[cfe-commits] cfe-commits Digest, Vol 47, Issue 5

[Matthieu Monrocq]

Shameless bump:

clang_stringrefize_diags_plus.diff replaces char const* by llvm::StringRef
in the DiagnosticIDs interface, and updates the ProcessWarningOptions in
lib/FrontEnd/Warnings.cpp  and PragmaDiagnosticHandler::HandlePragma in
lib/Lex/Pragma.cpp to take advantage of this.

Please review.

2011/5/1 Matthieu Monrocq <matthieu.monrocq at gmail.com>

> Date: Sun, 1 May 2011 15:45:33 +0200
>> From: Matthieu Monrocq <matthieu.monrocq at gmail.com>
>> Subject: [cfe-commits] Improving Clang Diagnostics [1.3]: StringRefize
>>        DiagnosticIDs API
>> To: cfe-commits at cs.uiuc.edu
>> Message-ID: <BANLkTim=6Zk6VYSWQ0-EDKZ3xYUJmATktw at mail.gmail.com>
>> Content-Type: text/plain; charset="iso-8859-1"
>>
>> Hello,
>>
>> This patch StringRefize DiagnosticIDs API (and internals).
>>
>> I had some troubles because of macro expansion: not easy to see that a
>> StringRef was built from a null pointer and track down the crash... but
>> could the issue via an assert in the StringRef class itself. I've
>> submitted
>> the patch for this already, hopefully it'll get applied so that others
>> have
>> an easier time figuring out what's going wrong when they introduce
>> StringRef.
>>
>> Apart from this, there are two functions that could perhaps benefit from
>> deeper changes, in the clients of DiagnosticIDs:
>> - ProcessWarningOptions in lib/FrontEnd/Warnings.cpp, which deals with
>> processing the -W family of options. There is a lot of manipulation of
>> OptStart / OptEnd pointers that I think could be easily translated into
>> using StringRef
>> - PragmaDiagnosticHandler::HandlePragma in lib/Lex/Pragma.cpp, I believe
>> WarningName could be a StringRef, but I need to investigate what
>> Literal.GetString returns to be sure it's possible
>>
>> I'll handle this as two follow-up patches, dependent on this one, unless
>> you
>> think it's better to merge the changes into this patch.
>>
>> Please review.
>> Matthieu.
>> -------------- next part --------------
>> An HTML attachment was scrubbed...
>> URL:
>> http://lists.cs.uiuc.edu/pipermail/cfe-commits/attachments/20110501/51ef5ee2/attachment.html
>> -------------- next part --------------
>> A non-text attachment was scrubbed...
>> Name: clang_stringrefize_diags.diff
>> Type: application/octet-stream
>> Size: 16999 bytes
>> Desc: not available
>> Url :
>> http://lists.cs.uiuc.edu/pipermail/cfe-commits/attachments/20110501/51ef5ee2/attachment.obj
>>
>> ------------------------------
>>
>
> I had some time on my hands so I went ahead and did the patches.
>
> The patch in lib/Lex/Pragma.cpp is relatively innocuous and to be frank
> does not bring much to the table (syntaxically), it removes a construction
> of  std::string  anyway so might be worth it.
>
> The patch in lib/Frontend/Warnings.cpp is much more important. It really
> cleans up the code consequently, trading awkward pointer differences and
> manual call to memcmp for  llvm::StringRef  smart interface (especially for
> prefix comparisons).
>
> Both patches are based on the patch mentionned above.
>
> I also provide a "all-in-one" patch (clang_stringrefize_diags_plus.diff),
> which bundles the StringRefication of DiagnosticIDs API with those two
> patches, for simpler application.
>
> Please review.
> Matthieu.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20110507/139136db/attachment.html>