r208345 - Remove -Wnon-modular-include

[Argyrios Kyrtzidis]

On May 9, 2014, at 12:45 AM, Alp Toker <alp at nuanti.com> wrote:

> 
> On 09/05/2014 08:32, Argyrios Kyrtzidis wrote:
>> On May 9, 2014, at 12:05 AM, Alp Toker <alp at nuanti.com> wrote:
>> 
>>> On 09/05/2014 07:34, Argyrios Kyrtzidis wrote:
>>>> Hmm, there is a difference between a warning that is designed to find problems but ends up too noisy in practice and a "remark"
>>> ... and a remark is also a noisy warning designed to find problems.
>>> 
>>>> , we should not conflate the two concepts.
>>> The two are the same thing.
>> Let me give you a specific example:
>> We recently introduced -Wmodule-build to "make it easy to see when modules are (re)built". Do you regard this as designed to find problems ?
>> 
>> Now consider that people have the workflow of having -Weverything, so that when clang introduces a new warning they can see it, be aware of it, and see if it finds new problems in their code.
>> What is the value for these people to get a brand new compiler, compile their code and see "warning: building module 'Foundation' as '/path/to/Foundation.pcm'" ?
> 
> I don't see how that message is more or less valuable than any other low-value diagnostics triggered by -Weverything. What makes it a special case?

It's not intended for users.

> There are lots of highly specific DefaultIgnore warnings, plenty that are less interesting.

Could you give specific examples for warnings that are less interesting than -Wmodule-build ?

Any warning that is less interesting than something that is information for clang development is a candidate to become a "remark" or removed entirely.

> 
> Note that with -Reverything, these low value diagnostics would all get emitted as "remark" messages, not warnings, thus providing a version of -Weverything that's actually useful.
> 
> With those semantics you could do a build, serialize diagnostics for each TU and filter them after the fact without losing information. An IDE could request "everything" without losing the correct warning and error mappings. And the pressure to not add sundry warnings because they add noise to -Weverything would be avoided. What's not to like?

We lose the distinction between warnings intended for users and internal development notifications.

> 
> Alp.
> 
> 
>> 
>>> (Whether they get treated as remarks, warnings or errors should be a question of mapping only.)
>>> 
>>> Alp.
>>> 
>>> -- 
>>> http://www.nuanti.com
>>> the browser experts
>>> 
> 
> -- 
> http://www.nuanti.com
> the browser experts
>