MachO support

[djc]

Would be awesome if this project at some point grew support for Darwin-based systems like macOS.

Enjoyed your RustWeek talk, figured I'd at least register interest for this in the issue tracker. 👍

[lu-zero]

Would be nice to provide a checklist on what is needed and see if some quick preparation work could be contributed.

[davidlattimore]

On Linux, our trivial binaries are statically linked executables, but given that Mac (and Windows) don't as far as I know have a stable interface to the kernel, it might be better for the trivial binaries to be dylibs. That way the dylib can just define a function that returns a number. Adding support for this kind of test (a shared object that gets opened and executed) is something that could be done before everything else - whoever is doing it would need Linux though.

After that, the next next step would be to move as much of the elf-specific code as possible behind a trait. Working out what is and isn't elf-specific is likely to be a bit of guesswork, so the trait would certainly need adjusting as it starts to actually get a second implementation.

Then work on producing a trivial dylib test that passes, probably in conjunction with adding basic macho support to linker-diff.

[marxin]

I've started working on the MachO port, where I collected the early observations and wrote them down in a document:
https://github.com/marxin/wild/blob/macos/macOS.md

[marxin]

@davidlattimore I think I'll need a bit of Platform refactoring related to the build_output_order_and_program_segments (and DEFAULT_SECTION_RULES) functionality where we define the single part sections (and their corresponding rules). I would like to use the very same concept, but with a different section names and segment names. Can you work on that?

[davidlattimore]

Happy to help. Could you elaborate a bit?

[davidlattimore]

I moved build_output_order_and_program_segments - #1750.

DEFAULT_SECTION_RULES was already previously moved. You can implement default_layout_rules on Platform.