r177218 - Remove -Wspellcheck and replace it with a diagnostic option.

Argyrios Kyrtzidis akyrtzi at gmail.com
Sat Mar 16 10:32:32 PDT 2013

On Mar 16, 2013, at 10:18 AM, David Blaikie <dblaikie at gmail.com> wrote:

> On Sat, Mar 16, 2013 at 10:13 AM, Dmitri Gribenko <gribozavr at gmail.com> wrote:
>> On Sat, Mar 16, 2013 at 3:40 AM, Argyrios Kyrtzidis <akyrtzi at gmail.com> wrote:
>>> Author: akirtzidis
>>> Date: Fri Mar 15 20:40:35 2013
>>> New Revision: 177218
>>> URL: http://llvm.org/viewvc/llvm-project?rev=177218&view=rev
>>> Log:
>>> Remove -Wspellcheck and replace it with a diagnostic option.
>>> Thanks to Richard S. for pointing out that the warning would show up
>>> with -Weverything.
>> If we are going to start testing clang this way, it would be better to
>> design this first, so that adding new 'testing' diagnostics is easy
>> *and* does not slow down the normal compilation.  I think the second
>> part is addressed already.
>> For example, adding a command line option every time is excessive.
>> This option could be renamed to -fclang-debugging-diagnostics, and all
>> such diagnostics could be placed under a special flag
>> -Wclang-debugging.
> I still don't understand the need for this at all. At a glance it
> seems like we're adding a positive diagnostic so we can check for the
> absence of a diagnostic - but we've never had a need to do this in the
> past. "-verify" fails if a diagnostic is emitted where it isn't
> expected so the absence of expected-blah lines is sufficient to test
> that we don't emit a diagnostic.
> Am I missing something here? Why are we doing this?

This code snippet of an objc method

-(void)objc_method: {
  super.x = 0;

would trigger typo-correction for 'super' silently, without emitting any diagnostic.
For the regression test I added I put:

typedef int super1;

so typo-correction "succeeds" in correcting 'super' to 'super1' and errors are emitted.
For regression testing purposes this would be sufficient though I don't like that we would be inferring that a typo-correction did not trigger indirectly (it is possible, though unlikely, that typo-correction would trigger without resolving to the intended identifier)

Beyond regression testing I'd like to have a way to get notified when typo-correction is silently triggered for general testing.

More information about the cfe-commits mailing list