Improve `size_t` detection using LLVM data layout instead of triple string matching

[m-akhil-reddy]

Summary

Summary

Currently size_t is determined in _get_size_t_type_from_triple() by checking architecture names inside the target triple string. While this works for common architectures, LLVM already exposes pointer size information through target_machine.target_data, which could be used to determine the correct size_t width more reliably.

Using LLVM’s data layout instead of string matching may simplify the implementation and make it more robust across different targets.

Current Implementation

In src/irx/builders/llvmliteir.py, _get_size_t_type_from_triple() determines the width of size_t by matching architecture names in the target triple.

Example logic:

• x86_64, aarch64, arm64, etc → 64-bit
• i386, i686, arm, etc → 32-bit
• fallback → host ctypes size

This requires maintaining architecture string checks manually.

Proposed Improvement

Instead of inferring the width from the triple string, it may be possible to derive size_t directly from the LLVM target data layout.

Example idea:

ptr_size = self.target_machine.target_data.pointer_size
return ir.IntType(ptr_size * 8)

This would rely on LLVM’s target information rather than architecture string matching.

Benefits

• More robust across different architectures
• Removes manual triple string checks
• Automatically supports additional targets
• Simpler and cleaner implementation

Additional Context

This builds on the work from PR #134 which fixed issue #132 by:

• setting module.triple
• setting module.data_layout
• removing the forced 64-bit size_t override

If this approach sounds reasonable, I’d be happy to work on implementing it and adding tests.

Additional Information

Code of Conduct

• I agree to follow the Code of Conduct

[m-akhil-reddy]

@xmnlab @yuvimittal ,
presently i am working on this issue .
can you assign it to me.