[cfe-dev] compilation database for tooling

[Manuel Klimek via cfe-dev]

On Fri, Feb 19, 2016 at 12:58 AM Laszlo Nagy <rizsotto.mailinglist at gmail.com>
wrote:

> hi guys,
>
> would like to come back to this topic... the inconsistency between the
> compilation database [CDB] implementation and the documentation about the
> file paths.
>
> would like to clarify what is our expectation about CDB file paths. at the
> same time would like to pin down these as test cases. and fix the
> implementation when test cases are not passing.
>
> so, for the first part: #1 shall CDB portable and allow relative paths?
> (it means when i generate a CDB and then i move my sources into another
> directory (with the CDB), tools shall be happy and run as before the move.)
> #2 shall CDB support mixture of relative paths and absolute paths? (eg.:
> the `directory` field in CDB use absolute path, while include paths are
> relative in the `command` field.) #3 shall CDB support absolute paths too?
> (my guess that's the only supported case at the moment.)
>

The CDB should support:
- absolute paths in directory (relative paths there seem to brittle)
- relative & absolute paths everywhere else


> so, for the test cases: i checked the current functional test against CDB.
> have not found specific test in the `clang-tools-extra` repo, but find a
> few in `clang` source tree. #4 where test cases shall go? my other
> questions about `lit`. CDB is a JSON file. using `lit` usually works as
> adding comment section. JSON does not support comments. (but have seen
> 'test/Index/skip-parsed-bodies/compile_commands.json' in clang repo that
> using it.) #5 will such file parsed? and when i'm writing test with
> absolute paths, i can't wire in the current absolute path into the file.
> so, thought to use CMake to substitute those values. #6 but then will `lit`
> find the generated test files in the build dir too? (or just scans the
> source dir?)
>

For lit you give it a couple of RUN files at the start of the file anyway.
In there, you can just strip the comments out to create the actual CDB
yourself.


>
> thanks,
> Laszlo
>
> On Fri, Feb 5, 2016 at 11:37 PM, Manuel Klimek <klimek at google.com> wrote:
>
>> +clang-tidy folks, as this seems to be more of an issue with clang-tidy
>> than anything else
>>
>> On Fri, Feb 5, 2016 at 12:27 PM Laszlo Nagy via cfe-dev <
>> cfe-dev at lists.llvm.org> wrote:
>>
>>> hi Everyone,
>>>
>>> i have a question about compilation database (CDB). i wrote tools which
>>> generate CDB. it follows the format what is mentioned in the
>>> documentation <http://clang.llvm.org/docs/JSONCompilationDatabase.html>.
>>> my question related to the source file. a source file is mentioned two
>>> fields for a single entry, the 'file' and the 'command' fields. i remember
>>> a discussion earlier to make the source file to be relative to the
>>> directory (for portability). but at the same time, i receive complains
>>> <https://llvm.org/bugs/show_bug.cgi?id=24710> about my tool when it
>>> generates non absolute paths (in the 'command' field).
>>>
>>> to make it more fun, the 'directory' field could be also relative or
>>> absolute (by the documentation). but two years ago, that was the first
>>> thing i needed to fix (and always use absolute path).
>>>
>>> can somebody tell me what is the current status of the file paths in
>>> CDB? is there any test suite around the CDB reading in the Clang repo? or
>>> shall i just blindly copy the CMake generated CDBs and ignore the
>>> documentation?
>>>
>>> thanks,
>>> Laszlo
>>> _______________________________________________
>>> cfe-dev mailing list
>>> cfe-dev at lists.llvm.org
>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20160219/d89e8130/attachment.html>