rustc: Fail immediately if linking returns status code != 0

[msiemens]

If rustc fails to link a library (as described in #17951), it still tries to rename the non-existent file on Windows. I'm not sure how to test this as we'd have to reproduce a linking failure...

Closes#17951

[rust-highfive]

Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @nikomatsakis (or someone else) soon.

[alexcrichton]

I think this may actually be somewhat of a red herring, the fs::rename(..).unwrap() should always succeed if the linker succeeded, but in this case the linker didn't succeed.

Could you instead modify line 717 (above this change) to actually look at the status returned by .status()? The return code is currently ignored as Ok(_), but we should test it for status.success() and fail immediately if it didn't have a 0 exit status.

[msiemens]

Sure, will do.

[msiemens]

@alexcrichton I've added the changes you proposed. For that I reverted my original commit. Would it be better if I remove it completely?

[alexcrichton]

Thanks @msiemens! Could you squash the commits together (removing the revert), and also it may be helpful to print the command line that failed (just use the Show impl for cmd)

[msiemens]

Done. Is the new error message okay?

[nikomatsakis]

(Sorry, to be clear, that was just a "for future reference" sort of comment, not a criticism or reason not to land the patch.)