[cfe-dev] [clang-tidy] problem with relative paths in compile_commands.json

[Manuel Klimek]

+Alexander Kornienko <alexfh at google.com>

On Fri Feb 20 2015 at 12:18:49 AM David Blaikie <dblaikie at gmail.com> wrote:

> +Sam who's been doing a fair bit with clang-tidy recently (Sam, feel free
> to add Manuel, Daniel, whomever else might be more appropriate if you
> prefer)
>
> On Thu, Feb 19, 2015 at 2:57 PM, Johann Klähn <kljohann at gmail.com> wrote:
>
>> clang-tidy seems to have problems with relative paths in command lines
>> in compile_commands.json.
>>
>> Consider the following case:
>>
>> $ cat > test.cpp
>> int main(int argc, char* argv[]) {
>>     return 0;
>> }
>> $ cat > compile_commands.json
>> [
>> {
>>   "directory": "/home/kljohann/test/build",
>>   "command": "/usr/bin/clang++ -Wunused-parameter -o test.cpp.o -c
>> ../test.cpp",
>>   "file": "/home/kljohann/test/test.cpp"
>> }
>> ]
>> $ mkdir build
>> $ clang-tidy test.cpp
>> 2 warnings generated.
>> clang-tidy: ../tools/clang/lib/Basic/SourceManager.cpp:422: const
>> clang::SrcMgr::ContentCache*
>> clang::SourceManager::getOrCreateContentCache(const clang::FileEntry*,
>> bool): Assertion `FileEnt && "Didn't specify a file entry to use?"'
>> failed.
>> [1]    23129 abort      clang-tidy test.cpp
>>
>> My guess is the following: In ClangTool::run we chdir() into the
>> "directory" that is specified in compile_commands.json. Then, the
>> relative file name is stored in ClangTidyMessage (I checked that this
>> is the case using gdb). Eventually handleErrors() is called, but by
>> then we already chdired back and thus in ErrorReporter::getLocation
>>   const FileEntry *File = SourceMgr.getFileManager().getFile(FilePath);
>> is a null pointer (checked using gdb).
>> On the other hand this is my first venture into the clang source code,
>> so I may very well be wrong. I hope you can reproduce this bug.
>>
>
> Yeah, I can reproduce this. I know little about clang-tidy and what the
> expected behavior is here - certainly crashing isn't ideal, but even beyond
> that I'm not sure if the assumption is taht all the paths in the compile
> commands database will be absolute, or relative to some specific place, etc.
>
> The CMake compilation database support seems to create databases with all
> absolute paths so I imagine this doesn't come up there.
>
>
> - David
>
>
>>
>> All the best,
>> Johann
>> _______________________________________________
>> cfe-dev mailing list
>> cfe-dev at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>>
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20150220/e74abd7a/attachment.html>