Today work was focused on three things.
Trying to figure out why function pointer checks are skipped in Windows. This happens in the libc crate’s test suite. It ensures the function pointers we get from C are equivalent to those we bind to in Rust. This makes it possible to catch mismatches in identifier link names and expected ABIs.
Trying to test on Fuchsia. This follows from yesterday’s attempts. The goal here is to make sure the changes in my Fuchsia PR are relatively correct.
Trying to find the AIX headers. This is part of ongoing efforts to check time_t and file offset types across all supported targets.
I’ve miserably failed on all three fronts.
The function pointer checks on Windows remain a mystery. A number of routines follow the cdecl convention. Others expecte a “C” calling convention. This is supposed to be documented in the MSVC docs. Yet for some reason the function pointers are mismatched between Rust and C. We currently bind to them all under the “C” ABI. This is potentially wrong. The symbol is meant to have both a decorated name and an undecorated name. The latter does not seem to correspond with the symbol’s link name. Otherwise just changing that in Rust would do. I’ve also tried gating a number of those routines under extern "cdecl" blocks for x86 only. This doesn’t work either. It yields undefined references at link time on the C side of things (from the generated C test files using the link name.) I’ve also tried using the ctest methods for renaming functions on the Rust side of things. This doesn’t work either. The routines link just fine under the "C" ABI. all other ABIs (applicable mostly to x86) cause linking errors even on supported x86 Windows targets.
A question from my mentor propmted me to attempt to solve this. It’s not part of the proposed timeline. I won’t be attempting to solve it again in some time.
Testing on Fuchsia seemed simple. It is simple. It just so happens the instructions don’t align with the (my) reality. They seem to have a binary/package/repo pipeline set up in Fuchsia. You have to configure a bunch of Fuchsia-specific files after compiling your Rust crate. These are all package metadata-related files. You’re then meant to create a local repository of packages. These are then registered on the repository. The emulator is then launched with parameters specifying both your local repository and the package. It’s a bit involved. The issue lies in a mismatch between the compiled artifacts I get from the Fuchsia-specific metadata compiler and the artifacts the instructions get. There’s meant to be a package that can be registered with the repository. That’s just nowhere to be found in my metadata compilation artifacts.
I’ve not found the AIX headers. I genuinely don’t know why IBM doesn’t provide a downloadable SDK with native headers. They already have a bunch of pages documenting the headers. Those are insufficient to ensure we bind to the correct definitions.
None at present.
The timeline will quite definetely be met. The only modules remaining in the unix module are the hurd and nuttx modules. These should be fairly simple to go through. The proposal expected this to be finished by the sixth to seventh week. It will be finished by the sixth week. Work on the aix module will stop. This should leave plenty of time for the GNU Hurd and NuttX.