Jets FFI fails on MacOS

[sanket1729]

Running cargo test on fbe88d6f703c2f3b74f52eb3eebf60411606ed42 . This need not be prioritized, but mainly an informational issue.

failures:
    bit_machine::tests::crash_regression1
    human_encoding::parse::tests::bip340_program
    human_encoding::parse::tests::simple_program
    human_encoding::tests::filled_witness
    jet::elements::tests::test_ffi_env
    jet::tests::test_ffi_jet
    node::commit::tests::shared_witnesses
    policy::satisfy::tests::satisfy_after
    policy::satisfy::tests::satisfy_and
    policy::satisfy::tests::satisfy_asm
    policy::satisfy::tests::satisfy_older
    policy::satisfy::tests::satisfy_or
    policy::satisfy::tests::satisfy_pk
    policy::satisfy::tests::satisfy_sha256
    policy::satisfy::tests::satisfy_thresh
    policy::serialize::tests::execute_after
    policy::serialize::tests::execute_and
    policy::serialize::tests::execute_and_true
    policy::serialize::tests::execute_older
    policy::serialize::tests::execute_or
    policy::serialize::tests::execute_pk
    policy::serialize::tests::execute_sha256
    policy::serialize::tests::execute_threshold

[apoelstra]

Could this be an endianness issue?

[sanket1729]

My system is little endian. Which I think is the most common and the one rust-simplicity expects?

[sanket1729]

Okay, I am failing the sanity_checks for FFI. Maybe, we can somehow make this a compile time check.

std::mem::size_of::<crate::ffi::UWORD>() != crate::ffi::c_sizeof_UWORD

[sanket1729]

I am not familiar with MacOS, but it looks like, UWORD uint_fast16_t typically expands to 2 byte. In contrast, on Linux, it often expands to 8 bytes (64 bits), where 64-bit operations are considered faster or more efficient given the architecture.

When we wrote the bindings, we assume UWORD which is uint_fast16_t maps to usize

[sanket1729]

fast_* types which don't really define particular sizes are really hard to work with FFI across systems :( .

[sanket1729]

And as expected changing

- pub type c_uint_fast16_t = usize;
+ pub type c_uint_fast16_t = u16;

fixes the issue. Figuring out a programmatic way to do this might be annoying. Maybe the new std::ffi has things that can help with fast* type