[cfe-dev] PList output for clang Tidy

[Gábor Horváth]

On 5 January 2015 at 11:02, Manuel Klimek <klimek at google.com> wrote:

> +chandler and daniel; I think that are all the chefs we need for this
> particular porridge
>
> On Mon Jan 05 2015 at 10:05:02 AM Gábor Horváth <xazax.hun at gmail.com>
> wrote:
>
>> Hello everyone,
>>
>> The Clang Static analyzer can output the diagnostics in plist format. It
>> is a useful feature, because it is easy to parse that format with 3rd party
>> tools, hence integrating clang tools with others.
>>
>> Unfortunately the plist reporting format is not supported by Clang Tidy.
>> We would like to add plist support to it. This involves a lot of changes
>> both to the format and the public API, so I want your opinion, how to do it.
>>
>> In my opinion we need to extend the plist format to:
>> * Support notes that are not events
>> * Support fixits
>>
>> What do you think, what would be the bast way to extend the format with
>> those informations?
>>
>> The plist reporting related functionality is not part of the Clang public
>> API at the moment. The best would be, if the Static Analyzer and regular
>> diagnostics could be reported to the same plist output file. To achieve
>> this, the diagnostic consumer that outputs to the plist should support both
>> PathDiagnostics and regular Diagnostics. It would be redundant to
>> reimplement the whole functionality in Clang Tidy. To reduce the
>> redundancy, we would like to refactor several plist related helper
>> functions out from PlistDiagnostics and make it available in public
>> headers. We would also like to make PlistDiagnostics class available so
>> we can inherit from it. What do you think, what would be the best way to
>> organize these changes?
>>
>
> If I remember correctly, Chandler has long ago proposed to merge the
> different diagnostic types we have into one central clang diagnostic type
> that supports all our use cases.
>
> I personally would need to see a CL to judge whether it makes sense, but
> generally, I think what you say sounds like it's the right direction (if
> you agree with the sentence above: make clang's basic diagnostic system
> powerful enough to support the analyzer's use cases, and then switch the
> analyzer and clang-tidy to use it).
>
>
In my proposal it would be possible to create a Diag Consumer that can
consume both PathDiagnostics and regular Diagnostics. It would be possible,
to make the same object to consume both Clang and Static Analyzer diags. It
may not be as clean as merging the two kinds of diagnostics, but I suspect
this approach is faster to implement. In the long term I do agree that, it
would be desirable to use the same mechanisms for both Static Analyzer and
Clang. But I am not sure that, I have enough time for a refactoring like
that.


>
> I'd vote for renaming PList to something non-horrible in the process, (I
> still have no idea what the "P" stands for), but that's bikeshedding.
>

I don't know if we could rename that. PList is not Clang specific, Apple
uses this format in other projects as well.


>
> Cheers,
> /Manuel
>
>

Thanks,
Gábor
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20150105/ef1b4754/attachment.html>