[lldb-dev] Enabling warnings as errors

[Todd Fiala]

My vote is for this:

> I think there should be a WERROR mode in the CMake build, and build bots
should run with it. That way releases etc aren't impacted by new warnings,
but developers *do* notice when they introduce a warning.

In particular, warning with clang.


With the caveat of this:

> You can easily disable warnings that don't have value. We do this all the
time in LLVM.

where there are likely a handful of warnings I'd definitely vote for
turning off (like the full-coverage-enum case with default present).  We
can discuss these as we break warnings that we think are worthwhile (and
fix them) or not (and then disable the warning).

-Todd


On Fri, Aug 15, 2014 at 7:05 AM, Nico Weber <thakis at chromium.org> wrote:

> In the clang/llvm world, I often get a small handful of warnings when
> building locally, almost always due to my local compiler having buggy
> warnings. If -Werror was enabled by default in the clang/llvm build, I'd
> have a broken build fairly often.
>
> Maybe we could detect compiler versions and only enable -Werror for
> compilers that are in some whitelist.
>
>
> On Thu, Aug 14, 2014 at 5:42 PM, Zachary Turner <zturner at google.com>
> wrote:
>
>> There are alot of compilers, but only a small number that we support
>> building with.  And really, the burden would only be on making sure that
>> the compiler you personally build with builds warning free (which,
>> unfortunately, even that doesn't always happen).  If someone checks
>> something in on Mac that introduces warnings on Windows with MSVC, then a
>> Windows person would need to fix it, but I don't think that's too big of a
>> deal (well, I don't mind anyway).  The difference is just that "strive"
>> would become "require".  Sometimes you can't fix warnings for whatever
>> reason, but you could still disable them in that case.
>>
>> I guess in an ideal world I would agree with you, that we should strive
>> to keep warnings down to as low a level as possible.  But in practice I
>> think it's too easy to ignore them, and so people do.
>>
>>
>> On Thu, Aug 14, 2014 at 5:28 PM, Greg Clayton <gclayton at apple.com> wrote:
>>
>>> There are so many compilers and each compiler is good at detecting
>>> things that other compilers don't warn about.
>>>
>>> So I don't think we should enable warnings as errors, but I do think we
>>> should strive to keep warnings down to as low a level as possible.
>>>
>>> Greg
>>>
>>> > On Aug 14, 2014, at 10:38 AM, Zachary Turner <zturner at google.com>
>>> wrote:
>>> >
>>> > What would it take to enable warnings as errors on all platforms?
>>> >
>>> > I've done alot of work getting the windows build warning free, but it
>>> seems like very time I sync the code on linux or Mac, there's many many new
>>> warnings introduced.  I'm always vigilant about fixing warnings in my own
>>> code, it would be nice if everyone else had the same level of vigilance.
>>>  If warnings broke the build, then I think that would be a good motivator
>>> to fix your warnings.
>>> >
>>> > Thoughts?
>>> > _______________________________________________
>>> > lldb-dev mailing list
>>> > lldb-dev at cs.uiuc.edu
>>> > http://lists.cs.uiuc.edu/mailman/listinfo/lldb-dev
>>>
>>>
>>
>> _______________________________________________
>> lldb-dev mailing list
>> lldb-dev at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/lldb-dev
>>
>>
>
> _______________________________________________
> lldb-dev mailing list
> lldb-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/lldb-dev
>
>


-- 
Todd Fiala | Software Engineer | tfiala at google.com | 650-943-3180
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-dev/attachments/20140815/6e591b14/attachment.html>