[PATCH] D51729: [Tooling] JSONCompilationDatabasePlugin infers compile commands for missing files

[Sam McCall via cfe-commits]

Sorry about that. I wasn't familiar with the python bindings.
Your bisect is correct, we changed the behavior that test is testing: now
an approximate match will be returned.

I guess either:
 - just remove that test (depending on how we feel about smoke-testing vs
exhaustively testing wrapperl like this)
 - change it to verify exactly what the fuzzy-matched result is, e.g.
    - add -DFEATURE=0 to the first compile command for "Project.cpp" in the
test CDB so it's easy to recognize
    - make the lookup key "/home/john.doe/MyProject/Project.h" so it's
obvious what the match should be
    - verify that the lookup result has "-DFEATURE=0" and "Project.h" in
its args
It seems to me that test_lookup_succeed should also verify what gets
returned (otherwise it's not really testing anything at this point).

I can try to put together a patch, though I don't currently have a setup to
test it - let me know if this would be useful!

On Fri, Oct 12, 2018 at 8:38 AM Michał Górny via Phabricator <
reviews at reviews.llvm.org> wrote:

> mgorny added a comment.
>
> Unless my bisect is mistaken, this broke Clang Python binding test:
>
> https://github.com/llvm-mirror/clang/blob/master/bindings/python/tests/cindex/test_cdb.py#L34
>
> The test apparently assumes that compilation database will not return
> anything for a file that does not exist. Could you please suggest how the
> test should be adjusted for this change?
>
>
> Repository:
>   rC Clang
>
> https://reviews.llvm.org/D51729
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20181012/acab3414/attachment.html>