use NativeEndian in symbolize::gimli::Context

[cuviper]

Object uses NativeEndian, so the Context should too.

Cc: rust-lang/rust#77410

[camelid]

Should this comment be updated? Isn't gimli the default now? (Not necessary to do here, just wondering.)

[cuviper]

I do see big-endian powerpc64 in CI, so I'm not sure how this has been passing -- does it really run the tests?

• before: https://github.com/rust-lang/backtrace-rs/runs/1183751745?check_suite_focus=true#step:6:897
• now: https://github.com/rust-lang/backtrace-rs/pull/373/checks?check_run_id=1228634621#step:6:978

I used native s390x to confirm the problem and this fix.

[cuviper]

Oh:

backtrace-rs/ci/docker/powerpc64-unknown-linux-gnu/Dockerfile

Lines 13 to 15 in 2b51e00

	# TODO: should actually run these tests
	#CARGO_TARGET_POWERPC64_UNKNOWN_LINUX_GNU_RUNNER="qemu-ppc64 -L /usr/powerpc64-linux-gnu" \
	CARGO_TARGET_POWERPC64_UNKNOWN_LINUX_GNU_RUNNER=echo \

[alexcrichton]

Thanks! You've confirmed locally this fixes the issue too?

I actually thought that ELF was always little-endian for common platforms, regardless of the endianness of the platform itself, but this would make sense why things are broken on big-endian!

[philipc]

I actually thought that ELF was always little-endian for common platforms, regardless of the endianness of the platform itself

I think that's true for PE but not for ELF.

[cuviper]

Thanks! You've confirmed locally this fixes the issue too?

On s390x, yes.

I actually thought that ELF was always little-endian for common platforms, regardless of the endianness of the platform itself, but this would make sense why things are broken on big-endian!

AFAIK it's always native in ELF, and from a quick skim of glibc it looks like the loader asserts that they match:
https://sourceware.org/git/?p=glibc.git;a=blob;f=elf/dl-load.c;h=0c8fa72c4d1d2396f956ce3eb64c02b425b6fd22;hb=HEAD#l1713

Any recollection why that powerpc64 CI has the qemu-ppc64 line commented out?

[cuviper]

I think that's true for PE but not for ELF.

Hmm, then should we make this import part of the platform-specific cfg_if?

[philipc]

Hmm, then should we make this import part of the platform-specific cfg_if?

This import seems to only be used for the DWARF, and I don't know what endianness DWARF in PE uses. My intuition is that NativeEndian would be correct for it too.

[alexcrichton]

Ok sounds good! I'm gonna go ahead and merge this since it fixes an issue for rust-lang/rust, we can sort out CI afterwards if necessary. PE I think only affects Windows (right?) and while Gimli is used for MinGW targets I don't think that there's any big-endian Windows targets right now.

Any recollection why that powerpc64 CI has the qemu-ppc64 line commented out?

Ah I'm sure at some point in time in the past it was broken, but it was most likely a bug in qemu. If it works now seems good to uncomment!

[mati865]

I think that's true for PE but not for ELF.

Correct, docs: https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=vs-2019#endianness

There are no official Rust MinGW ARM targets yet.
There are C/C++ MinGW ARM/Aarch64 targets though and some folks would like to see official Rust target already.

[alexcrichton]

I'm pretty sure arm/aarch64 are little endian...

[nagisa]

I'm pretty sure arm/aarch64 are little endian...

ARM-the-architecture can go both ways, though its not awfully useful as the ecosystem around it pretty much universally uses LE. Context.