Revert #1564

[illicitonion]

It looks like the reported issue is probably related to the ios
toolchain in use, rather than our general toolchain flag propagation.

[illicitonion]

@keith - I would be interested in your thoughts on the CI failure we get from reverting this PR...

The original PR filtered out the -target flag from the cc_toolchain when propagating to cargo_build_scripts. This is causing problems for toolchains which actually need the -target flag.

AFAICT what's actually currently needed to fix the issues with building iOS applications is to pass --cpu=ios_x86_64 on the command line, and that this is effectively a dupe of bazelbuild/bazel#16784 - that Bazel decides what cc_toolchain and flags &c to configure based on --cpu rather than --platforms...

Is there a nicer way of making the iOS stuff work that doesn't require setting the --cpu flag? Or do you have thoughts on what kind of manipulating of cc_toolchain stuff we should be doing to satisfy iOS builds? I'm pretty convinced that just stripping out -target is incorrect, but it's a little sad if setting --cpu is required as a workaround...

Otherwise, presumably we should start setting --cpu on both of these CI builds?

rules_rust/.bazelci/presubmit.yml

Lines 446 to 459 in ddf2a4c

	ios_examples:
	name: iOS Examples
	platform: macos
	working_directory: examples/ios
	build_targets:
	- "//..."
	ios_build:
	name: iOS build script cross compile test
	platform: macos
	working_directory: examples/ios_build
	test_flags:
	- "--platforms=//:ios_x86_64"
	test_targets:
	- "//..."

- the first one doesn't appear to need it because the code involved isn't complex enough (or we don't have sufficiently pedantic asserts), but the second does seem to...

[UebelAndre]

Seems a test needs to be updated for iOS?

[illicitonion]

Seems a test needs to be updated for iOS?

Yeah - the test is fixed by adding --cpu=ios_x86_64 to the iOS example build, which I think is the correct fix for the issue, but I don't know the iOS stuff well enough to know if there's a better one...

[dae]

Sorry about the unintended breakage; no objections from me on reverting the change.

[UebelAndre]

@dae would you be able to confirm what the right solution is for #1663 (comment) ?

[dae]

If @illicitonion confirmed the test I added passes when --cpu=ios_x86_64 is provided, that sounds like the best solution, as it addresses the problem at the source, and it's the solution I would have used had I known specifying a platform was not enough.

[illicitonion]

/cc again @keith post-Thanksgiving - I'm adding a --cpu flag to the iOS Rust example bazelrc files - does that look reasonable to you, or is there a better way to make the iOS toolchain pieces work?

[keith]

What part of this change breaks it in that way (or maybe it was just broken at HEAD?)

For ios_build I think an issue is not having a platform_mappings file, did the issue appear for both ios and ios_build? (not sure why those are separate either

[keith]

Overall I guess I don't think passing that flag hurts for various examples here, but I'm not sure what impact this might have on downstream users

[illicitonion]

Aha - yeah, it looks like we were missing the platform_mappings file, and instead were directly setting --platforms in the CI for the ios_build example - we were only seeing the issue in that one repo. Thanks for looking!

I added a platform_mappings, and changed the .bazelci file to set the flags mentioned in the platform_mappings file instead of the one it sets. This fixes things up :)