rustc 1.30.0's linker flavor inference is a non-backwards compat change to -Clinker

[jsgf]

In our environment we end up setting -Clinker to a script called ld.sh; this script is implemented in terms of gcc as rustc used to expect. Now it treats it as a direct invocation of ld.

This was introduced in #52101.

I think rather than inferring the linker flavor from the name of the executable, rustc should either:

1. invoke the binary with --version to try to work out what its dealing with, and/or
2. have a stable -Clinker-flavor option to set it explicitly

There's an unstable -Zlinker-flavor option, which I think should have been stabilized at the same time as flavor inference. Ideally inference should be opt-in with -Clinker-flavor=infer, with the default being the old default of -Clinker-flavor=gcc (at least for Linux platforms).

[alexcrichton]

Oh dear, sorry for the breakage! I think there's probably a few things we can do to help mitigate this:

• For backwards compatibility, inference should only chop off *.exe and not other extensions when inferring a flavor. I think we'll still want to infer the flavor by default as it fixed a number of issues and empowered a number of embedded use cases to become stable. I also think we want to avoid running --version as running a binary has quite a bit of overhead unfortunately
• Simultaneously I think we should likely stabilize -C linker-flavor as mentioned here to allow explicit configuration via the command line.

@jsgf does it work for now for y'all to rename to gcc.sh? Do the above steps sound ok to you?

Also, cc @japaric

[jsgf]

The script is currently called cxx-gcc-5-glibc-2.23-clang-ld.sh so I wonder if the inference heuristic should be tweaked a bit. I think in the meantime I'll try adding -Zlinker-flavor=gcc rather than adjusting the name of the script.

[alexcrichton]

@jsgf ah yeah the heuristic includes looking for -ld at the end, and we chop off the file extension to handle .exe on Windows. You should be able to work around it locally by renaming to -gcc.sh.

[pnkfelix]

triage: P-high.

[pnkfelix]

assigning to @nagisa to take point on moving forward on how we're going to address this

[pnkfelix]

T-compiler triage: this has one work-item PR (#56349) that landed, and then the final step is in PR #56351, which has gotten approval via rfcbot so I think all that remains is to bors r+ PR #56351 ?

[pnkfelix]

T-compiler triage. PR #56351 landed, so closing this as fixed.

[jsgf]

Thanks everyone!