GCC Cross Compiler Strip

[teburd]

Describe the bug
Currently cross compiler toolchains using gcc set dontStrip = true. This leads to larger closures, especially for C++ projects targeting other architectures using musl as the libc.

To Reproduce
Steps to reproduce the behavior:

1. build any software with crossSystem set to a different target arch, see gcc has -debug added, and all debug symbols
2. If you choose musl and build a C++ project you'll see the unexpected closure contained glib through the debug symbols and musl -dev package containing scripts pointing to bash which needs glibc.

Expected behavior

1. Only build -debug variant when override is set
2. Not have stray references to glibc from musl-dev (this can be done by setting dontPatchShebangs in musl's build)

Additional context
The root of the problem seems to stem from strip.sh using the same strip binary for both libraries and binaries, where in a cross compiler toolchain what probably should happen is the target strip should be used against libraries, and the host strip should be used against binaries.

Notify maintainers
@Synthetica9

[teburd]

I'm happy to sponsor a fix for this if its reasonable!

[teburd]

@matthewbauer thought you might be interested in this one

[matthewbauer]

Yeah! So I can definitely merge #108060, but you're right that it doesn't actually strip true cross-compilation.

@Ericson2314 originally introduced the TARGET_STRIP & STRIP distinction. It's possible we can just set "stripped" always to true, but I would want to make sure we aren't running "strip" on the wrong things.

[Ericson2314]

Yeah TARGET_STRIP & STRIP was something I made for this purpose, but never found time to put it at the test. At this point, I rather invest in the real solution which is building GCC itself and the target libraries separately. This will improve and simplify things on so many levels.

[stale]

I marked this as stale due to inactivity. → More info

[teburd]

Not stale

[stale]

I marked this as stale due to inactivity. → More info

[teburd]

Not Stale, Still Relevant

[Mindavi]

I thought this was done to prevent broken binaries, since (IIRC correctly) ARM binaries were sometimes broken when stripped. But maybe the situation has improved?

[minijackson]

So I tried cross-compiling a minimal all-stripped armv7l NixOS image and tested it a bit on a Raspberry PI 2, and it boots and works without any issue. Note that comments in compilers' default.nix state:

Strip kills static libs of other archs (hence no cross)

I'm not sure how to test that.

[trofi]

Possible fix for cross-compiler stripping: #182513