Summary: | ccache doesn't notice when split-specs compilers are switched | ||
---|---|---|---|
Product: | Gentoo Linux | Reporter: | Kevin F. Quinn (RETIRED) <kevquinn> |
Component: | [OLD] Development | Assignee: | Lisa Seelye (RETIRED) <lisa> |
Status: | RESOLVED UPSTREAM | ||
Severity: | normal | CC: | solar, toolchain |
Priority: | High | ||
Version: | unspecified | ||
Hardware: | All | ||
OS: | Linux | ||
Whiteboard: | |||
Package list: | Runtime testing required: | --- | |
Bug Depends on: | 128810 | ||
Bug Blocks: | |||
Attachments: |
Adds contents of specs files listed in GCC_SPECS to the hash.
add output of '<compiler> -v' to the hash |
Description
Kevin F. Quinn (RETIRED)
2006-02-02 11:38:35 UTC
Created attachment 78744 [details, diff]
Adds contents of specs files listed in GCC_SPECS to the hash.
hmm; occurs to me this isn't good enough, as we set GCC_SPECS in the gcc wrapper, in which case ccache won't see it. I think to be complete, we'd need to some or all of: 1) output of 'gcc -v' 2) output of 'gcc -dumpspecs' 3) contents of spec files listed by 'gcc -v' in practice it might be enough just (1). Created attachment 78754 [details, diff]
add output of '<compiler> -v' to the hash
Ok; this is a simple enough patch to add the output of '<compiler> -v' to the hash.
I think this is enough to capture changes in compiler driven by our specs trickery. It doesn't actually hash the specs themselves, but I don't think that's necessary.
You should submit the patch upstream to the ccache mail list, http://lists.samba.org/mailman/listinfo/ccache/ This should have been closed UPSTREAM as it is an upstream improvement request. Well, as I said before, the problem is caused by the GCC_SPECS patch we apply to our compilers to support switching specs in the environment. Without that patch there's no need to hash the output of 'gcc -v'. This means that as far as upstream are concerned, it's not relevant - unless they maintain distribution-specific variations (which I think they don't). Also I'm not sure they would be keen on adding a separate invocation of the compiler driver to every ccache call. Perhaps "WONTFIX" might be a better resolution. |