl
$c->add( \<<~'RUST', lang => 'rs' );
    use std::ffi::CStr;
    use std::os::raw::c_char;

    // We must define the struct layout to match C
    #[repr(C)]
    pub struct Point {
        x: i32,
        y: i32,
    }

    // #[no_mangle] keeps the symbol name "add_points"
    // extern "C" ensures the ABI uses standard registers
    #[no_mangle]
    pub extern "C" fn add_points(a: Point, b: Point) -> Point {
        Point {
            x: a.x + b.x,
            y: a.y + b.y
        }
    }

    // Example of reading a Perl string
    // We accept a raw pointer (*const c_char)
    #[no_mangle]
    pub unsafe extern "C" fn hello_rust(name: *const c_char) {
        if !name.is_null() {
            let c_str = CStr::from_ptr(name);
            if let Ok(s) = c_str.to_str() {
                println!("Hello, {} from Rust!", s);
            }
        }
    }
RUST
my $lib = $c->link;

# 2. Define Types
typedef Point => Struct [ x => Int, y => Int ];

# 3. Bind
# We treat Rust structs exactly like C structs.
affix $lib, 'add_points', [ Point(), Point() ] => Point();
affix $lib, 'hello_rust', [String]             => Void;

# 4. Execute
my $p1 = { x => 10, y => 20 };
my $p2 = { x => 5,  y => 5 };

# Pass by value
my $p3 = add_points( $p1, $p2 );
say "Result: $p3->{x}, $p3->{y}";    # 15, 25

# Pass string
hello_rust('Perl');

How It Works

• 1. #[repr(C)] Rust creates its own memory layout for structs by default, which usually does not match C (or Affix). Adding this attribute forces the compiler to lay out the struct exactly like a C struct, ensuring Affix can write to fields x and y correctly.

• 2. #[no_mangle] and extern "C" Just like C++, Rust mangles function names (e.g., _ZN7example3add...) to support namespaces and generics.

  • #[no_mangle] turns off name mangling, keeping the symbol as add_points.
  • extern "C" sets the calling convention to the system standard (cdecl/SystemV).

• 3. Pointers and Unsafe When accepting pointers from Perl (like the string in hello_rust), Rust requires an unsafe block to dereference them. Affix handles the memory management of the string buffer (ensuring it is NULL-terminated); Rust just reads it.

Kitchen Reminders

• Panics If your Rust code panics (crashes), it may "unwind" the stack across the FFI boundary. This is undefined behavior and will likely abort the Perl interpreter immediately. Pro Tip: Use std::panic::catch_unwind inside your extern "C" functions to catch Rust errors and return a clean error code to Perl instead of crashing.

• Windows MinGW If you are using Strawberry Perl, Affix::Build automatically handles the ABI mismatch between Rust (which defaults to MSVC on Windows) and Perl (which uses GCC/MinGW), ensuring the libraries link correctly.

In previous chapters, we compiled code directly or used Alien::Base to find system libraries. But sometimes you want a library that isn't installed on the system, isn't on CPAN, but is available in the wider C/C++ ecosystem.

This is where xmake and its package manager, xrepo, shine.

xrepo is a modern, cross-platform C/C++ package manager. It can download, compile, and install thousands of libraries (like libpng, zlib, pcre, ffmpeg) with a single command.

We have created a helper module, Alien::Xrepo, to bridge the gap. It asks xrepo to install a library, finds the resulting shared object (.dll / .so), and hands it to Affix::Wrap for binding.

The Recipe

We will install libpng using Alien::Xrepo and use it to generate a gradient image file. We'll set up the script so you can easily define the size and color range. The current config produces a PNG that looks like this:

use v5.40;
use Affix::Wrap;
use Alien::Xrepo;
use List::Util qw[min max];
use constant PI => 3.14159265359;

# Configuration
my $width  = 400;
my $height = 400;
my $angle  = 135;    # Degrees (0 = Horizontal, 90 = Vertical)

# Define gradient stops (Red -> Yellow -> Green -> Blue)
my @stops = (
    [ 255, 0,   0 ],     # Red
    [ 255, 255, 0 ],     # Yellow
    [ 0,   255, 0 ],     # Green
    [ 0,   0,   255 ]    # Blue
);

# 1. Initialize the Repo Helper
my $repo = Alien::Xrepo->new( verbose => 1 );

# 2. Install Library
# This runs `xrepo install -k shared libpng`
# It downloads source, compiles a DLL/SO, and returns the paths.
my $pkg = $repo->install('libpng');
say 'Found libpng version ' . $pkg->version . ' at ' . $pkg->libpath;

# 3. Locate Header
# We need the full path to png.h to parse function signatures.
my $header = $pkg->find_header('png.h');

# 4. Wrap
# We must pass include_dirs so Affix::Wrap's parser can find zlib (which libpng depends on).
my $wrapper = Affix::Wrap->new( project_files => [$header], include_dirs => $pkg->includedirs );

# We define a helper type that png.h relies on but doesn't define clearly for FFI
$wrapper->wrap( $pkg->libpath );

# 5. Use the Library
# We will use the high-level API of libpng to write a file.
# A. Generate Data (Math!)
my $buffer = '';

# Convert angle to vector
my $rad = $angle * PI / 180;
my $dx  = cos($rad);
my $dy  = sin($rad);

# Calculate projection bounds to normalize the gradient
# We project the 4 corners of the image onto the gradient vector
my @corners = ( 0 * $dx + 0 * $dy, $width * $dx + 0 * $dy, 0 * $dx + $height * $dy, $width * $dx + $height * $dy );
my $min_p   = min @corners;
my $max_p   = max @corners;
my $dist    = $max_p - $min_p || 1;         # Avoid division by zero
for my $y ( 0 .. $height - 1 ) {
    for my $x ( 0 .. $width - 1 ) {
        my $p = $x * $dx + $y * $dy;        # Project pixel onto gradient line
        my $t = ( $p - $min_p ) / $dist;    # Normalize to 0.0 - 1.0 range

        # Find which two colors we are blending between
        # Map t (0..1) to index (0 .. #stops-1)
        my $pos     = $t * $#stops;
        my $idx     = int $pos;       # Lower color index
        my $local_t = $pos - $idx;    # Blend factor between lower and upper

        # Clamp for safety at the very edge
        if ( $idx >= $#stops ) { $idx = $#stops - 1; $local_t = 1.0; }
        my $c1 = $stops[$idx];
        my $c2 = $stops[ $idx + 1 ];

        # Lerp
        my $r = int( $c1->[0] + ( $c2->[0] - $c1->[0] ) * $local_t );
        my $g = int( $c1->[1] + ( $c2->[1] - $c1->[1] ) * $local_t );
        my $b = int( $c1->[2] + ( $c2->[2] - $c1->[2] ) * $local_t );
        $buffer .= pack( 'CCC', $r, $g, $b );
    }
}

# B. Setup Struct
# Affix lets us pass a Perl HashRef for the C struct 'png_image'.
my $image = {
    version => PNG_IMAGE_VERSION(),    # Macro wrapped by Affix
    width   => $width,
    height  => $height,
    format  => PNG_FORMAT_RGB(),       # Macro wrapped by Affix
    flags   => 0,
    opaque  => undef
};
say 'Writing to test_affix.png...';

# C. Call Function
# int png_image_write_to_file(png_imagep image, const char *file, ...);
my $result = png_image_write_to_file(
    $image, 'test_affix.png', 0,    # convert_to_8bit (auto)
    $buffer,                        # Raw pixel data
    0,                              # row_stride (auto)
    undef                           # colormap
);
#
say $result ? 'SUCCESS: Image written.' : 'FAILURE: ' . $image->{message};

How It Works

• $repo->install('libpng') This command invokes the external xrepo tool. It checks if libpng is cached. If not, it downloads the source, compiles it as a shared library (critical for FFI), and returns an object containing the paths to the binary and headers.

• $pkg->includedirs Libraries often depend on other libraries (e.g., libpng needs zlib). xrepo handles this dependency tree. When we pass these directories to Affix::Wrap, we ensure that the Clang or Regex parser can find zlib.h when it parses png.h.

• The Math While we use C for the I/O, we kept the pixel logic in Perl. We use a Linear Projection to calculate where every pixel falls along the angled line defined by $angle. We then map that position to our array of @stops and interpolate the RGB values. This demonstrates that Perl is more than capable of handling complex logic, while Affix handles the heavy lifting of interfacing with system libraries.

Why use Alien::Xrepo?

1. Cross-Platform: xmake works on Windows, Linux, and macOS.
2. Huge Repository: It supports thousands of libraries via conan, vcpkg, brew, and its own repository.
3. Compilation: Unlike system package managers (apt/yum) which often only provide static .a files for development, xrepo can be forced to build .dll / .so files, which is exactly what dynamic FFI needs.

This approach gives you the power of CPAN's Alien:: namespace but with access to the entire C/C++ open source ecosystem instantly.

In the previous chapter, we built a script that compiles its own dependencies. To share this on CPAN, we want a standard module structure.

Most modern Perl developers use Module::Build to manage the installation of their distributions. Module::Build can build an XS based module automatically but has no idea how to build with Affix::Build. To work around this, we must create a custom builder class to hook into the build process and compile our C library exactly when the user runs ./Build.

We're going to use Minilla to mint our dists. A complete copy of this demo can be found here on github.

1. The Setup

Initialize your project:

minil new Acme-Image-Stb
cd Acme-Image-Stb
mkdir builder

2. Configuration (minil.toml)

We need to tell Minilla two things:

1. Use Module::Build (instead of Module::Build::Tiny).
2. Use our custom class (builder::MyBuilder) to drive the process.

Edit minil.toml:

name = "Acme-Image-Stb"
badges = ["github-actions/test.yml"]
module_maker="ModuleBuild"
static_install = "auto"

[build]
build_class = "builder::MyBuilder"

3. The Custom Builder (builder/MyBuilder.pm)

This is where the magic happens. We subclass Module::Build and override the ACTION_code method. This method runs when the user types ./Build.

We use Affix::Build to build the library and place it directly into the staging directory (blib/arch), ensuring it gets installed correctly.

package builder::MyBuilder;
use v5.40;
use parent 'Module::Build';
use Affix::Build;
use HTTP::Tiny;
use Path::Tiny;
use Config;

sub ACTION_code ($self) {
    unless ( defined $self->config_data('lib') ) {
        say 'Building embedded C library...';

        # Setup source directory
        my $src_dir = path('_src');
        $src_dir->mkpath;

        # Download headers if missing. In reality, you would probably just bundle the headers with the dist.
        # But this isn't reality.
        my $http = HTTP::Tiny->new;
        for my $file (qw(stb_image.h stb_image_resize2.h stb_image_write.h)) {
            my $path = $src_dir->child($file);
            next if $path->exists;
            say "  Fetching $file...";
            my $res = $http->get("https://raw.githubusercontent.com/nothings/stb/master/$file");
            die "Failed to download $file" unless $res->{success};
            $path->spew_raw( $res->{content} );
        }

        # Determine output path
        # We want the DLL to end up in: blib/arch/auto/Acme/Image/Stb/stb.so.xx.xx
        # This ensures it is installed in the architecture-specific library path.
        my $dist_name = $self->dist_name;    # "Acme-Image-Stb"
        my @parts     = split /-/, $dist_name;
        my $arch_dir  = path( $self->blib, 'arch', 'auto', @parts );
        $arch_dir->mkpath;

        # Compile with Affix
        my $c = Affix::Build->new(
            version   => $self->dist_version,
            name      => 'stb',
            build_dir => $arch_dir,
            flags     => { cflags => "-I$src_dir -O3", ldflags => ( $^O eq 'MSWin32' ? '-Wl,--export-all-symbols' : '' ) }
        );
        $c->add( \<<~'C', lang => 'c' );
        #if defined(_WIN32)
          #define STBIDEF __declspec(dllexport)
          #define STBIWDEF __declspec(dllexport)
          #define STBIRDEF __declspec(dllexport)
        #else
          #define STBIDEF __attribute__((visibility("default")))
          #define STBIWDEF __attribute__((visibility("default")))
          #define STBIRDEF __attribute__((visibility("default")))
        #endif

        #define STB_IMAGE_IMPLEMENTATION
        #define STB_IMAGE_WRITE_IMPLEMENTATION
        #define STB_IMAGE_RESIZE_IMPLEMENTATION

        #include "stb_image.h"
        #include "stb_image_resize2.h"
        #include "stb_image_write.h"
    C
        my $lib_file = $c->link;
        say "  Compiled: $lib_file";
        $self->config_data( lib => $lib_file->basename );
    }

    # Run standard build steps (copying .pm files to blib/)
    $self->SUPER::ACTION_code;
}
1;

4. Dependencies (cpanfile)

We need to add our build tools to the configure phase.

requires 'perl', '5.040';
requires 'Affix', 'v1.0.3';

on 'configure' => sub {
    requires 'Module::Build';
    requires 'HTTP::Tiny';
    requires 'Path::Tiny';
    requires 'Affix::Build', 'v1.0.3';
};

on 'test' => sub {
    requires 'Test2::V0';
};

5. The Module (lib/Acme/Image/Stb.pm)

Since we installed the library into blib/arch/auto/..., we can't just look in the current directory. We need to look where Perl installs architecture-specific files (XS modules usually live here).

We use a helper to scan @INC.

package Acme::Image::Stb 0.01 {
    use v5.40;
    use Affix qw[:all];
    use Carp  qw[croak];
    use Config;
    use Acme::Image::Stb::ConfigData;
    use parent 'Exporter';
    our %EXPORT_TAGS = ( internals => [qw[stbi_load stbi_write_png stbir_resize_uint8_linear]], core => [qw[load_and_resize]] );
    $EXPORT_TAGS{all} = [ our @EXPORT_OK = sort map {@$_} values %EXPORT_TAGS ];

    # Locate Library
    # We look for: .../auto/Image/Stb/stb.so (or .dll)
    my $lib_name = Acme::Image::Stb::ConfigData->config('lib');
    my $lib_path;
    for my $dir (@INC) {
        my $check = "$dir/auto/Acme/Image/Stb/$lib_name";
        if ( -e $check ) {
            $lib_path = $check;
            last;
        }
    }
    croak "Could not find compiled library '$lib_name' in \@INC" unless $lib_path;
    my $lib = Affix::load_library($lib_path);

    # Bindings
    use constant STBIR_RGBA => 4;
    affix $lib, 'stbi_load',                 [ String, Pointer [Int], Pointer [Int], Pointer [Int], Int ] => Buffer;
    affix $lib, 'stbi_write_png',            [ String, Int, Int, Int, Buffer, Int ]                       => Int;
    affix $lib, 'stbir_resize_uint8_linear', [ Buffer, Int, Int, Int, Buffer, Int, Int, Int, Int ]        => Buffer;

    # API
    sub load_and_resize ( $input, $output, $scale ) {
        my ( $w, $h, $ch ) = ( 0, 0, 0 );

        # Load
        my $img = stbi_load( $input, \$w, \$h, \$ch, 4 );
        return undef if is_null($img);

        # Calculate
        my $nw      = int( $w * $scale );
        my $nh      = int( $h * $scale );
        my $out_buf = "\0" x ( $nw * $nh * 4 );

        # Resize
        my $res = stbir_resize_uint8_linear( $img, $w, $h, 0, $out_buf, $nw, $nh, 0, STBIR_RGBA );
        return undef if is_null($res);

        # Save
        stbi_write_png( $output, $nw, $nh, 4, $out_buf, 0 );
    }
}
1;
__END__

=pod

=encoding utf-8

=head1 NAME

Acme::Image::Stb - It's new $module

=head1 SYNOPSIS

    use Acme::Image::Stb;

=head1 DESCRIPTION

Acme::Image::Stb is ...

=head1 LICENSE

Copyright (C) You.

This library might be free software; you may or may not be able to redistribute it
and/or modify it under the same terms of the author.

=head1 AUTHOR

You E<lt>yournamehere@cpan.orgE<gt>

=cut

6. Development Workflow

With this setup, the standard Minilla commands work perfectly, but now they trigger a C compiler in the background.

# Install dependencies
$ cpanm --installdeps .

# Build and test
$ minil test

# Make the dist
$ minil dist

Kitchen Reminders

By subclassing Module::Build, we have seamlessly integrated Affix::Build into the standard Perl toolchain.

1. Transparency: Users installing via cpanm won't know you aren't using XS. It just works.
2. Correctness: By targeting blib/arch/auto, we respect Perl's directory structure for binary binaries. This demo does it in a rather unsophisticated way...
3. Flexibility: You can add logic to MyBuilder.pm to detect system libraries (via pkg-config or xrepo) and only compile the bundled version if the system version is missing.

Pat yourself on the back. You have now created a binary distribution ready for CPAN without writing a single line of XS.

For our next trick, we'll tackle the most common headache in the C ecosystem: Dependency Hell. Usually, if you want to manipulate images, you need libpng, libjpeg, zlib, and headers for all of them installed on your system. If your user is on a different OS, your script fails. To avoid that, let's use Affix::Build to compile the famous stb single-header libraries.

The Recipe

Here we'll load a JPG/PNG, resize it to a thumbnail using high-quality linear scaling, and save it as a PNG.

In other words, we'll turn this:

...into this:

Simple enough.

This recipe demonstrates two advanced techniques:

1. Output Parameters: Passing pointers to variables so C can populate them with data.
2. Raw Buffers: Passing the memory address of Perl scalars to C for high-performance data manipulation.

use v5.40;
use Affix qw[:all];
use Affix::Build;
use HTTP::Tiny;
use Path::Tiny;
use Config;
$|++;

# 1. Setup Persistent Cache
# Like Inline::C, we keep artifacts in a local subdirectory.
my $build_dir = path('.')->child('_build');
$build_dir->mkpath;

# Calculate the expected library name (e.g., stb_lib.dll or libstb_lib.so)
my $lib_file = $build_dir->child( ( $^O eq 'MSWin32' ? '' : 'lib' ) . 'stb_lib.' . $Config{so} );

# 2. Check Cache
if ( $lib_file->exists ) {
    say "Using cached library: $lib_file";
}
else {
    say "Library not found. Building...";
    build_library($build_dir);
}

# 3. Bindings
my $lib = Affix::load_library($lib_file);

# stbir_pixel_layout enum
use constant STBIR_RGBA => 4;

# "Buffer" passes the raw memory address of a Perl string to C.
affix $lib, 'stbi_load',                 [ String, Pointer [Int], Pointer [Int], Pointer [Int], Int ] => Buffer;
affix $lib, 'stbi_write_png',            [ String, Int, Int, Int, Buffer, Int ]                       => Int;
affix $lib, 'stbir_resize_uint8_linear', [ Buffer, Int, Int, Int, Buffer, Int, Int, Int, Int ]        => Buffer;

# 4. The Application
my $input_file  = 'avatar.jpg';
my $output_file = 'avatar_small.png';

# Generate dummy input if needed
unless ( -e $input_file ) {
    say 'Creating dummy input file...';
    my $w      = 100;
    my $h      = 100;
    my $pixels = '';

    # Gradient Pattern
    for my $y ( 0 .. $h - 1 ) {
        for my $x ( 0 .. $w - 1 ) {
            $pixels .= pack( 'C4', int( $x * 2.5 ), int( $y * 2.5 ), 0, 255 );
        }
    }
    stbi_write_png( $input_file, $w, $h, 4, $pixels, $w * 4 );
}

# Step A: Load
my ( $w, $h, $ch ) = ( 0, 0, 0 );
say "Loading $input_file...";
my $img_data = stbi_load( $input_file, \$w, \$h, \$ch, 4 );
die 'Failed to load image.' if is_null($img_data);
say "Original: ${w}x${h} ($ch channels)";

# Step B: Resize
my $new_w      = int( $w / 2 );
my $new_h      = int( $h / 2 );
my $in_stride  = $w * 4;
my $out_stride = $new_w * 4;

# Allocate output memory (Buffer)
my $out_data = "\0" x ( $new_h * $out_stride );
say "Resizing to ${new_w}x${new_h}...";
my $res = stbir_resize_uint8_linear( $img_data, $w, $h, $in_stride, $out_data, $new_w, $new_h, $out_stride, STBIR_RGBA );
die 'Resize failed!' if is_null($res);

# Step C: Save
say "Saving to $output_file...";
stbi_write_png( $output_file, $new_w, $new_h, 4, $out_data, $out_stride );
say 'Done.';

# Subroutines
sub build_library($dir) {
    my $http = HTTP::Tiny->new;

    # Download headers if missing
    for my $file (qw(stb_image.h stb_image_resize2.h stb_image_write.h)) {
        my $path = $dir->child($file);
        next if $path->exists;
        say "Downloading $file...";
        my $url = "https://raw.githubusercontent.com/nothings/stb/master/$file";
        my $res = $http->get($url);
        die "Failed to download $file" unless $res->{success};
        $path->spew_raw( $res->{content} );
    }
    my $c = Affix::Build->new(
        name      => 'stb_lib',
        build_dir => $dir,
        flags     => { cflags => "-I$dir -O3", ldflags => ( $^O eq 'MSWin32' ? '-Wl,--export-all-symbols' : '' ) }
    );
    $c->add( \<<~'C', lang => 'c' );
        #if defined(_WIN32)
          #define STBIDEF __declspec(dllexport)
          #define STBIWDEF __declspec(dllexport)
          #define STBIRDEF __declspec(dllexport)
        #else
          #define STBIDEF __attribute__((visibility("default")))
          #define STBIWDEF __attribute__((visibility("default")))
          #define STBIRDEF __attribute__((visibility("default")))
        #endif

        #define STB_IMAGE_IMPLEMENTATION
        #define STB_IMAGE_WRITE_IMPLEMENTATION
        #define STB_IMAGE_RESIZE_IMPLEMENTATION

        #include "stb_image.h"
        #include "stb_image_resize2.h"
        #include "stb_image_write.h"
    C
    $c->compile_and_link;
}

How It Works

• 1. The Buffer Type We defined Buffer type in Affix to handle passing an opaque pointer. Perl strings, when passed as a Pointer type, usually copy the data. However, for Pointer[Void], Affix optimizations kick in: it passes the raw address of the Perl scalar's memory. This allows stbir_resize to write directly into $out_data without any copy overhead.

• 2. Output Parameters (int *width) When calling stbi_load, we pass references (\$w, \$h). Affix detects this, allocates integer pointers, passes them to C, and then updates your Perl variables with the values C wrote (the image dimensions).

• 3. The Pixel Layout The stb_image_resize2 library requires us to specify the memory layout of the pixels. We pass STBIR_RGBA (4) to tell it that every 4 bytes represent Red, Green, Blue, and Alpha. If we passed 0 (default), it would treat the image as Grayscale, resulting in a garbled output.

Standard Perl scalars are designed for flexibility, not raw math throughput. When you need to process millions of coordinates, pixels, or audio samples, you want SIMD (Single Instruction, Multiple Data).

Affix makes SIMD vectors (like __m128, __m256, float32x4_t) first-class citizens and properly manages support on both x84 and ARM64.

The Recipe

Let's write a hypothetical 3D math library that uses 128-bit vectors (4 floats).

use v5.40;
use Affix qw[:all];
use Affix::Build;

# 1. Compile C Library
# We use GCC/Clang vector extensions for this example.
my $c = Affix::Build->new( flags => { cflags => '-O3 -mavx' } );
$c->add( \<<~'END', lang => 'c', );
    #if defined(__GNUC__) || defined(__clang__)
        typedef float v4f __attribute__((vector_size(16)));

        // Add two vectors
        v4f add_vecs(v4f a, v4f b) {
            return a + b;
        }

        // Scale a vector by a scalar
        v4f scale_vec(v4f v, float s) {
            return v * s;
        }

        // Dot product (returns scalar)
        float dot_vec(v4f a, v4f b) {
            v4f res = a * b;
            return res[0] + res[1] + res[2] + res[3];
        }
    #endif
END
my $lib = $c->link;

# 2. Bind
# We define a generic "Vec4" type: v[4:float]
typedef Vec4 => Vector [ 4, Float ];
affix $lib, 'add_vecs',  [ Vec4(), Vec4() ] => Vec4();
affix $lib, 'scale_vec', [ Vec4(), Float ]  => Vec4();
affix $lib, 'dot_vec',   [ Vec4(), Vec4() ] => Float;

# 3. Use (Packed Strings - The Fast Way)
# Pack 4 floats into a 16-byte string
my $v1 = pack( 'f4', 1.0, 2.0, 3.0, 4.0 );
my $v2 = pack( 'f4', 5.0, 6.0, 7.0, 8.0 );

# Add
my $sum_ref = add_vecs( $v1, $v2 );

# Results come back as ArrayRefs by default (for convenience)
say 'Sum: ' . join ', ', @$sum_ref;    # 6, 8, 10, 12

# 4. Use (ArrayRefs - The Convenient Way)
# You can also pass ArrayRefs directly. Affix handles the packing.
my $scaled_ref = scale_vec( [ 1.0, 1.0, 1.0, 1.0 ], 0.5 );
say 'Scaled: ' . join( ', ', @$scaled_ref );    # 0.5, 0.5, 0.5, 0.5

# 5. Dot Product
my $dot = dot_vec( $v1, $v2 );
say 'Dot Product: ' . $dot;                     # 1*5 + 2*6 + 3*7 + 4*8 = 70

How It Works

• 1. First-Class Types

  Vector[ 4, Float ]
  

  Affix understands that this is not just an array, but a specific machine type (128-bit register). On x64, this maps to __m128 and is passed in SSE registers (XMM0, etc.). On ARM64, it maps to NEON registers (V0, etc.).

• 2. Packed String Input

  my $v1 = pack( 'f4', ... );
  

  This is the most efficient way to work. You are creating the raw binary layout of the vector in a Perl scalar. Affix simply passes the pointer to this data directly to the C function (or loads it into a register). This avoids the overhead of iterating over a Perl array and converting individual values.

• 3. ArrayRef Output

  By default, Affix unpacks returned vectors into a Perl ArrayRef (e.g., [x, y, z, w]). This is convenient for debugging or light math.

Kitchen Reminders

• Alignment

  SIMD instructions are picky about alignment. When you use pack, Perl usually handles the string buffer allocation, but it doesn't guarantee 16-byte alignment. Affix handles the move to the register safely. If you use malloc to store vectors for C functions, be sure to use an aligned allocation if the C library requires it.

• Aliases

  Affix exports aliases for common SIMD types if you import :types:

  • M128 -> Vector[4, Float]
  • M128d -> Vector[2, Double]
  • M256 -> Vector[8, Float] (AVX)

So, you read Chapters 22 and 23, but the ultimate question remains: "Is it fast?"

To find out, we'll benchmark Affix against the two titans of Perl performance:

1. PDL (Perl Data Language) The standard for high-performance numerical computing in Perl. It is heavily optimized for processing large arrays ("piddles") in C.

2. Inline::C The traditional way to write C extensions. It compiles C code on the fly and links it to Perl via XS. It is generally considered the speed limit of Perl.

Benchmark 1: SIMD Vector Addition

This set of benchmarks tests using 128-bit SIMD vectors (adding four floats at once):

1. Micro-Benchmark: Call a function many times (measuring call overhead).
2. Macro-Benchmark: Call a function once to process 100,000 items (measuring raw throughput).

use v5.40;
use Benchmark qw[cmpthese];
use Affix     qw[:all];
use Affix::Compiler;
use PDL;
use Inline C => Config => ( CCFLAGSEX => '-O3 -mavx' );
use Inline C => <<~'END_XS';
    typedef float v4f __attribute__((vector_size(16)));

    // Micro: Add single vector
    SV* inline_add(SV* a_sv, SV* b_sv) {
        STRLEN len_a, len_b;
        float* a = (float*)SvPVbyte(a_sv, len_a);
        float* b = (float*)SvPVbyte(b_sv, len_b);
        if (len_a < 16 || len_b < 16) croak("Bad length");
        v4f vc = *(v4f*)a + *(v4f*)b;
        return newSVpvn((const char*)&vc, 16);
    }

    // Macro: Add arrays of vectors
    void inline_bulk(SV* a_sv, SV* b_sv, SV* out_sv, int count) {
        v4f* a = (v4f*)SvPV_nolen(a_sv);
        v4f* b = (v4f*)SvPV_nolen(b_sv);

        SvUPGRADE(out_sv, SVt_PV);
        SvGROW(out_sv, count * 16 + 1);
        SvCUR_set(out_sv, count * 16);
        v4f* out = (v4f*)SvPV_nolen(out_sv);

        for(int i=0; i<count; i++) {
            out[i] = a[i] + b[i];
        }
    }
END_XS

# Compile C Library for Affix
my $c = Affix::Compiler->new( flags => { cflags => '-O3 -mavx' } );
$c->add( \<<~'END', lang => 'c' );
    typedef float v4f __attribute__((vector_size(16)));

    v4f affix_add(v4f a, v4f b) {
        return a + b;
    }

    void affix_bulk(v4f *a, v4f *b, v4f *out, int count) {
        for(int i=0; i<count; i++) {
            out[i] = a[i] + b[i];
        }
    }
END
my $lib = $c->link;

# Standard (Safe) Binding
my $std_add  = wrap $lib, 'affix_add',  [ Vector [ 4, Float ], Vector [ 4, Float ] ] => Vector [ 4, Float ];
my $std_bulk = wrap $lib, 'affix_bulk', [ Pointer [Void], Pointer [Void], Pointer [Void], Int ] => Void;

# Setup Data
# A = [1, 2, 3, 4], B = [5, 6, 7, 8] -> Sum = [6, 8, 10, 12]
my $v1      = pack( 'f4', 1, 2, 3, 4 );
my $v2      = pack( 'f4', 5, 6, 7, 8 );
my $p1      = pdl( [ 1, 2, 3, 4 ] );
my $p2      = pdl( [ 5, 6, 7, 8 ] );
my $count   = 100_000;
my $bytes   = $count * 16;
my $big_a   = $v1 x $count;
my $big_b   = $v2 x $count;
my $big_out = "\0" x $bytes;
my $big_p1  = $p1->dummy( 1, $count )->clump(2);
my $big_p2  = $p2->dummy( 1, $count )->clump(2);

# Helper to unpack Vector result if Affix returns ArrayRef
sub unpack_if_ref($val) {
    return ref($val) eq 'ARRAY' ? pack( 'f4', @$val ) : $val;
}

# Test 1: Micro-Benchmark
say 'Micro-Benchmark (Single Vector Add)';
cmpthese(
    -5,
    {   PDL         => sub { my $r = $p1 + $p2; },
        'Inline::C' => sub { inline_add( $v1, $v2 ) },
        Affix       => sub { $std_add->( $v1, $v2 ) }
    }
);

# Test 2: Macro-Benchmark
say 'Macro-Benchmark (100k Vectors / 1.6MB)';
cmpthese(
    -5,
    {   PDL         => sub { my $r = $big_p1 + $big_p2; },
        'Inline::C' => sub { inline_bulk( $big_a, $big_b, $big_out, $count ) },
        Affix       => sub { $std_bulk->( \$big_a, \$big_b, \$big_out, $count ) }
    }
);

The Results

Micro-Benchmark (Single Vector Add)
               Rate       PDL Inline::C     Affix
PDL        623532/s        --      -92%      -92%
Inline::C 7706580/s     1136%        --       -3%
Affix     7933356/s     1172%        3%        --
Macro-Benchmark (100k Vectors / 1.6MB)
             Rate       PDL Inline::C     Affix
PDL        1436/s        --      -94%      -94%
Inline::C 23291/s     1522%        --       -1%
Affix     23473/s     1535%        1%        --

Analysis

1. Micro (Latency): Affix Wins by a Nose Affix is 3% faster than compiled XS (Inline::C) and more than 10x faster than PDL for small operations. This result is remarkable. Inline::C compiles static C code, but the XS macro system (SvPV, newSV) adds overhead. Affix generates a custom JIT trampoline at runtime that accesses the Perl stack and CPU registers directly, effectively behaving like inline assembly for Perl.

2. Macro (Throughput): Virtual Tie In the bulk data test, Affix and Inline::C perform identically. This confirms that Affix has zero overhead once the pointer is passed to C.

Benchmark 2: Matrix Math (PDL Logic)

PDL is famous for its terse syntax and "broadcasting" capabilities. A classic example from the PDL synopsis is filling a matrix based on its coordinates and summing the results.

Before we race, we will verify that our C implementation produces the exact same results as PDL.

The Logic: $y = $x + 0.1 * xvals($x) + 0.01 * yvals($x)

use v5.40;
use Benchmark qw[cmpthese];
use Affix     qw[:all];
use Affix::Compiler;
use PDL;

# 1. Compile C Matrix Logic
my $c = Affix::Compiler->new( flags => { cflags => '-O3' } );
$c->add( \<<~'END', lang => 'c' );
    void calc_matrix(double *out, int rows, int cols) {
        for(int y=0; y < rows; y++) {
            for(int x=0; x < cols; x++) {
                int i = y * cols + x;
                // logic: 0 + 0.1*x + 0.01*y
                out[i] = 0.1 * x + 0.01 * y;
            }
        }
    }
END
my $mat_lib = $c->link;
my $calc    = wrap $mat_lib, 'calc_matrix', [ Pointer [Void], Int, Int ] => Void;

# 2. Setup (1000x1000 Matrix)
my $N     = 1000;
my $p_mat = zeroes $N, $N;
my $c_buf = "\0" x ( $N * $N * 8 );    # 8 bytes per double

# 3. Verification
say "\nVerifying results...";

# Run PDL Logic
my $p_res = $p_mat + 0.1 * xvals($p_mat) + 0.01 * yvals($p_mat);

# Run Affix Logic
$calc->( \$c_buf, $N, $N );

# Compare value at specific coordinate [500, 500]
# Expected: 0.1*500 + 0.01*500 = 50 + 5 = 55
my $idx     = 500 * $N + 500;
my $val_c   = unpack( 'd', substr( $c_buf, $idx * 8, 8 ) );
my $val_pdl = $p_res->at( 500, 500 );
if ( abs( $val_c - $val_pdl ) < 1e-9 ) {
    say "Verification Successful: Matches at [500, 500] ($val_c)";
}
else {
    die "Mismatch! C=$val_c PDL=$val_pdl";
}

# 4. Benchmark
say "Matrix Benchmark (1000x1000)";
cmpthese(
    -5,
    {   PDL => sub {
            my $y = $p_mat + 0.1 * xvals($p_mat) + 0.01 * yvals($p_mat);
        },
        Affix => sub {

            # The raw C logic via Affix
            $calc->( \$c_buf, $N, $N );
        }
    }
);

The Results

Verifying results...
Verification Successful: Matches at [500, 500] (55)
Matrix Benchmark (1000x1000)
        Rate   PDL Affix
PDL   78.5/s    --  -99%
Affix 6795/s 8555%    --

Analysis

In this specific case, Affix is nearly 9x faster than PDL.

While PDL is highly optimized C, it still has to allocate temporary objects for xvals, yvals, and the intermediate multiplication results before summing them. The custom C function loaded by Affix does the calculation in a single pass over the memory with no intermediate allocations, allowing the CPU to cache lines efficiently.

Summary

• Speed: Affix is consistently the fastest way to bridge Perl and C, beating XS in call overhead and beating generic numerical libraries in raw throughput via custom C implementations.
• Simplicity: We achieved this performance with one line of Perl (affix), versus lines of XS macros or complex PDL threading logic.
• Portability: Affix requires no compilation at install time. You can ship your script without a C compiler dependency (if binding to pre-compiled libraries) and still get maximum performance.

Of course, none of these points matter if you can't write the C or Fortran to emulate the features of PDL.

In Chapter 30, we wrapped C++ classes using extern "C" helper functions in a shim. That is the 'Right Way.' But what if you are stuck with a pre-compiled C++ library and can't recompile it? What if you're just super bored? Let's say you have an object pointer, but no C functions to pass it to.

To call a method, we must act like a C++ compiler: manually look up the Virtual Method Table (vtable) and call the function pointer directly.

The Recipe

You know the drill. Let's create a C++ class, then write a Perl script that "scans" the object to find and call its methods.

use v5.40;
use Affix qw[:all];
use Affix::Build;
use Config;

# Compile C++ Library
# We define a class with virtual methods.
my $c = Affix::Build->new();
$c->add( \<<~'END', lang => 'cpp' );
    #include <stdio.h>

    class Calculator {
    public:
        // Explicit virtual destructor
        // (Often occupies vtable slots 0 & 1 on GCC/Clang)
        virtual ~Calculator() {}

        // Target Methods
        virtual void say_hello(const char* name) {
            printf("Hello, %s!\n", name);
        }

        virtual int add(int a, int b) {
            return a + b;
        }
    };

    extern "C" Calculator* new_calc() {
        return new Calculator();
    }
END
my $lib = $c->link;

# 2. Get the Object
affix $lib, 'new_calc', [] => Pointer [Void];
my $this = new_calc();

# Find the VTable
# The first 8 bytes (on 64-bit) of a C++ object are usually the vptr.
# Structure: $this -> [ vptr ] -> [ func0, func1, func2 ... ]
# Cast $this to a Pointer to a Size_t (Pointer-sized integer).
# This allows us to read the address stored in the first 8 bytes.
my $vptr_ref = cast( $this, Pointer [Size_t] );

# Dereference to get the address of the vtable array.
my $vtable_addr = $$vptr_ref;
say sprintf 'Object: 0x%X', address($this) ;
say sprintf 'VTable: 0x%X', $vtable_addr ;

# Scan the VTable
# We interpret the vtable as an array of 5 pointers (Size_t).
my $slots_pin = cast( $vtable_addr, Array [ Size_t, 5 ] );
my $addrs     = $$slots_pin;

# Dump the slots to find our functions.
# Code pointers will look like valid memory addresses.
# Slots 0/1 are often destructors or RTTI info.
for my $i ( 0 .. 4 ) {
    say sprintf( "Slot [%d]: 0x%X", $i, $addrs->[$i] );
}

# Heuristic:
# GCC/Clang (Itanium ABI): Slots 0,1 are dtors. Methods start at 2.
# MSVC (Windows ABI): Methods usually start at 0 or 1.
my ( $idx_hello, $idx_add ) = ( 2, 3 );

# Simple detection for MSVC
if ( $^O eq 'MSWin32' && $Config{cc} =~ /cl/i ) {
    ( $idx_hello, $idx_add ) = ( 0, 1 );
}
my $addr_hello = $addrs->[$idx_hello];
my $addr_add   = $addrs->[$idx_add];

# Bind and Call
# wrap() can take a raw integer address instead of a library handle.
#
# CRITICAL: C++ methods pass 'this' as the hidden first argument!
# We must include Pointer[Void] at the start of the signature.
if ( $addr_hello && $addr_add ) {
    my $say_hello = wrap( $addr_hello, [ Pointer [Void], String ] => Void );
    my $add       = wrap( $addr_add,   [ Pointer [Void], Int, Int ] => Int );

    # Execute
    $say_hello->( $this, 'Perl Hacker' );
    say '10 + 20 = ' . $add->( $this, 10, 20 );
}
else {
    say 'Could not resolve function addresses.';
}

The output looks like this:

Object: 0x11ECCA01160
VTable: 0x7FFBF85E4360
Slot [0]: 0x7FFBF85E2E40
Slot [1]: 0x7FFBF85E2E10
Slot [2]: 0x7FFBF85E2DC0
Slot [3]: 0x7FFBF85E2DA0
Slot [4]: 0x694D28203A434347
10 + 20 = 30
Hello, Perl Hacker!

How It Works

• 1. The Memory Layout

  $this (0x1000)
  .-----------.
  | vptr      | ------> vtable (0x5000)
  +-----------+        +--------------------------+
  | member... |        | [0] Destructor           |
   `---------'         | [1] Destructor (Deleting)|
                       | [2] say_hello()   <-- TARGET
                       | [3] add()         <-- TARGET
                       +--------------------------+
  

• 2. Double Indirection

  To get the function address, we must dereference twice:

  1. Read memory at $this to get $vtable_addr.
  2. Read memory at $vtable_addr + offset to get the function address.

  We achieve this in Affix by casting $this to Pointer[Size_t] to read the pointer-sized integer at that location.

• 3. Wrapping Raw Pointers

  my $sub = wrap( $address, [Args] => Ret );
  

  The wrap function usually takes a library handle. However, if you pass a raw integer address (or a Pin) as the first argument, Affix treats it as the function address itself. This converts the raw vtable entry into a callable Perl subroutine.

Kitchen Reminders

• Fragility

  This relies on the Itanium C++ ABI (Linux/macOS/MinGW) or the MSVC ABI (Visual Studio). The layout changes if you use Multiple Inheritance or Virtual Inheritance. This recipe works best for simple classes or COM-style interfaces (which use pure virtual classes).

• The 'this' Argument

  C++ methods are just C functions with a hidden first argument (this). When binding manually, you must include this argument in the signature (Pointer[Void]) and pass the object instance every time you call the method.

C++ libraries are notoriously difficult for FFI systems to bind. Unlike C, which uses standard symbol names, C++ "mangles" function names (for example, turning Warrior::attack() into _ZN7Warrior6attackEv) to support features like overloading. Furthermore, C++ methods require a hidden this pointer to know which object instance they are acting upon.

To capture a C++ class, we must bridge the gap by creating a C shim. This is a thin layer of extern "C" functions that expose the C++ class creation, destruction, and methods as standard C functions. This is the cheap and easy route.

The Recipe

This time, we're creating a simple C++ Droid class, wrap it in a C-compatible API, and then build a Perl class to manage the object's lifecycle automatically.

use v5.40;
use Affix qw[:all];
use Affix::Build;
$|++;

# 1. The C++ Class and C Shim
# We include both the class definition and the extern "C" wrappers.
my $src = <<~'END_CPP';
    #include <iostream>

    /* Pure C++ Class */
    class Droid {
        int serial;
    public:
        Droid(int s) : serial(s) {}
        ~Droid() {
            std::cout << "Droid " << serial << " is being scrapped." << std::endl;
        }
        void speak() {
            std::cout << "Droid " << serial << " says: Roger roger." << std::endl;
        }
    };

    /* The C ABI Shim */
    extern "C" {
        // Constructor Wrapper
        void* Droid_new(int serial) {
            return new Droid(serial);
        }

        // Method Wrapper (Pass object pointer explicitly)
        void Droid_speak(void* d) {
            static_cast<Droid*>(d)->speak();
        }

        // Destructor Wrapper
        void Droid_delete(void* d) {
            delete static_cast<Droid*>(d);
        }
    }
    END_CPP

# 2. Compile
# Affix::Build detects C++ by file extension or flags.
# We'll treat this string as .cpp content.
my $c = Affix::Build->new;
$c->add( \$src, lang => 'cpp' );
my $lib = $c->link;

# 3. Define the Perl Class
package Droid {
    use Affix qw[:all];

    # Define the pointer type for clarity
    # We treat the C++ object as an opaque Void Pointer in Perl
    typedef DroidPtr => Pointer [Void];

    # Bind the Shim functions
    affix $lib, 'Droid_new',    [Int]          => DroidPtr();
    affix $lib, 'Droid_speak',  [ DroidPtr() ] => Void;
    affix $lib, 'Droid_delete', [ DroidPtr() ] => Void;

    sub new( $class, $serial ) {

        # Call C++ new, get the pointer
        my $ptr = Droid_new($serial);

        # Bless the pointer into a reference to make it a Perl object
        return bless \$ptr, $class;
    }

    sub speak($self) {

        # Dereference to get the raw pointer and pass to shim
        Droid_speak($$self);
    }

    sub DESTROY($self) {

        # Automatically clean up C++ memory when Perl object dies
        Droid_delete($$self);
    }
}

# 4. Run it
{
    say 'Fabricating unit...';
    my $r2 = Droid->new(101);
    say 'Commanding unit...';
    $r2->speak();
    say 'Unit going out of scope...';
}
say 'Done.';

Output:

Fabricating unit...
Commanding unit...
Droid 101 says: Roger roger.
Unit going out of scope...
Droid 101 is being scrapped.
Done.

How It Works

• 1. The extern "C" Block

  This tells the C++ compiler: "Do not mangle these names." This ensures Droid_new appears in the library exactly as written, making it findable by Affix.

• 2. The Opaque Pointer (void*)

  Perl does not need to know the layout of the class Droid. To Perl, the object is just a memory address. In the C shim, we accept void* and static_cast it back to Droid* to call methods.

• 3. Lifecycle Management

  • Construction: Droid_new allocates memory on the C++ heap (new Droid).
  • Destruction: We bind Droid_delete to Perl's DESTROY method. When the Perl variable $r2 goes out of scope, Perl calls DESTROY, which calls delete in C++, preventing memory leaks.

Kitchen Reminders

• Catching Exceptions

  You must catch all C++ exceptions within your C shim. If a C++ exception throws across the FFI boundary into Perl, your program will crash instantly. Use try { ... } catch (...) { ... } blocks inside your shim functions to handle errors gracefully (perhaps returning an error code).

• Virtual Methods

  If you need to extend a C++ class in Perl (i.e., override a virtual method in Perl), the C shim must be more complex. You would do something like create a C++ child class that holds a Perl callback, forwarding the virtual call back to Perl.

While runtime wrapping is convenient, it has a startup cost (parsing headers every time). For a distributable CPAN module, you want the parsing to happen once (on your machine), generating a pure Perl module that users can load instantly.

Affix::Wrap objects have an affix_type method that returns the Perl code required to bind that entity. We can use this to write a code generator.

The Recipe

This recipe will generate a file named MyLib.pm from a C header, complete with POD documentation extracted from C comments.

use v5.40;
use Affix::Wrap;
use Path::Tiny;

# 1. The Source Header
my $header = Path::Tiny->tempfile( SUFFIX => '.h' );
$header->spew_utf8(<<~'H');
    /**
     * @brief A user profile.
     */
    typedef struct {
        int id;
        char name[32];
    } User;

    /**
     * @brief Fetch a user from the database.
     * @param id The user ID to lookup
     * @return A User struct
     */
    User get_user(int id);
    H

# 2. Parse
my $binder = Affix::Wrap->new( project_files => ["$header"] );
my @nodes  = $binder->parse;

# 3. Generate Perl Code
my $code = <<~'PERL';
    package MyLib;
    use v5.40;
    use Affix;

    # In a real module, you might locate the lib via Alien::Base or File::ShareDir
    my $lib = Affix::load('mylib');
    PERL
for my $node (@nodes) {

    # Spew POD
    if ( my $doc = $node->pod ) {
        $code .= "\n$doc=cut\n\n";
    }

    # Generate Affix calls
    if ( $node isa Affix::Wrap::Function ) {
        $code .= $node->affix_type . "\n";
    }
    elsif ( $node isa Affix::Wrap::Typedef ) {
        $code .= $node->affix_type . ";\n";
    }
    else {
        ...;    # etc.
    }
}
$code .= "\n1;";

# 4. Save
path('MyLib.pm')->spew_utf8($code);
say $code;

Output (The generated file):

package MyLib;
use v5.40;
use Affix;

# In a real module, you might locate the lib via Alien::Base or File::ShareDir
my $lib = Affix::load('mylib');

=head2 User

A user profile.

=cut

typedef User => Struct[ id => Int, name => Array[Char, 32] ];

=head2 get_user

Fetch a user from the database.

=over

=item C<id>

The user ID to lookup

=back

B<Returns:> A User struct

=cut

affix $lib, get_user => [Int], User;

1;

How It Works

• affix_type This method returns the string representation of the binding.

  • For User: Struct[ id => Int, name => Array[Char, 32] ].
  • For get_user: affix $lib, get_user => [Int], User;.

• pod Affix::Wrap parses Doxygen-style comments (@brief, @param) and converts them into standard Perl POD, meaning your generated module is automatically documented!

Kitchen Reminders

• Dependencies: The generated .pm file does not depend on Affix::Wrap or clang. It only depends on Affix. This means your end-users do not need a compiler installed to use your module.

The fastest way to use Affix::Wrap is Runtime Wrapping. In this workflow, you parse the headers and bind the functions immediately when your script starts. This is ideal for rapid prototyping, internal scripts, or testing, as you don't need to generate a separate .pm file.

The Recipe

In this demo, we'll use Affix::Build to build a library and Affix::Wrap to wrap it instantly.

use v5.40;
no warnings 'once';
use Affix qw[:all];
use Affix::Build;
use Affix::Wrap;
use Path::Tiny;

# 1. Setup the C Project
my $dir = Path::Tiny->tempdir( CLEANUP => 0 );
my $src = $dir->child('physics.c');
my $hdr = $dir->child('physics.h');

# A header with documentation
$hdr->spew_utf8(<<~'H');
    typedef struct { double x, y, z; } Vector3;

    /**
     * Adds two vectors.
     */
    Vector3 vec_add(Vector3 a, Vector3 b);

    // Global gravity constant
    extern double GRAVITY;
    H

# Implementation
$src->spew_utf8(<<~'C');
    #include "physics.h"
    #ifdef _WIN32
    __declspec(dllexport)
    #endif
    double GRAVITY = 9.81;

    #ifdef _WIN32
    __declspec(dllexport)
    #endif
    Vector3 vec_add(Vector3 a, Vector3 b) {
        Vector3 r = { a.x + b.x, a.y + b.y, a.z + b.z };
        return r;
    }
    C

# 2. Compile the library
my $compiler = Affix::Build->new( build_dir => $dir );
$compiler->add($src);
my $lib_path = $compiler->link;

# 3. The Magic: Bind it!
Affix::Wrap->new( project_files => [$hdr], include_dirs => [$dir] )->wrap($lib_path);

# 4. Use it
# The functions and typedefs are now installed in our package!
say "Gravity is: $main::GRAVITY";
my $v1 = { x => 1, y => 2, z => 3 };
my $v2 = { x => 4, y => 5, z => 6 };

# Affix::Wrap automatically handled the Struct definition for Vector3
my $v3 = vec_add( $v1, $v2 );
say "Result: $v3->{x}, $v3->{y}, $v3->{z}";

Output:

Gravity is: 9.81
Result: 5, 7, 9

How It Works

• ->wrap($lib) This method iterates over every node found in the AST.
  1. Typedefs/Structs: It calls Affix::typedef, registering types like Vector3 so they can be used in signatures.
  2. Variables: It calls Affix::pin, tying the Perl variable $GRAVITY to the symbol in the shared library.
  3. Functions: It calls Affix::affix, creating the Perl subroutine vec_add.

Note that we didn't write a single Struct[...] or affix signature manually.

Kitchen Reminders

• Namespaces: By default, wrap installs symbols into the caller's package. If you want to keep things clean, you can wrap into a dedicated class:

  package Physics {
      Affix::Wrap->new(...)->wrap($lib, 'Physics');
  }
  say Physics::vec_add(...);
  

Writing affix signatures manually is great for control, but it becomes tedious when wrapping a library with hundreds of functions and structs. Instead of staring at a .h file and manually translating C types to Affix types, let Affix::Wrap automate it!

Affix::Wrap is a friction-less introspection engine that parses C/C++ header files and builds a structured Abstract Syntax Tree (AST) of the library. It can detect functions, variables, macros, enums, and complex nested structs.

The Drivers

Affix::Wrap uses a dual-driver approach to read C code:

1. Clang Driver (Preferred): If the clang executable is found in your path, Affix::Wrap uses it to compile the header and dump the AST in JSON format. This is extremely accurate. It handles preprocessor macros, include paths, and complex typedef chains exactly as a C compiler would.
2. Regex Driver (Fallback): If Clang is missing, it falls back to a pure Perl, regex based parser. This is fast and has zero external dependencies but it may struggle with highly complex C++ templates or obscure preprocessor magic.

The Recipe

This time, let's parse a sample C header to see what Affix::Wrap sees.

use v5.40;
use Affix::Wrap;
use Path::Tiny;
use Data::Dump;

# 1. Create a dummy header file
my $header = Path::Tiny->tempfile( SUFFIX => '.h' );
$header->spew_utf8(<<~'C');
    /**
     * @brief A 2D Point
     */
    typedef struct {
        int x;
        int y;
    } Point;

    #define MAX_POINTS 100

    /**
     * @brief Calculate distance
     * @param p The point to measure
     */
    double distance(Point p);
    C

# 2. Parse the header
my $binder = Affix::Wrap->new( project_files => ["$header"] );
my @nodes  = $binder->parse;

# 3. Inspect the AST
foreach my $node (@nodes) {
    if ( $node isa Affix::Wrap::Function ) {
        say 'Found Function: ' . $node->name;
        say ' - Returns: ' . $node->ret->name;
    }
    elsif ( $node isa Affix::Wrap::Macro ) {
        say 'Found Macro: ' . $node->name . ' = ' . $node->value;
    }
    elsif ( $node isa Affix::Wrap::Typedef ) {
        say 'Found Typedef: ' . $node->name;
        say '  ' . $node->affix_type;
    }
}

Output:

Found Typedef: Point
  typedef Point => Struct[ x => Int, y => Int ]
Found Macro: MAX_POINTS = 100
Found Function: distance
 - Returns: double

How It Works

The parse method returns a list of Affix::Wrap::Entity objects. These objects understand the C types they represent. For example, the Struct object contains a list of Member objects, which in turn hold Type objects.

Affix::Wrap flattens complex relationships. Notice in the C code, Point was a typedef around an anonymous or named struct. Affix::Wrap detects this common pattern and merges them, presenting you with a single named Struct entity.

Kitchen Reminders

• Includes: If your header includes other files (like <stdio.h> or "my_types.h"), pass their locations to the constructor via include_dirs => ['/path/to/includes'].
• Documentation: Affix::Wrap also captures comments! In the next chapters, we will use this to generate documentation automatically.

Sometimes, the best tool for the job involves three different tools. You might have legacy math models in Fortran, performance-critical loops in Assembly, and a clean C API to orchestrate them.

By mastering the Polyglot strategy, you're no longer forced to rewrite battle-tested legacy code or compromise on performance. You can let Fortran handle the numerics, Assembly handle the vectorization, and C handle do coordination, all while driving the whole machine from Perl. Affix::Build turns what used to be a nightmare of Makefiles and linker flags into a simple script, freeing you to choose the absolute best tool for every specific problem.

The Recipe

Here, we'll create a library called number_cruncher. It uses C as the API controller, Fortran for the calculation logic, and Assembly for a low-level incrementer. It's not a very useful example but it's a good demonstration of our task.

use v5.40;
use Affix qw[:all];
use Affix::Build;
use Config;

# 1. Initialize
# We assume we are compiling different languages into one binary
my $c = Affix::Build->new( name => 'number_cruncher' );

# 2. C Source (The Orchestrator)
# Standard C function that will be the entry point or helper
$c->add( \<<~'', lang => 'c' );
    #include <stdio.h>
    #ifdef _WIN32
    __declspec(dllexport)
    #endif
    int core_version() { return 1; }

# 3. Fortran Source (The Math Engine)
# Uses ISO C Binding to export 'fortran_add' as a standard C symbol
$c->add( \<<~'', lang => 'f90' );
    function fortran_add(a, b) bind(c, name='fortran_add')
        use iso_c_binding
        integer(c_int), value :: a, b
        integer(c_int) :: fortran_add
        fortran_add = a + b
    end function

# 4. Assembly Source (The Optimizer)
# A simple function 'asm_inc(x)' that returns x + 1.
# We must detect the CPU architecture to provide the correct instructions.
my $arm = $Config{archname} =~ /arm64|aarch64/;
$c->add( \( $arm ? <<~'' : $^O eq 'MSWin32' ? <<~'': <<~'' ), lang => $arm ? 's' : 'asm' );
        ; ARM64: Linux, macOS, Windows ARM uses standard system assembler (.s)
        .global asm_inc
        .text
        .align 2
        asm_inc:
            add w0, w0, #1  ; Increment w0 (first arg) by 1
            ret

        ; Win64 x86_64: Uses RCX for first argument
        global asm_inc
        section .text
        asm_inc:
            mov eax, ecx
            inc eax
            ret

        ; SysV x86_64: Uses RDI for first argument
        global asm_inc
        section .text
        asm_inc:
            mov eax, edi
            inc eax
            ret


# 5. Build
# The compiler detects mixed languages and switches to 'Polyglot Mode'
my $lib = $c->link;
say 'Compiled Polyglot Library: ' . $lib;

# 6. Bind
affix $lib, 'core_version', []           => Int;
affix $lib, 'fortran_add',  [ Int, Int ] => Int;
affix $lib, 'asm_inc',      [Int]        => Int;

# 7. Run
say 'Core Version: ' . core_version();
say 'Fortran Add:  ' . fortran_add( 10, 20 );
say 'ASM Inc:      ' . asm_inc(99);

Output:

Compiled Polyglot Library: C:/Users/S/AppData/Local/Temp/Fye21DPXMw/number_cruncher.dll.1
Core Version: 1
Fortran Add:  30
ASM Inc:      100

How It Works

• The Polyglot Strategy

  When Affix::Build detects multiple languages, it switches from 'Native' mode (one compiler does everything) to 'Polyglot' mode.

  1. It asks gcc to compile the C code to math_core.o.
  2. It asks gfortran to compile the Fortran code to math_algos.o.
  3. It asks nasm (or the system assembler on ARM) to assemble the ASM code to fast.o.
  4. It invokes the system linker (usually via cc or c++) to combine all three objects into the final shared library.

Kitchen Reminders

• ABI Compatibility

  The reason this works is that C, Fortran, and Assembly all speak the same "C ABI" (Application Binary Interface). Rust, Zig, and Odin also support this. Languages like Go and C# require heavier runtimes, but Affix::Build handles the initialization flags for you.

• Name Mangling

  If you mix C++ into this stack, remember to wrap your exported functions in extern "C". Without it, the C++ compiler will mangle the names (like _Z8func_cppv or whatever your compiler's ABI defines), and the linker won't be able to match them if another language tries to call them. We'll demonstrate how you'd work around this in a future chapter.

C might be the lingua franca of system programming, but Affix::Build is a true polyglot. We've covered building shared libraries in C and C++ but Affix::Build supports over 15 languages out of the box.

This chapter is a quick-reference gallery. Each recipe demonstrates how to compile a single function that adds two integers, exported in a way that Affix can easily bind to it.

Assembly

For raw speed, you can write directly in Assembly.

Affix::Build supports both x86_64 (via NASM) and ARM64 (via the system assembler). Calling conventions differ by architecture and operating system, so we check %Config to determine which assembly code to use.

use v5.40;
use Affix qw[:all];
use Affix::Build;
use Config qw[%Config];
$|++;
#
my $c   = Affix::Build->new( name => 'asm_lib' );
my $arm = $Config{archname} =~ /arm64|aarch64/;
$c->add( \( $arm ? <<~'' : $^O eq 'MSWin32' ? <<~'': <<~'' ), lang => $arm ? 's' : 'asm' );
        ; ARM64: add w0, w0, w1
        .global add
        .text
        .align 2
        add:
            add w0, w0, w1
            ret

        ; Win64 x86_64: RCX + RDX -> RAX
        global add
        section .text
        add:
            mov eax, ecx
            add eax, edx
            ret

        ; SysV x86_64: RDI + RSI -> RAX
        global add
        section .text
        add:
            mov eax, edi
            add eax, esi
            ret

#
affix $c->link, 'add', [ Int, Int ] => Int;
say add( 4, 5 );

C# (.NET 8+)

With .NET 8's NativeAOT feature, you can compile C# directly to native machine code (no .NET runtime required on the target machine).

use v5.40;
use Affix qw[:all];
use Affix::Build;
$|++;
#
my $c = Affix::Build->new( name => 'cs_lib' );
$c->add( \<<~'CS', lang => 'cs' );
    using System.Runtime.InteropServices;
    namespace Demo {
        public class Math {
            [UnmanagedCallersOnly(EntryPoint="add")]
            public static int Add(int a, int b) => a + b;
        }
    }
    CS
#
affix $c->link, 'add', [ Int, Int ] => Int;
say add( 4, 5 );

D (Dlang)

D is unique: on Windows, a specific initialization mixin is required to ensure its runtime starts correctly when loaded as a DLL.

use v5.40;
use Affix qw[:all];
use Affix::Build;
$|++;
#
my $c   = Affix::Build->new( name => 'd_lib' );
my $src = ( $^O eq 'MSWin32' ? <<~'WIN32' : '' ) . <<~'D';
    import core.sys.windows.dll;
    mixin SimpleDllMain;
    export
WIN32
    extern(C) int add(int a, int b) { return a + b; }
D
$c->add( \$src, lang => 'd' );
#
affix $c->link, 'add', [ Int, Int ] => Int;
say add( 4, 5 );

Fortran

Modern Fortran (2003+) offers the iso_c_binding module, which ensures your functions use standard C types and calling conventions, avoiding the historic "name mangling" issues (like appended underscores).

use v5.40;
use Affix qw[:all];
use Affix::Build;
$|++;
#
my $c = Affix::Build->new( name => 'fortran_lib' );
$c->add( \<<~'F90', lang => 'f90' );
    function add(a, b) bind(c, name='add')
        use iso_c_binding
        integer(c_int), value :: a, b
        integer(c_int) :: add
        add = a + b
    end function
F90
#
affix $c->link, 'add', [ Int, Int ] => Int;
say add( 4, 5 );

Go (Golang)

Go functions must be marked with the special //export comment to be visible.

Note: The Go runtime spins up background threads (for GC and scheduling) that do not shut down cleanly when a shared library is unloaded. This often causes access violations on Windows during program exit.

Affix attempts to detect Go libraries (by looking for the _cgo_dummy_export symbol) and pin them in memory to prevent this crash. However, if you still encounter segmentation faults during global destruction, you can force a safe exit using POSIX::_exit(0).

use v5.40;
use Affix qw[:all];
use Affix::Build;
$|++;
#
my $c = Affix::Build->new( name => 'go_lib' );
$c->add( \<<~'GO', lang => 'go' );
    package main
    import "C"

    //export add
    func add(a, b C.int) C.int {
        return a + b
    }

    func main() { }
    GO
#
affix $c->link, 'add', [ Int, Int ] => Int;
say add( 4, 5 );

Rust

Rust is an excellent candidate for FFI because it has no garbage collector and strict ABI control. You must use #[no_mangle] and extern "C" to prevent the compiler from changing your function names.

use v5.40;
use Affix qw[:all];
use Affix::Build;
#
my $c = Affix::Build->new( name => 'rust_lib' );
$c->add( \<<~'RUST', lang => 'rust' );
    #[no_mangle]
    pub extern "C" fn add(a: i32, b: i32) -> i32 {
        a + b
    }
RUST
#
my $lib = $c->link;
affix $lib, 'add', [ Int, Int ] => Int;
say add( 4, 5 );

Zig

Zig is fully C-ABI compatible. Use the export keyword to make functions available.

use v5.40;
use Affix qw[:all];
use Affix::Build;
$|++;
#
my $c = Affix::Build->new( name => 'zig_lib' );
$c->add( \<<~'ZIG', lang => 'zig' );
    export fn add(a: i32, b: i32) i32 {
        return a + b;
    }
ZIG
#
affix $c->link, 'add', [ Int, Int ] => Int;
say add( 4, 5 );

Chef's Notes

This gallery is just a sample but demonstrates that Perl's role as the ultimate glue language is stronger than ever. By abstracting away the build chain complexities... from the strict safety of Rust to the raw speed of Assembly, Affix::Build allows you to reach far outside the CPAN ecosystem. You can now leverage the unique strengths of almost any language, modern or ancient, without ever leaving the comfort of Perl.

In our next chapter, we'll discuss merging some of these languages into a single shared library.

Throughout this cookbook, you've seen me use Affix::Build to generate shared libraries on the fly from C. However, building shared libraries isn't a concept limited to just C so neither are the capabilities of Affix::Build.

Affix::Build, first introduced in our very first chapter, is a cross-platform, multi-lingual compilation wrapper. While it supports over 15 languages (including Rust, Go, Fortran, and D), it is particularly useful for C and C++ because it abstracts away the differences between GCC, Clang, and MSVC, handles operating system extensions (.dll, .so, .dylib), automatically locates the correct driver to link the standard libraries properly, and applies critical flags like -fPIC where required.

In this chapter, we'll demonstrate that as we build a library that uses the C++ Standard Library (STL) to reverse a string in place.

The Recipe

use v5.40;
use Affix qw[:all];
use Affix::Build;

# 1. Configure the Compiler
# We enable debug mode to see the underlying compiler commands
my $compiler = Affix::Build->new( name => 'str_util', debug => 1 );

# 2. Add C++ Source
# We pass a reference to a string containing the source code.
# We must specify 'lang => cpp' so the compiler knows to use a C++ driver.
$compiler->add( \<<~'CPP', lang => 'cpp' );
    #include <algorithm>
    #include <string>
    #include <cstring>

    extern "C" {
        // We use C linkage so Affix can find the symbol easily
        void reverse_it(char* buffer) {
            if (!buffer) return;

            // Use high-level C++ features
            std::string s(buffer);
            std::reverse(s.begin(), s.end());

            // Copy back to the raw buffer
            std::strcpy(buffer, s.c_str());
        }
    }
    CPP

# 3. Compile and Link
# This generates the shared object in a temp folder and returns the path.
my $lib_path = $compiler->link;
say 'Library compiled to: ' . $lib_path;

# 4. Bind and Call
affix $lib_path, 'reverse_it', [String] => Void;
my $text = 'Affix makes C++ easy';
reverse_it($text);
say $text;

Output:

[Affix] Exec: g++ -shared C:/Users/S/AppData/Local/Temp/Yx9npzzak9/source_001.cpp -o C:/Users/S/AppData/Local/Temp/Yx9npzzak9/str_util.dll.1
Library compiled to: C:/Users/S/AppData/Local/Temp/Yx9npzzak9/str_util.dll.1
ysae ++C sekam xiffA

How It Works

• 1. Language Detection

  When you call $compiler->add( ..., lang => 'cpp' ), the compiler updates its internal linking strategy. It knows that standard C compilers (gcc) cannot link C++ standard libraries (libstdc++) by default. It resolves the correct driver on your system (preferring g++ or clang++) to ensure the STL is available.

• 2. Inline Source

  Passing a SCALAR reference (\<<~'CPP') tells the compiler to write that content to a temporary file before compiling. You can also pass a standard file path (e.g., $compiler->add('src/utils.cpp')), in which case the lang parameter is optional (auto-detected by extension).

• 3. Platform Abstraction

  The call to $compiler->link performs a lot of heavy lifting:

  • Windows: It knows to use the .dll extension and creates a PE binary.
  • Linux/Unix: It uses .so or .dylib, adds the -fPIC (Position Independent Code) flag required for shared libraries, and ensures the library name is prefixed with lib if required by the OS loader.

Bonus: The Polyglot Strategy

Affix::Build can mix languages in a single library. If you add files from different languages (e.g., C and Rust), the compiler switches strategies:

1. It compiles every source file into a static object file (.o or .a) using the appropriate compiler for each language (rustc for Rust, cc for C, etc.).
2. It invokes the system linker to combine all static objects into one shared library.

# Example: Using a Rust function inside a C wrapper
$compiler->add( 'src/calc.rs' );  # Contains #[no_mangle] pub extern "C" fn add...
$compiler->add( 'src/main.c' );   # Calls 'add' from Rust
$compiler->link;                  # Links them into one DLL

We'll revisit this concept in future chapters.

Kitchen Reminders

• Compiler Flags

  You can optimize your build or include external headers using the flags parameter in the constructor.

  my $c = Affix::Build->new(
      flags => {
          cxxflags => '-O3 -I/usr/local/include/my_lib',
          ldflags  => '-L/usr/local/lib -lmy_lib'
      }
  );
  

• Windows MinGW

  If you are using Strawberry Perl, Affix::Build automatically handles the whole-archive linking issues often encountered when bundling static C++ libraries into a DLL, ensuring no symbols are stripped during the link phase.

"Is Affix faster than... everything else?"

Let's find out!

The Perl ecosystem offers several ways to call C code. How does Affix stack up?

We will benchmark four approaches:

1. Inline::C

   Compiles C code into a custom XS module. This is historically the fastest way to call C from Perl, as it eliminates the "FFI" layer entirely, but it requires a C compiler at installation time (or runtime).

2. FFI::Platypus

   The current dominant FFI module on CPAN. It uses libffi (a portable, general-purpose Foreign Function Interface library) to handle argument shuffling.

3. Affix

   Our pride and joy. Wraps our custom JIT engine: infix.

The Recipe

We'll compile a shared library performing a simple integer addition, then bind it using all three libraries.

use v5.40;
use Benchmark qw[cmpthese];
use Affix     qw[:all];
use Affix::Build;
use FFI::Platypus 2.00;

# All tests run on the same C.
my $c_code;
BEGIN { $c_code = <<~'' }
    int add_ints(int a, int b) {
        return a + b;
    }


# Inline::C setup
use Inline C => Config => ( ccflags => '-O3' );
use Inline C => $c_code;

# We must compile the library for Affix and Platypus
my $c = Affix::Build->new( flags => { cflags => '-O3' } );
$c->add( \$c_code, lang => 'c' );
my $lib = $c->link;

# FFI::Platypus setup
my $ffi_plat = FFI::Platypus->new( api => 2 );
$ffi_plat->lib($lib);
$ffi_plat->attach( [ add_ints => 'add_platypus' ], [ 'int', 'int' ], 'int' );

# Affix setup
affix $lib, [ add_ints => 'add_affix' ], [ Int, Int ] => Int;

# The race
say 'Benchmarking for 5 seconds...';
my $x = 100;
my $y = 200;
cmpthese - 5, {
    Inline   => sub { add_ints( $x, $y ) },
    Platypus => sub { add_platypus( $x, $y ) },
    Affix    => sub { add_affix( $x, $y ) }
    }

The Results

On my system, I see results similar to this:

Benchmarking for 5 seconds...
               Rate Platypus   Inline    Affix
Platypus  5026460/s       --     -57%     -82%
Inline   11641793/s     132%       --     -58%
Affix    28016636/s     457%     141%       --

Analysis

• Affix vs. XS (Inline::C)

  To the reader, the most startling result is that Affix is over 120% faster than Inline::C. Traditionally, XS (which Inline::C generates) was considered the speed limit for Perl. Affix beats this limit by generating a custom JIT trampoline that reads Perl's internal SVs directly, skipping the overhead of the Perl stack macros (dXSARGS, ST(n)) that XS relies on. It is effectively "Inline Assembly" for Perl.

• Affix vs. Platypus

  Affix runs over 4x faster than Platypus in this test. While Platypus relies on libffi (a stable, generic library), libffi must handle platform logic at runtime. Affix generates specific machine code for the exact call signature, allowing for drastic optimizations.

Summary

• Use Affix for flexibility, safety, and complex types (Structs, Arrays). As the benchmarks show, you sacrifice absolutely nothing in speed; in fact, you'll likely gain performance over hand-written XS.
• Use Inline::C if you are already comfortable writing XS macros and want static compilation, though you may actually lose performance compared to Affix unless you really focus on optimizing your own hot path.
• Use FFI::Platypus if you already already have code written and don't feel like moving to Affix?

"Is Affix faster than pure Perl?"

This is the most important question. Migrating code to C requires an investment of time (and maybe the development of a new skillset) so the payoff must be worth it.

The simple answer is usually, "Yes, but..."

You must understand that there is a fixed cost to crossing the boundary between Perl and C (marshalling arguments, setting up the stack, the JIT trampoline, etc.). If your C function does very little (like adding two integers), the overhead might outweigh the speed gain. However, if your C function does heavy lifting (matrix math, cryptography, parsing), the gain is massive.

Let's benchmark a naive Fibonacci calculation in Perl versus C to see exactly how much faster native code can be.

The Recipe

use v5.40;
use Affix qw[:all];
use Affix::Build;
use Benchmark qw[cmpthese];
$|++;

# 1. Compiled C Implementation
# We use -O3 to ensure the C compiler optimizes the recursion as much as possible
my $c = Affix::Build->new( flags => { cflags => '-O3' } );
$c->add( \<<~'', lang => 'c' );
    int fib_c(int n) {
        if (n < 2) return n;
        return fib_c(n-1) + fib_c(n-2);
    }

affix $c->link, 'fib_c', [Int] => Int;

# 2. Perl Implementation
sub fib_p($n) {
    return $n if $n < 2;
    return fib_p( $n - 1 ) + fib_p( $n - 2 );
}

# 3. Benchmark
say 'Small N (High overhead impact)';
cmpthese(
    -5,
    {   Perl  => sub { fib_p(10) },
        Affix => sub { fib_c(10) }
    }
);

say 'Large N (Raw compute dominance)';
cmpthese(
    -5,
    {   Perl  => sub { fib_p(30) },
        Affix => sub { fib_c(30) }
    }
);

These are the results from my machine:

Small N (High overhead impact)
            Rate   Perl  Affix
Perl     47892/s     --  -100%
Affix 10607023/s 22048%     --
Large N (Raw compute dominance)
        Rate   Perl  Affix
Perl  3.16/s     --  -100%
Affix  959/s 30258%     --

How It Works

• 1. The Boundary Cost

  In the "Small N" test, Affix leans on C to overcome the overhead of marshalling data back and forth. Even with the FFI 'tax,' the compiled C code processes the recursion so efficiently that it still runs 200x faster than the Perl subroutine.

• 2. The Compute Gain

  In the "Large N" test, the difference widens. The C code runs entirely in native machine code for millions of recursive iterations before crossing back to Perl. As you can see, this results in the C version being over 300x faster than pure Perl.

Kitchen Reminders

• Batching

  If you have a C function that sets a single pixel, calling it 2 million times from Perl will be slow. Instead, expose a C function that takes an Array[Pixel] and updates the whole image in one call. We'll get to even more advanced features like SIMD in later chapters.

In C, OOP is often simulated using structs containing function pointers. This pattern is commonly known as a vtable and allows a library to call different implementations of a function depending on the object it's holding. With Affix, you can populate these fields with Perl subroutines, effectively creating a Perl class that C can call into.

We're going to use a vtable to build a hypothetical plugin system. The C library expects a struct containing handlers for init and process.

The Recipe

use v5.40;
use Affix qw[:all];
use Affix::Build;

# 1. Compile C Library
my $c = Affix::Build->new();
$c->add( \<<~'END', lang => 'c' );
    typedef struct {
        const char* name;
        int (*init)(void);
        void (*process)(const char* data);
    } Plugin;

    void run_plugin(Plugin* p, const char* data) {
        if (p->init()) {
            p->process(data);
        }
    }
    END
my $lib = $c->link;

# 2. Define Types
# The struct fields must include the Callback signature.
typedef Plugin => Struct [ name => String, init => Callback [ [] => Int ], process => Callback [ [String] => Void ] ];

# 3. Bind
affix $lib, 'run_plugin', [ Pointer [ Plugin() ], String ] => Void;

# 4. Create the Perl "Object"
# We use a HashRef to represent the struct.
# The callback fields are populated with anonymous subs.
my $my_plugin = {
    name => 'PerlPlugin v1.0',
    init => sub {
        say 'Initializing Perl Plugin...';
        return 1;    # Success
    },
    process => sub ($data) {
        say 'Processing data in Perl: ' . $data;
    }
};

# 5. Call C
# We pass the reference to our hash. Affix packs it into the C struct,
# creating trampolines for the subroutines automatically.
run_plugin( $my_plugin, 'Some Input Data' );

How It Works

• 1. Nested callbacks

  init => Callback[ [] => Int ]
  

  When you define a struct member as a Callback, Affix expects a Perl code reference in that hash field.

• 2. Automatic trampolining

  When you pass the HashRef $my_plugin to the C function, Affix:

  1. Allocates the C struct.
  2. Copies the string 'PerlPlugin v1.0'.
  3. Asks infix to generate reverse trampolines for the init and process subs.
  4. Stores the function pointers to those trampolines in the struct and passes the whole thing on to our library.

• 3. Execution

  When C calls p->process(data), it hits the trampoline which bounces everything over to your Perl sub.

Kitchen Reminders

• Lifetime management

  The trampolines generated for struct members exist only for the duration of the call to run_plugin. If the C library stores this struct pointer globally and tries to call the callbacks after run_plugin returns, the program will crash.

  If you need persistent callbacks, you must manually allocate the struct using calloc and assign the fields, ensuring the Perl variables stay in scope.

Manual free() is error-prone. We can use Perl's DESTROY phase to automate it.

The Recipe

use v5.40;
no warnings 'experimental';
use feature 'class';
use Affix qw[:all];

class AutoPointer {
    field $ptr : reader : param;

    method DESTROY {
        Affix::free($ptr);    # Call C/malloc free()
    }
}

# Usage
{
    my $safe = AutoPointer->new( ptr => malloc(1024) );

    # ... use $safe->ptr ...
    Affix::dump( $safe->ptr, 1024 );
}    # $safe goes out of scope -> DESTROY called -> free() called.

How It Works

When the last reference to $safe disappears, Perl calls DESTROY. By wrapping our raw C pointer in a blessed object, we tie the C memory lifecycle to the Perl variable's scope. This is known as RAII (Resource Acquisition Is Initialization).

When things go wrong in Perl, you probably get an error message. When they go wrong in C, you get a segfault, a frozen terminal, or data corruption that only shows up three hours later. This chapter provides a survival kit for when things go very wrong and it's your job to figure it all out.

The Bug: The Vanishing Callback

A common source of crashes in FFI is the scope mismatch. Perl relies on reference counting while C expects you to know what you are doing.

# 1. C library (A Fake Event System)
# void register_handler(void (*cb)(void));
# void trigger_event();

# 2. Perl binding
affix $lib, 'register_handler', [ Callback[ [] => Void ] ] => Void;
affix $lib, 'trigger_event',    [] => Void;

# 3. The Buggy Code
sub setup {
    # We pass an anonymous subroutine.
    # Affix creates a trampoline for it.
    register_handler( sub { say "Event!" } );

    # END OF SCOPE
    # The anonymous sub may have a refcount of 0. If so, perl frees it and the reverse trampoline is destroyed.
}

setup();

# ... later ...
trigger_event(); # SEGFAULT! C jumps to freed memory.

The Fix

You must ensure the Perl CodeRef lives as long as the C library needs it.

my $KEEP_ALIVE; # Global or Object-level storage

sub setup {
    $KEEP_ALIVE = sub { say "Event!" };
    register_handler( $KEEP_ALIVE );
}

# ...
trigger_event(); # Works!

The Debugging Toolbox

Affix provides tools to inspect raw memory and internal states.

Affix::dump( $ptr, $bytes )

If you suspect struct alignment issues or garbage data, look at the raw bytes.

my $ptr = malloc(16);
# ... C writes to ptr ...

# Dump 16 bytes to STDOUT in Hex/ASCII format
Affix::dump($ptr, 16);

address( $ptr ) & is_null( $ptr )

Sanity check your pointers. A pointer value of 0 (NULL) or 0xFFFFFF... (Uninitialized memory) is a smoking gun.

my $ptr = some_c_function();
say sprintf("Pointer address: 0x%X", address($ptr));

errno()

If a C function returns -1 or NULL to indicate failure, it usually sets a system error code (errno/GetLastError).

if ( some_call() < 0 ) {
    die "C Error: " . errno();
}

4. sv_dump( $scalar )

Debugging the Perl side with this function that's a part of perl's internal API. If Affix refuses to accept a variable, checking its internal flags (Integer vs String, Readonly status) often reveals why.

my $val = 10;
Affix::sv_dump($val);

SV = IV(0x25b7a6b4118) at 0x25b7a6b4128
  REFCNT = 1
  FLAGS = (IOK,pIOK)
  IV = 10

Common Pitfalls

• Writing to Strings

  String arguments (const char*) are read-only copies. If the C function modifies the string in-place, you MUST use Pointer[Char] and pass a mutable scalar.

• Struct padding

  C compilers insert padding bytes to align members. If your Perl Struct[...] definition doesn't match the C compiler's packing rules, fields will be read from the wrong offsets. Use dump() to verify alignment.

• VarArg types

  Passing a Perl float to a variadic function? It promotes to double. Passing an integer? It promotes to int64_t. If the C function expects a 32-bit int, use coerce(Int32, $val).

System calls fail. In C, you check the global errno variable (or GetLastError() on Windows). In Affix, we expose this via errno().

The Recipe

We will try to open a file that doesn't exist.

use v5.40;
use Affix qw[:all];

# fopen returns NULL on failure
affix libc, 'fopen', [ String, String ] => Pointer [Void];
my $file = fopen( 'non_existent_file.txt', 'r' );
if ( is_null($file) ) {

    # get_system_error returns the numeric error code
    # In numerical context: errno
    # In string context:    strerror(errno)
    my $err = Affix::errno();
    die sprintf 'Could not open file: %s (%d)', $err, $err;
}

Kitchen Reminders

• errno returns a dualvar

  • Numeric: 2 (ENOENT)
  • Stringified: "No such file or directory"

  This makes logging errors rather convenient.

Bridging I/O between languages is historically one of the most fragile aspects of FFI. While handling file descriptors manually as opaque pointers may be effective on POSIX systems, that approach is a minefield on Windows. Affix solves this by introducing two smart types: File and PerlIO. These types handle the extraction, translation, and lifecycle management of file handles automatically.

This chapter covers the correct usage patterns and advanced scenarios like multiplexing I/O between Perl and C on the same handle.

The Basic Recipe: Writing from C

Let's start with the basics: passing a Perl filehandle to a C function that expects FILE*.

Note the use of Pointer[File] in the signature; using just File would attempt to pass the opaque structure by value, which is not what we want.

use v5.40;
use Affix;
use Affix::Build;
#
# 1. Build the lib
my $c = Affix::Build->new();
$c->add( \<<~'', lang => 'c' );
    #include <stdio.h>
    void write_recipe(FILE* stream, const char* title) {
        if (stream) {
            fprintf(stream, "RECIPE: %s\n", title);
            fprintf(stream, "1. Preheat oven.\n");
            fprintf(stream, "2. Mix ingredients.\n");
            fflush(stream); // Important: C buffers IO. Flush if mixing with Perl IO.
        }
    }

my $lib = $c->link;

# 2. Bind the function
# CRITICAL STEP: Check your signature!
# Wrong:  [ File, String ]          <- Passes the whole skillet (Struct Copy)
# Right:  [ Pointer[File], String ] <- Passes the handle (Pointer)
affix $lib, 'write_recipe', [ Pointer [File], String ] => Void;

# 3. Start Cooking
open my $fh, '>', 'dinner_plans.txt' or die "Oven broke: $!";

# Affix unwraps the Perl glob, extracts the FILE*, and hands it to C.
write_recipe( $fh, 'Spicy Meatballs' );

# You may use normal file I/O functions
close $fh;

# 4. Verify
print $^O eq 'MSWin32' ? `type dinner_plans.txt` : `cat dinner_plans.txt`;

Advanced: Multiplexing I/O (Perl & C Interleaved)

A common point of failure in FFI is buffering. Perl has its own IO buffers, and C stdio has its own. If you mix print (Perl) and fprintf (C) on the same handle, the output often appears out of order.

To handle this, we rely on flushing. In this example, we mix Perl's unbuffered syswrite with C's standard I/O.

use v5.40;
use Affix;
use Affix::Build;

# 1. Add a C function that writes a specific block
my $c = Affix::Build->new();
$c->add( \<<~'', lang => 'c' );
    #include <stdio.h>
    void c_append_middle(FILE* f) {
        if (!f) return;
        fprintf(f, "[C] ...Middle Content...\n");
        fflush(f); // FLUSH is mandatory to sync with Perl's layer
    }

my $lib = $c->link;
affix $lib, 'c_append_middle', [ Pointer [File] ] => Void;
open my $fh, '>', 'multiplex.log' or die $!;

# 1. Perl writes the Header
syswrite $fh, "[Perl] Header\n";

# 2. C writes the Body
# We pass the same handle. Affix extracts the FILE* from it.
c_append_middle($fh);

# 3. Perl writes the Footer
syswrite $fh, "[Perl] Footer\n";
close $fh;

# Verify order is preserved
say 'Multiplex Results:';
print do { local ( @ARGV, $/ ) = 'multiplex.log'; <> };

Advanced: Reading from C

You can also open a file in Perl (handling errors and paths in high-level code) and hand it off to C for parsing or reading.

use v5.40;
use Affix;
use Affix::Build;
my $c = Affix::Build->new();
$c->add( \<<~'', lang => 'c' );
    #include <stdio.h>
    // Reads a single char from the stream and returns it as an Int
    int c_get_char(FILE* f) {
        if (!f) return -1;
        return fgetc(f);
    }

my $lib = $c->link;
affix $lib, 'c_get_char', [ Pointer [File] ] => Int;

# Create a dummy file
open my $w, '>', 'input.txt';
print $w 'ABC';
close $w;

# Open for reading in Perl
open my $reader, '<', 'input.txt';

# Read byte 1 via C
say 'Read from C: ' . chr c_get_char($reader);    # Output: A

# Read byte 2 via Perl
read $reader, my $buf, 1;
say 'Read from Perl: ' . $buf;                    # Output: B

# Read byte 3 via C
say 'Read from C: ' . chr c_get_char($reader);    # Output: C
close $reader;

The PerlIO Interface

If you are interfacing with XS modules or Perl internals (someday...), the API might expect PerlIO* instead of standard C FILE*. Affix provides the Pointer[PerlIO] type for this.

While Pointer[PerlIO] behaves similarly to Pointer[File] in usage (it unwraps the handle), it creates a distinct type check in the signature to prevent mixing up abstract PerlIO streams with raw stdio streams.

use v5.40;
use Affix;
use Affix::Build;
use Fcntl qw[SEEK_CUR SEEK_SET];
my $c = Affix::Build->new();
$c->add( \<<~'', lang => 'c' );
    // A function that theoretically accepts a PerlIO stream.
    // Since we aren't linking libperl in this stub, we use void*
    // to simulate an opaque handle, but in real XS, this would be PerlIO*.
    void* identity(void* stream) {
        return stream;
    }

my $lib = $c->link;

# We define the signature using Pointer[PerlIO] to enforce intent
affix $lib, 'identity', [ Pointer [PerlIO] ] => Pointer [PerlIO];
open my $fh, '+<', 'input.txt';
syswrite $fh, "Test\n";
sysseek $fh, 0, SEEK_SET;    # both handles will be at this position

# Round-trip the handle through C
my $returned_fh = identity($fh);

# Affix wraps the returned pointer in a new GLOB reference
say 'Original Handle: ' . $fh;
say 'Returned Handle: ' . $returned_fh;
say 'Original Handle Pos: ' . sysseek $fh,          0, SEEK_CUR;
say 'Returned Handle Pos: ' . sysseek $returned_fh, 0, SEEK_CUR;
say 'Reading returned handle: ' . <$returned_fh>;    # Could just as easily read the original
say 'Original Handle Pos: ' . sysseek $fh,          0, SEEK_CUR;
say 'Returned Handle Pos: ' . sysseek $returned_fh, 0, SEEK_CUR;

Kitchen Reminders

• Key takeaway: Multiplexing file IO will likely require you pay close attention to buffering. Use fflush(f) or a similar function from C and syswrite or other unbuffered IO functions in Perl.

• Summary of Types

  Perl Type	C Equivalent	Usage
  Pointer[File]	FILE*	Standard C I/O (stdio.h)
  Pointer[PerlIO]	PerlIO*	Perl Internal I/O or XS Extensions
  File	struct FILE	Opaque struct (Rarely used directly)

Speed is the reason I wrote infix and Affix and I think I've achieved a good balance of features vs. speed. I've done a lot of work to reduce overhead on the hot path but what if we could go... faster?

Standard Affix calls involve a "marshalling" phase: Perl values are converted to C values, stored in a buffer, passed to the JIT trampoline, and then passed to the C function. This is flexible, but adds overhead. Direct marshalling removes the middleman; Affix generates a specialized JIT trampoline that knows how to read your Perl variables directly from SV*.

The Recipe

We will compare the standard wrap against direct_wrap using a simple addition function.

use v5.40;
use Affix qw[:all];
use Affix::Build;
use Benchmark qw[cmpthese];

# 1. Compile
my $c = Affix::Build->new();
$c->add( \<<~'END', lang => 'c' );
    int add(int a, int b) { return a + b; }
END
my $lib = $c->link;

# 2. Standard Bind
my $std_add = wrap $lib, 'add', [ Int, Int ] => Int;

# 3. Direct Bind
# The syntax is identical, just a different function name.
my $fast_add = direct_wrap $lib, 'add', [ Int, Int ] => Int;

# 4. Benchmark
cmpthese(
    -5,
    {   Standard => sub { $std_add->( 10, 20 ) },
        Direct   => sub { $fast_add->( 10, 20 ) }
    }
);

The Results

On my machine, direct marshalling can be faster for simple primitives.

               Rate Standard   Direct
Standard 21102786/s       --      -9%
Direct   23192942/s      10%       --

The difference between the two grows as the number of parameters grows. With a C function with just one more int param (int add(int a, int b, int c) { return a + b + c; }), the benchmarks look like this:

               Rate Standard   Direct
Standard 20533088/s       --     -18%
Direct   25184993/s      23%       --

Ideally, I'll close the gap a little more between the two but that's a problem for future me.

How It Works

• Standard trampoline

  perl stack -> argument buffer (void **) -> trampoline -> function

  The trampoline is generic. It expects arguments to be laid out in a specific C array format.

• Direct trampoline

  trampoline -> perl stack -> function

  infix generates machine code that calls directly into the perl API (e.g. SvIV) to fetch values from the Perl stack, puts them into CPU registers, and jumps straight to the target function. We cut out the middleman and reduces pointer indirection layers and memory reads. These are micro-optimizations that produce real world results.

Kitchen Reminders

• Experimental

  This feature is currently marked as experimental because it is not as feature complete as the mainline infix trampolines and Affix's standard functions. It supports primitives (Int, Float, Pointer) robustly, but complex aggregates (Structs by value) may fallback to standard marshalling or behave unexpectedly in edge cases.

• Restrictions

  Direct marshalling relies on knowing exact types at compile time. ...the JIT's compile time not application compile time. It's less forgiving of type mismatches (like passing a Struct where an Int is expected) than the standard path, which often attempts coercion.

C headers are full of enum definitions. To the C compiler, these are just integers. To the programmer, they have semantic meaning. When binding these functions to Perl, passing magic numbers (like 0, 1, 2) makes your code unreadable. Affix solves this by allowing you to define Enums that generate Perl constants and return dualvars.

The Recipe

Let's wrap a hypothetical state machine library.

use v5.40;
use Affix qw[:all];
use Affix::Build;

# 1. Compile C
my $c = Affix::Build->new();
$c->add( \<<~'END', lang => 'c' );
    typedef enum {
        STATE_IDLE     = 0,
        STATE_RUNNING,
        STATE_PAUSED,
        STATE_ERROR    = 99,
        // Bitmasks
        FLAG_READ      = 1 << 0,  // 1
        FLAG_WRITE     = 1 << 1,  // 2
        FLAG_RDWR      = FLAG_READ | FLAG_WRITE // 3
    } MachineState;

    int set_state(MachineState s) {
        return s; // Returns the new state
    }
END
my $lib = $c->link;

# 2. Define the Enum
# We can use:
#   'NAME'            -> Auto-increments
#   [ NAME => Value ] -> Explicit value
#   [ NAME => 'Expr'] -> C-style expression strings
typedef State => Enum [
    [ STATE_IDLE => 0 ],                           # 0
    'STATE_RUNNING',                               # Auto-set to 1
    'STATE_PAUSED',                                # Auto-set to 2
    [ STATE_ERROR => 99 ],                         # Manually set to 99
    [ FLAG_READ   => '1 << 0' ],                   # Calculated value is 1
    [ FLAG_WRITE  => '1 << 1' ],                   # Calculated value is 2
    [ FLAG_RDWR   => 'FLAG_READ | FLAG_WRITE' ]    # Calculated value is 3
];

# 3. Bind
# We use the typedef alias '@State'
affix $lib, 'set_state', [ State() ] => State();

# 4. Use Constants (Input)
# typedef installs these constants into your package.
set_state( STATE_RUNNING() );
set_state( FLAG_RDWR() );

# 5. Use Dualvars (Output)
# The return value behaves like a number AND a string.
my $current = set_state( STATE_PAUSED() );
if ( $current == 2 ) {
    say 'Numeric check passed.';
}
if ( $current eq 'STATE_PAUSED' ) {
    say 'String check passed.';
}

# Debugging is free
say sprintf 'Current State: %s (%d)', $current, $current;    # Prints "Current State: STATE_PAUSED (2)"

How It Works

• 1. Constants Generation

  When you call typedef ... Enum [...], Affix calculates the integer value of every element in the list. When you pass that to typedef, Affix installs a constant subroutine (e.g., sub STATE_IDLE () { 0 }) into your current package. This allows you to use the names directly in your Perl code.

• 2. Expression parsing

  Affix includes a small expression parser (see shunting yard algorithm). This allows you to copy and paste definitions from C headers like bit shifts (1 << 4) or combinations (FLAG_A | FLAG_B) directly into your Affix definition without manually calculating the result.

• 3. Dualvars

  When a C function returns a value defined as an Enum type, Affix looks up the integer in the definition map. If found, it returns a Scalar with both slots filled:

  • IV (Integer Value): 2
  • PV (String Value): "STATE_PAUSED"

  This allows you to write readable code (eq 'STATE_...') while maintaining high performance for numeric comparisons.

Kitchen Reminders

• Unknown values

  If the C library returns an integer that you didn't define in your Enum list (e.g., a new error code added in a library update), Affix simply returns the integer.

• Scope

  The constants are installed into the package where typedef is called. If you put your Affix setup in a utility module (e.g., Acme::LibSample), callers will need to import those constants or refer to them fully qualified (Acme::LibSample::STATE_IDLE).

In Chapter 12, we passed a simple callback. But real-world C libraries often take a callback and an opaque pointer (usually something like void * user_data) which allows you to store context data. However, if the library's author decided not to provide that, Affix allows you to bake the context into the callback itself using closures.

The Recipe

We will wrap a hypothetical iteration function that takes a callback but no data argument.

use v5.40;
use feature 'class';
no warnings 'experimental::class';
use Affix qw[:all];
use Affix::Build;

# 1. C Library
my $c = Affix::Build->new();
$c->add( \<<~'END', lang => 'c' );
    void iterate_3( void (*cb)(int) ) {
        cb(1); cb(2); cb(3);
    }
END
my $lib = $c->link;
affix $lib, 'iterate_3', [ Callback [ [Int] => Void ] ] => Void;

# 2. The Problem
# We want to sum these numbers into a Perl object.
# But the C function doesn't give us a slot to hand it the object directly.
class Sum {
    field $total = 0;
    method add($n) { $total += $n; }
    method get()   { $total; }
}
my $obj = Sum->new;

# 3. The Closure Solution
# We define the callback inside the scope where $obj exists.
# Perl closes over $obj.
my $cb = sub ($n) { $obj->add($n); };

# 4. Execute
iterate_3($cb);
say 'Total: ' . $obj->get();    # 6

How It Works

• 1. Closures

  When you create sub { ... } in Perl, it captures the surrounding lexical variables ($obj). Event the new perlclass objects are just SV*s at their core.

• 2. The trampoline

  When you pass $cb to Affix, infix generates a reverse trampoline. This is a C function pointer that exists just to bounce arguments back to us. When C calls it, the trampoline invokes the specific Perl CODE reference (CV*), which restores the Perl environment and finds $obj.

  To the C library, it looks like a standard function pointer. To you, it looks like magic.

Sometimes you need to peek inside the oven. If a function expects a C style struct to be passed by value or if you want to read struct members without writing C helper functions, you can define the struct layout in Perl and Affix handles the rest.

The Recipe

We will create a simple geometry library that operates on Points and Rectangles.

use v5.40;
use Affix qw[:all];
use Affix::Build;

# 1. Compile C Library
my $c = Affix::Build->new();
$c->add( \<<~'END', lang => 'c' );
    typedef struct {
        int x, y;
    } Point;

    typedef struct {
        Point top_left;
        Point bottom_right;
        int color;
    } Rect;

    // Pass by Value
    int area(Rect r) {
        int w = r.bottom_right.x - r.top_left.x;
        int h = r.bottom_right.y - r.top_left.y;
        return w * h;
    }

    // Pass by Reference (Modify in place)
    void move_rect(Rect *r, int dx, int dy) {
        r->top_left.x += dx;
        r->top_left.y += dy;
        r->bottom_right.x += dx;
        r->bottom_right.y += dy;
    }
END
my $lib = $c->link;

# 2. Define Types
# Order matters! Defined in the same order as the C struct.
typedef Point => Struct [ x => Int, y => Int ];

# We can nest types. Point() refers to the typedef above.
typedef Rect => Struct [ tl => Point(), br => Point(), color => Int ];

# 3. Bind
affix $lib, 'area',      [ Rect() ]                       => Int;
affix $lib, 'move_rect', [ Pointer [ Rect() ], Int, Int ] => Void;

# 4. Use (Pass by Value)
# We represent the struct as a hash reference.
my $r = { tl => { x => 0, y => 0 }, br => { x => 10, y => 20 }, color => 0xFF00FF };
say 'Area: ' . area($r);    # 200

# 5. Use (Pass by Reference)
# We pass a reference to the hash.
# Affix updates the hash members after the call.
move_rect( $r, 5, 5 );
say "New TL: $r->{tl}{x}, $r->{tl}{y}";    # 5, 5

How It Works

• 1. Structural typing

  typedef Point => Struct [ x => Int, y => Int ];
  

  Affix (by way of infix's type introspection) calculates the memory layout (padding, alignment, size) of the struct matching the platform's C ABI.

• 2. Hashref marshalling

  When you pass a hashref (HV*) to a Struct argument, Affix packs the hash values into a temporary C memory block. When the call returns, if you passed a reference (as in step 5), Affix automagically unpacks the C memory back into your hashref, seemingly updating any changed values.

• 3. Nesting

  Structs can contain other Structs, Arrays, and Pointers. Affix handles the recursive packing and unpacking automatically.

Kitchen Reminders

• Performance

  Marshalling a deep hashref into a C struct (and back) involves copying data. For high-performance scenarios (like processing millions of points), use calloc to allocate the struct in C memory once, and pass the pointer around.

• Missing keys

  If your hashref is missing a key defined in the Struct's list of fields, Affix assumes 0 (or NULL).

C libraries often use callback functions to delegate logic back to the user. The standard library's qsort is the classic example: it knows how to sort, but it doesn't know how you'd like to compare your data. Affix allows you to pass a standard Perl subroutine (CodeRef) where C expects a function pointer.

The Recipe

Let's sort a list of integers using qsort.

use v5.40;
use Affix;

# 1. Bind qsort
# void qsort(void *base, size_t nmemb, size_t size,
#            int (*compar)(const void *, const void *));
#
# The callback signature is defined inside the argument list.
# Callback[ [Args...] => ReturnType ]
affix libc, 'qsort', [
    Pointer [Int], Size_t, Size_t,    
    Callback [ [ Pointer [Int], Pointer [Int] ] => Int ]
] => Void;

# 2. Prepare Data
# qsort works on a raw memory array.
# We create a C array of integers.
my @nums  = ( 88, 56, 100, 2, 25 );
my $count = scalar @nums;

# 3. Define the Comparator
# C passes us pointers to the two items being compared.
# We must dereference them to get the values.
my $compare_fn = sub ( $p_a, $p_b ) { $$p_a <=> $$p_b };

# 4. Call
# We pass the ArrayRef directly. Affix handles the pointer decay
# and write-back (see Chapter 10).
# sizeof(int) is usually 4.
qsort( \@nums, $count, 4, $compare_fn );
say join ', ', @nums;    # 2, 25, 56, 88, 100

How It Works

• 1. The Callback type

  Callback[ [Pointer[Int], Pointer[Int]] => Int ]
  

  This tells Affix to create a reverse trampoline. It generates a small piece of C code that looks like a standard C function. When that C code is called by qsort, it:

  1. Marshals the C arguments (two pointers here) into Perl variables (SV*).
  2. Calls your Perl subroutine $compare_fn.
  3. Takes the integer return value and passes it back to C.

• 2. Pointer dereferencing

  $$p_a <=> $$p_b;
  

  In the callback signature, we claimed the arguments were Pointer[Int]. Affix receives the raw address from C, wraps it in a pin with the type Int, and passes it to your sub. Dereferencing it reads the integer value from memory.

Kitchen Reminders

• Scope and lifecycle

  The magic trampoline created for your subroutine exists only as long as the C call is running. If you pass a callback to a C function that stores it for later use (like setting an event handler in a GUI library), you must ensure your Perl CodeRef stays alive.

• Exceptions

  If your Perl callback throws an exception (croak, die, etc.), Affix catches it, issues a warning, and returns a zero/void value to C to prevent crashing the host application. C does not understand Perl exceptions and the C function might not understand this either. Be careful!

Some C functions don't have a fixed number of arguments. The most famous example is printf, which takes a format string and... almost everything else. Affix supports variadic functions, but they require a little more care than standard functions because the C compiler isn't there to cast types for you.

The Recipe

We will wrap the standard C library's snprintf. This is safer than printf because it writes to a buffer.

We also handle a common cross-platform annoyance: on Windows, this function is often named _snprintf, while on Linux/macOS it is snprintf. We can normalize this using symbol aliasing.

use v5.40;
use Affix;

# 1. Bind snprintf
# Windows uses _snprintf, POSIX uses snprintf.
# We detect the OS and bind the correct symbol to the name 'snprintf'.
my $symbol = ( $^O eq 'MSWin32' ? '_' : '' ) . 'snprintf';

# Signature: int snprintf(char *str, size_t size, const char *format, ...);
# We use '...' (or VarArgs) to indicate the variadic part.
affix libc, [ $symbol => 'snprintf' ], [ Pointer [Char], Int, String, VarArgs ] => Int;

# 2. Prepare a buffer
# We need a place for C to write the formatted string.
my $size   = 100;
my $buffer = "\0" x $size;

# 3. Call with Inference
# Affix guesses the C type based on the Perl value.
#   "A String" -> char*
#   123        -> int64_t
#   12.34      -> double
snprintf( $buffer, $size, 'Name: %s, ID: %d, Score: %.2f', 'Alice', 42, 99.9 );

# Clean up the null terminator and print
$buffer =~ s/\0.*//;
say $buffer;    # "Name: Alice, ID: 42, Score: 99.90"

# 4. Explicit Coercion
# Sometimes inference isn't enough.
# Here we want to print a pointer address (%p).
# Passing a Perl integer might be interpreted as a value, not a pointer.
# We use coerce() to force specific types.
my $ptr = 0xDEADBEEF;
snprintf( $buffer, $size, 'Pointer address: %p', coerce( Pointer [Void], $ptr ) );
say $buffer =~ s/\0.*//r;

How It Works

• 1. Symbol aliasing

  affix $lib, [ 'RealSymbol', 'PerlName' ], ...
  

  Often, C libraries have messy naming conventions or platform-specific quirks (like _snprintf vs snprintf), or the library has mangled symbol names. By passing an array reference as the symbol name, you tell Affix: "Find RealSymbol in the library, but create a Perl subroutine named PerlName."

• 2. VarArgs in signature

  [ ..., VarArgs ]
  

  The VarArgs constant (or the string ';') tells Affix: "Stop checking types here. Everything after this point is dynamic."

• 3. Dynamic JIT

  When you call a variadic function, Affix looks at the arguments you provided at that moment. It generates a custom, temporary infix style signature on the fly (e.g., (*char, size_t, *char, *char, int64, double)->int) and compiles a trampoline for it.

  These trampolines are cached, so calling snprintf repeatedly with the same argument types is fast.

• 4. Coercion

  coerce( Type, $value )
  

  If you need to pass a float (not double), a short, or a struct by value, the default inference won't work. coerce attaches a "Type Hint" to the value. Affix sees this hint and uses the exact type you requested.

Kitchen Reminders

• Format strings

  Affix does not validate your format string (e.g. "%s"). If you pass an integer to a %s placeholder, snprintf will assume it's a pointer address and try to read memory from it which will likely result in a segfault.

In C, arrays passed to functions are often used as output buffers. The function reads data from the array, modifies it in place, and expects the caller to see the changes. In XS, these are defined as IN_OUT parameters. Affix supports this functionality automatically.

The Recipe

We will define a C function that reverses an integer array in place.

use v5.40;
use Affix qw[:all];
use Affix::Build;

# 1. Compile C
my $c = Affix::Build->new();
$c->add( \<<~'END', lang => 'c' );
    // Reverses array 'a' in place
    int array_reverse(int a[], int len) {
        int tmp, i;
        int ret = 0; // Sum of first half (arbitrary logic)
        for (i = 0; i < len / 2; i++) {
            ret += a[i];
            tmp = a[i];
            a[i] = a[len - i - 1];
            a[len - i - 1] = tmp;
        }
        return ret;
    }

    // Wrapper for fixed-size 10-element array
    int array_reverse10(int a[10]) {
        return array_reverse(a, 10);
    }
END
my $lib = $c->link;

# 2. Bind
# Dynamic length array
affix $lib, array_reverse => [ Array[Int], Int ], Int;

# Fixed length array
affix $lib, array_reverse10 => [ Array[Int, 10] ], Int;

# 3. Call (Fixed Size)
my @a = ( 1 .. 10 );
array_reverse10( \@a );
say "@a"; # Output: 10 9 8 7 6 5 4 3 2 1

# 4. Call (Dynamic Size)
my $b = [ 1 .. 20 ];
array_reverse( $b, 20 );
say "@$b"; # Output: 20 19 ... 2 1

How It Works

• 1. In-Place Modification

  When you pass a Perl ArrayRef to an Array argument:

  1. Affix allocates a temporary C array.
  2. Copies your data into it.
  3. Calls the C function.
  4. Copies the data back from C to your Perl array.

  This ensures that any changes made by the C function are reflected in your SV*.

• 2. Fixed size padding

  affix ..., [ Array[Int, 10] ] ...
  

  If you define a fixed size (e.g. 10) but pass a shorter list (e.g. [1, 2]), Affix allocates the full 10 integers and zero-fills the rest. This protects the C function from reading garbage memory if it assumes the array is 10 elements long.

Kitchen Reminders

• Performance

  Copying arrays back and forth (Perl -> C -> Perl) takes time. For very large datasets where you only need to modify a few bytes, consider using malloc and Pointer types to keep the data in C memory (see Chapter 7).

• Reference requirement

  You must pass a reference (\@a or $array_ref). Passing a list directly (e.g. array_reverse([1,2], ...)) works, but the modifications will be lost because the anonymous array reference is discarded immediately after the call. This is a limitation of perl's handling of scalars.

In C, arrays and pointers are cousins. In function arguments, they are twins. Declaring void func(int a[]) is exactly the same as void func(int *a). However, in Perl and thus in Affix, this isn't the case.

• Pointer[Int] expects a reference to a scalar integer (memory address).
• Array[Int] expects a list of integers (data).

This recipe shows how to pass lists of data to C using the Array type, which handles allocation and copying for you.

The Recipe

We will wrap a C function that sums a zero-terminated array of integers.

use v5.40;
use Affix qw[:all];
use Affix::Build;

# 1. Compile C
my $c = Affix::Build->new();
$c->add( \<<~'END', lang => 'c' );
    #include <stdlib.h>

    int array_sum(const int * a) {
        int i, sum;
        if (a == NULL)
            return -1; // Error code for NULL

        // Loop until we hit a 0
        for (i = 0, sum = 0; a[i] != 0; i++)
            sum += a[i];
        return sum;
    }
END

# 2. Bind
# We use Array[Int] to tell Affix to expect a list of numbers.
# Affix handles the C "array decay" (passing as pointer) automatically.
affix $c->link, array_sum => [ Array [Int] ] => Int;

# 3. Call
# Pass undef -> NULL
say "Undef: " . array_sum(undef);    # -1

# Pass [0] -> {0} -> Sums to 0 (Immediate stop)
say "Zero:  " . array_sum( [0] );    # 0

# Pass List -> {1, 2, 3, 0} -> Sums to 6
say "List:  " . array_sum( [ 1, 2, 3, 0 ] );    # 6

How It Works

• 1. Array marshalling

  array_sum( [ 1, 2, 3, 0 ] );
  

  When you define an argument as Array[Int], Affix allocates a temporary C array, copies the values from your Perl array reference into it, and passes the pointer to the C function. This matches C's array decay rules.

• 2. The undef case

  array_sum(undef);
  

  C allows array pointers to be NULL. Affix supports this by checking for undef and passing NULL to the function, skipping the allocation entirely.

Kitchen Reminders

• Syntactic sugar

  You could have bound this function using Pointer[Int]. If you did, you would have to manually pack a binary string or use malloc to create the buffer. Array[Int] handles that boilerplate for you.

• Sentinels

  Notice we passed 0 at the end of our list: [1, 2, 3, 0]. Affix does not automatically append null terminators to numeric arrays (unlike Strings). Since our C function relies on finding a 0 to stop the loop, we must provide it explicitly. I could have designed this demonstration to accept a Size_t array length but this seemed like a better lesson to sneak in here.

Perl scripts often live in the terminal or deep inside a server handing out webpages or whatnot, but they don't have to stay there. Earlier, in chapter 2, we demonstrated using system libraries on Windows to display a simple MessageBox. This time, we turn to Linux (and BSDs with a desktop environment).

The libnotify library allows applications to send pop-up notifications to the user. These are the bubbles you see from apps like Slack, Discord, or your system updater.

This library uses the GLib object system (GObject), which is notoriously complex to bind manually. However, if we only want to trigger a notification, we can treat the objects as opaque pointers and ignore the internal complexity entirely.

The Recipe

use v5.40;
use Affix;

# 1. Find the library
# locate_lib searches system paths (like /usr/lib) for libnotify.so
# It returns the full path as a string.
my $libnotify = Affix::locate_lib('notify');
unless ($libnotify) {
    die "Could not find libnotify. Are you on Linux with libnotify-bin installed?";
}

# 2. Bind the lifecycle functions
# bool notify_init(const char *app_name);
affix $libnotify, 'notify_init',   [String] => Bool;
affix $libnotify, 'notify_uninit', []       => Void;

# 3. Bind the object functions
# NotifyNotification* notify_notification_new(const char *summary, const char *body, const char *icon);
affix $libnotify, 'notify_notification_new', [ String, String, String ] => Pointer [Void];

# bool notify_notification_show(NotifyNotification *notification, GError **error);
affix $libnotify, 'notify_notification_show', [ Pointer [Void], Pointer [Void] ] => Bool;

# 4. Use it
# Initialize the library with our app name
notify_init('Affix');

# Create the notification object
# We get back a Pin (opaque pointer), but we don't need to look inside it.
my $n = notify_notification_new(
    'Keep your stick on the ice! 🏒',                          #
    "Hello from Affix!\nWelcome to the fun world of FFI.",    #
    'dialog-information'                                      # Standard icon name
);

# Show it
# We pass undef for the second argument (GError**), meaning "ignore errors".
notify_notification_show( $n, undef );

# Clean up
notify_uninit();

How It Works

• 1. Locating libraries

  my $libnotify = Affix::locate_lib('notify');
  

  Unlike load_library (which loads the lib into memory immediately), locate_lib scans the system's dynamic linker paths (like LD_LIBRARY_PATH) and simply returns the filename. We pass this path to affix, which handles the loading.

• 2. Ignoring complexity

  The notify_notification_new function returns a NotifyNotification*. In C, this is a struct with many private fields and GObject metadata. In Perl, we don't care. We define the return type as Pointer[Void].

  As long as we pass that pointer back to other functions in the same library (like notify_notification_show), everything works.

• 3. Optional Error Handling

  The C signature for show is:

  gboolean notify_notification_show (NotifyNotification *notification, GError **error);
  

  The second argument is a pointer-to-a-pointer that the library fills if an error occurs. By passing undef, Affix sends a NULL, telling the library we don't want to receive error details.

Kitchen Reminders

• Platform specific

  This recipe works on Linux/BSD with a freedesktop.org compliant notification daemon (GNOME, KDE, XFCE, etc.). It will fail to find the library on Windows or vanilla macOS.

• Unicode

  Notice we used an emoji in the notification. Because we used the String type, Affix automatically encoded this as UTF-8, which is what modern Linux libraries expect.

Perl handles memory for you. C does not. When you start allocating raw memory buffers in Affix, you are stepping into the C world. This recipe demonstrates how to manually allocate, manipulate, and free memory using the standard C lifecycle.

The Recipe

use v5.40;
use Affix qw[:all];

# 1. Allocation
# We allocate 14 bytes.
# malloc returns a Pointer[Void] (a pin).
my $void_ptr = malloc( 14 );

# 2. String Duplication
# strdup allocates new memory containing a copy of the string.
my $ptr_string = strdup("hello there!!\n");

# 3. Memory Copy
# We copy 15 bytes (text + null terminator) from one pointer to another.
# memcpy works on raw pointers, so passing $void_ptr is fine.
memcpy( $void_ptr, $ptr_string, 15 );

# 4. Read (The Cast)
# Dereferencing $void_ptr ($$void_ptr) would just return the memory address
# as an integer, because Affix doesn't know what's inside.
# We must CAST it to a specific type to read it.
my $str_view = cast( $void_ptr, String );

print $str_view;

# 5. Cleanup
# These are managed pointers, so Perl *would* free them when the variables
# go out of scope. But in C, explicit is better than implicit.
free($ptr_string);
free($void_ptr);

How It Works

• 1. malloc returns an opaque pointer

  my $void_ptr = malloc( 14 );
  

  Just like in C, malloc returns a void*. It represents a chunk of raw memory with no type information attached.

• 2. cast adds meaning

  my $str_view = cast( $void_ptr, String );
  

  Casting creates a new pin that points to the same memory address but has different type metadata.

  $void_ptr  -> Address 0x1234, Type: *void
  $str_view  -> Address 0x1234, Type: *char
  

  When you dereference $$str_view, Affix sees the type is String and reads the memory until the null terminator.

• 3. Pointer arithmetic via memcpy

  memcpy( $void_ptr, $ptr_string, 15 );
  

  This is a raw memory operation. It doesn't care about types; it just moves bytes. This is extremely fast, but dangerous if you get the length wrong. Always ensure your destination buffer is large enough to hold the source data.

• 4. Explicit free

  free($void_ptr);
  

  Affix uses Perl's memory allocator (safemalloc) for malloc, calloc, and strdup. This means these pointers are distinct from pointers allocated by a loaded C library (which use the system malloc).

  Always use Affix's free for Affix-created pointers, and the library's free function for library-created pointers. You must keep them seperate!

Kitchen Reminders

• The void * Default

  If you see a large integer when you expected data (e.g. 2029945...), you are dereferencing a Pointer[Void]. Use cast($ptr, Type) to tell Affix how to read the data.

• Buffer Overflows

  Perl aims to protect you from buffer overflows. Affix does not. memcpy will happily write past the end of your allocation and corrupt your program's memory. Measure twice, cut once. Yes, that's a carpentry reference in a cookbook. It's fine.

Affix cannot directly instantiate a C++ class or read a C struct unless we define every single field with the correct type and in the proper order. Often, we don't care about the fields, we just want to hold onto the object and pass it back to the library later. In this case, we can treat the object as a black box, an opaque pointer (Pointer[Void]) and use helper functions to interact with it.

The Recipe

We will create a simple C++ Person struct, wrap it with C-compatible helpers, and bind it to Perl.

use v5.40;
use Affix qw[:all];
use Affix::Build;

# 1. Compile the C++ Library
my $c = Affix::Build->new();
$c->add( \<<~'END', lang => 'cpp' );
    #include <stdlib.h>
    #include <string.h>

    typedef struct {
        char * name;
        unsigned int age;
    } Person;

    // extern "C" prevents C++ name mangling. This ensures the symbols in the
    // DLL are just "person_new", not "_Z10person_newPKcj".
    extern "C" {
        Person * person_new(const char * name, unsigned int age) {
            Person * self = (Person *) malloc(sizeof(Person));
            self->name = strdup(name);
            self->age = age;
            return self;
        }

        const char * person_name(Person * self) {
            return self->name;
        }

        unsigned int person_age(Person * self) {
            return self->age;
        }

        void person_free(Person * self) {
            if (self) {
                free(self->name);
                free(self);
            }
        }
    }
END
my $lib = $c->link;

# 2. Define the Opaque Type
# We tell Affix: "Whenever you see 'Person', treat it as a void pointer."
typedef Person => Pointer [Void];

# 3. Bind
# Notice we use Person() as the type.
affix $lib, 'person_new',  [ String, UInt ] => Person();
affix $lib, 'person_name', [ Person() ]     => String;
affix $lib, 'person_age',  [ Person() ]     => UInt;
affix $lib, 'person_free', [ Person() ]     => Void;

# 4. Use
my $roger = person_new( 'Roger Frooble Bits', 35 );

# $roger is just a blessed scalar holding a memory address.
# We pass it back to C to get data.
say 'Name: ' . person_name($roger);
say 'Age:  ' . person_age($roger);

# Clean up
person_free($roger);

How It Works

• 1. extern "C"

  C++ compilers "mangle" function names (e.g., person_new might become _Z10person_newPKcj) to support function overloading. This is a nightmare to bind and depends on the ABI of the compiler. Wrapping your helper functions in extern "C" forces compilers to use standard C names.

• 2. The shim pattern

  We didn't bind the C++ struct fields directly. Instead, we wrote small C helper functions (shims) to access the data:

  const char * person_name(Person * self) { return self->name; }
  

  This insulates you from changes in the C++ class layout.

• 3. Typedef aliases

  typedef Person => Pointer[Void];
  

  This creates a named alias. While Pointer[Void] works, using Person() in your signatures makes the code self-documenting. If you later define Car => Pointer[Void], Affix won't stop you from passing a Car to a Person function (since they are both just pointers), but your code will be much easier to read and debug.

Kitchen Reminders

• Constructors must return

  It seems obvious but, in C++, forgetting to return self; in a constructor function is easy to do and leads to immediate segfaults or illegal instruction errors.

• Memory ownership

  Since person_new used malloc, the memory belongs to the C heap. You must explicitly call person_free to release it. Affix cannot garbage collect these opaque pointers automatically unless you attach a destructor using Perl's DESTROY (which is an advanced topic I might cover in another chapter).

Every operating system comes with a "C standard library" or, simply, libc. It contains the fundamental building blocks of C programming: memory allocation, file I/O, string manipulation, system calls, and much more.

Affix provides a shortcut to access this library without you needing to know its exact filename (which varies wildly between Linux, macOS, and Windows).

The Recipe

We will bind the standard C function puts which takes a string, prints it to stdout, and appends a newline character.

use v5.40;
use Affix;

# 1. Bind
# libc is a helper function exported by Affix.
# It returns the handle to the system's standard library.
affix libc, 'puts' => [String] => Int;

# 2. Call
# Note: puts() automatically adds a newline "\n"
puts('Hello World');

# 3. Check Result
# puts returns a non-negative integer on success, or EOF (-1) on error.
my $ret = puts('Affix makes FFI easy.');
say 'Successfully wrote to stdout.' if $ret >= 0;

How It Works

• 1. The libc Helper

  Finding the standard library is platform-dependent:

  • Linux: libc.so.6 (usually)
  • macOS: libSystem.B.dylib
  • Windows: msvcrt.dll (or ucrtbase.dll)

  The libc function (exported by Affix) handles this detection logic for you, returning the correct handle for your operating system.

• 2. The puts Function

  The C signature for puts is:

  int puts(const char *s);
  

  We map this to:

  [String] => Int
  

  When you call puts('Hello'), Affix takes your Perl string, ensures it is null-terminated, and passes the pointer to C. The C library writes the bytes to the file descriptor for STDOUT.

Kitchen Reminders

• Output Buffering

  Perl and C use their own IO buffers. If you mix print "...", say "...", and puts(...) in the same script, the output might appear out of order depending on when the buffers are flushed. Setting $| = 1; in Perl usually helps, but doesn't strictly control the C buffer.

• Implicit Newlines

  Unlike Perl's print, puts always adds a newline. If you want to print without a newline using C, you would need to bind printf.

Dealing with C strings usually involves asking "Who owns this memory?"

In Perl, strings are values. In C, they are pointers. A common (though not thread-safe) pattern in C libraries is to return a pointer to an internal static buffer. This buffer is overwritten or freed the next time the function is called.

If you were using raw pointers, this would be dangerous; your Perl variable would suddenly change its value or point to freed memory when you made a subsequent API call.

Affix solves this by copying string return values into Perl's memory space immediately.

The Recipe

In this example, we implement a C function that remembers its last result. It frees the old result before allocating a new one. We use a final call with undef to trigger a cleanup.

use v5.40;
use Affix qw[:all];
use Affix::Build;

# 1. Compile the C library
my $c = Affix::Build->new();
$c->add( \<<~'END', lang => 'c' );
    #include <stdlib.h>
    #include <string.h>

    const char * string_reverse(const char * input) {
        // Static pointer persists across function calls
        static char * output = NULL;
        int i, len;

        // Cleanup previous call's memory
        if (output != NULL) {
            free(output);
            output = NULL;
        }

        // Handle "Cleanup Mode" (NULL input)
        if (input == NULL)
            return NULL;

        // Allocate new buffer
        len = strlen(input);
        output = malloc(len + 1);

        // Reverse the string
        for (i = 0; input[i]; i++)
            output[len - i - 1] = input[i];
        output[len] = '\0';

        return output;
    }
END

my $lib = $c->link;

# 2. Bind the function
# String means "const char*" for input and output.
affix $lib, 'string_reverse', [String] => String;

# 3. Use it
# Input:  "\nHello world"
# Output: "dlrow olleH\n"
print string_reverse("\nHello world");

# 4. Cleanup
# Passing undef sends NULL to C, triggering the free() logic.
string_reverse(undef);

How It Works

• 1. Passing Strings (Input)

  [ String ] => ...
  

  When passing a Perl string to a String argument, Affix temporarily allocates a C buffer, copies the Perl string (encoded as UTF-8), and passes that pointer to C. This buffer is valid only for the duration of the call.

• 2. Returning Strings (Output)

  ... => String
  

  When a C function returns a String type, Affix reads the memory at that address (up to the first null byte), creates a new Perl scalar containing those bytes, and returns the scalar.

  It does not return a pointer to the C memory.

  This is crucial for this recipe. Even though the C function eventually calls free(output) on the next invocation, our Perl variable holding "dlrow olleH\n" remains safe because it is a distinct copy.

• 3. Passing Undef

  string_reverse(undef);
  

  For string arguments, undef is marshalled as NULL. Our C function checks for this to perform its cleanup duties.

Kitchen Reminders

• Copy vs. Reference

  If you want a copy of the text (standard Perl behavior), use the type String. If you want to modify the C memory buffer in place or hold onto the specific memory address, use Pointer[Char] (see Chapter 3).

• Thread Safety

  While Affix is thread-safe, C functions using static variables (like the one in this recipe) are generally not thread-safe. If two threads call string_reverse at the same time, they will race to free/overwrite the same static pointer.

C strings are simple: they start at a memory address and end at the first zero/null byte (\0). But what if your data contains nulls? If you use standard string types like String or Array[Char], C (and Affix) will assume the data ends at the first null byte. This is fatal for encryption, compression, and binary formats.

To handle this, we must stop thinking in characters and start thinking in bytes.

The Recipe

This is a demonstration of the XOR cipher. Since the data might contain nulls, we handle it as a raw buffer.

use v5.40;
use Affix qw[:all];
use Affix::Build;

# 1. Compile the C library
my $c = Affix::Build->new();
$c->add( \<<~'END', lang => 'c' );
    #include <stdlib.h>
    #include <string.h>

    char * string_crypt(const char * input, int len, const char * key) {
        char * output = malloc(len + 1);
        if (!output) return NULL;
        output[len] = '\0';

        int key_len = strlen(key);
        for (int i = 0; i < len; i++) {
            output[i] = input[i] ^ key[i % key_len];
        }
        return output;
    }

    void string_free(void * p) { free(p); }
END

my $lib = $c->link;

# 2. Bind the functions
# Return Pointer[Void] to get the raw address (Pin) without interpretation.
affix $lib, 'string_crypt', [ String, Int, String ] => Pointer[Void];
affix $lib, 'string_free',  [ Pointer[Void] ]       => Void;

# 3. The Wrapper
sub cipher( $input, $key ) {
    my $len = length($input);

    # Call C. We get back a Pin (unmanaged pointer).
    my $ptr = string_crypt( $input, $len, $key );
    return undef if is_null($ptr);

    # Create a view of the memory as an Array of Bytes (UInt8).
    # Important:
    #   Array[Char]  => Reads up to NULL (String behavior)
    #   Array[UInt8] => Reads exactly $len bytes (Binary behavior)
    my $view = cast( $ptr, Array[UInt8, $len] );

    # Dereference the view ($$view).
    # For byte arrays, Affix returns a binary string directly.
    my $binary_string = $$view;

    # Clean up the C memory immediately
    string_free($ptr);

    return $binary_string;
}

# 4. Run it
my $orig = 'hello world!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!';
my $key  = 'foobar';

my $enc = cipher( $orig, $key );
my $dec = cipher( $enc,  $key );

say "Original:  $orig";
say "Decrypted: $dec";

# Verify the binary data size
say "Encrypted length: " . length($enc);

use JSON::PP;
say "Encrypted blob:   " . encode_json([$enc]);

How It Works

• 1. Pointer[Void]

  We define the return type as Pointer[Void] to prevent Affix from automatically attempting to read it as a null-terminated string. We receive a raw Pin.

• 2. Casting to Array[UInt8]

  my $view = cast( $ptr, Array[UInt8, $len] );
  

  We use UInt8 (unsigned 8-bit integer) instead of Char. This tells Affix we are dealing with raw binary data, not text. Affix will read exactly $len bytes from memory, regardless of nulls.

• 3. Dereferencing

  my $binary_string = $$view;
  

  When you dereference a Pin defined as an Array of 8-bit types, Affix optimizes the operation. It copies the raw bytes directly into a Perl scalar.

Kitchen Reminders

• Char vs UInt8

  In Affix, Array[Char] is treated as a string buffer (stops at null). Array[UInt8] is treated as a byte buffer (reads full length). Use the one matching your data.

When interacting with the Windows API (Win32), you will inevitably face two facts: libraries are named things like user32.dll, and text is almost always expected to be UTF-16 (Wide Characters). In XS, bridging this gap involved tedious calls to MultiByteToWideChar and manual buffer management.

Affix handles the translation for you.

The Recipe

This recipe displays a native Windows message box containing Unicode characters (an emoji). It looks like this:

Okay, here's the sauce. ...the source:

use v5.40;
use utf8; # Required for literal emojis in source
use Affix;

# Define some standard Windows constants
use constant MB_OK                   => 0x00000000;
use constant MB_DEFAULT_DESKTOP_ONLY => 0x00020000;

# Bind the function
# Library: user32.dll
# Symbol:  MessageBoxW (The Unicode version)
# Alias:   MessageBox  (What we call it in Perl)
affix 'user32', [ MessageBoxW => 'MessageBox' ] =>
    [ Pointer [Void], WString, WString, UInt ] => Int;

# Call it
# undef becomes NULL (No owner window)
MessageBox( undef, 'Keep your stick on the ice.', '🏒', MB_OK | MB_DEFAULT_DESKTOP_ONLY );

How It Works

• 1. The A vs. W Suffix

  Most Windows functions come in two flavors: ANSI (ending in A) and Wide (ending in W). MessageBoxA expects system-codepage strings (like ASCII), while MessageBoxW expects UTF-16.

  Since modern Perl is Unicode-aware, we almost always want the W version.

• 2. Aliasing

  We bind the symbol MessageBoxW, but that is a clumsy name to type in Perl. By passing an array reference [ 'MessageBoxW' => 'MessageBox' ], we tell Affix to look up the real symbol in the DLL but install it into our namespace as MessageBox.

• 3. The WString Type

  This is the magic ingredient.

  [ ..., WString, WString, ... ]
  

  When you pass a Perl string to a WString argument, Affix performs the following steps on the fly:

  1. Check the internal encoding of the Perl scalar.
  2. Transcode the string into UTF-16LE (Little Endian).
  3. Append a double-null terminator (\0\0).
  4. Pass the pointer to the C function.

  This allows us to pass the hockey stick emoji '🏒' directly.

• 4. The Null Pointer

  The first argument to MessageBox is a handle to an owner window (HWND). Since we don't have a parent window, we pass NULL.

  In Affix, NULL is represented by Perl's undef. When passed to a Pointer type, undef is automatically converted to address 0x0.

Kitchen Reminders

• Input Only

  The WString type is designed for input strings (LPCWSTR or const wchar_t*). Affix creates a temporary buffer for the duration of the call. If the C function intends to modify the string buffer, you must use Pointer[WChar] and manage the memory yourself using calloc.

• use utf8

  If you are typing special characters like '🏒' directly into your Perl script, always remember use utf8; at the top of your file, or Perl might interpret the bytes incorrectly before Affix even sees them.

Sometimes, you don't have a pre-existing .dll or .so file. Sometimes, you just have an idea, a heavy mathematical loop, or a snippet of C code you found on Stack Overflow, and you want to run it right now inside your Perl script.

This is where Affix::Build shines. It turns your Perl script into a build system, a linker, and a loader, all in about five lines of code.

Let's look at a classic "Hello World" of pointer manipulation: the Integer Swap.

The Recipe

use v5.40;
use Affix;
use Affix::Build;

# 1. Spin up the compiler
my $c = Affix::Build->new();

# 2. Add some C code directly into the script
$c->add( \<<~'', lang => 'c' );
    void swap(int *a, int *b) {
        int tmp = *b;
        *b = *a;
        *a = tmp;
    }

# 3. Compile, Link, and Bind
affix $c->link, swap => [ Pointer [Int], Pointer [Int] ] => Void;

# 4. Profit
my $a = 1;
my $b = 2;
say "Before: [a,b] = [$a, $b]";

# Pass references so C can write back to Perl
swap( \$a, \$b );
say "After:  [a,b] = [$a, $b]";