[LLVMdev] Enabling stricter warnings for Windows builds

[Reid Kleckner]

Enabling /W4 in MSVC and disabling things as necessary seems reasonable.

You can also self-host with clang-cl, and that will give you the same set
of warnings as clang on other platforms. You can select the platform
toolset in the VS IDE, as described at http://llvm.org/builds/. I think we
are currently slower to compile than MSVC, though. =/

On Fri, Mar 20, 2015 at 4:26 PM, Kaylor, Andrew <andrew.kaylor at intel.com>
wrote:

>  I’ve been guilty several times recently of committing code that
> introduced build warnings.  It’s a poor excuse, but my excuse is that I’ve
> been working on code for Windows and LLVM doesn’t enable strict warnings by
> default on Windows and produces nearly half a million warnings (literally)
> if you manually turn them on.  As such, I thought I’d make an effort to see
> what it would take to clean things up on Windows.
>
>
>
> Looking just at the main llvm project I found that despite the multitude
> of warnings, there were only 16 unique warnings being produced.  I see in
> HandleLLVMOptions.cmake that there are already 17 warnings being explicitly
> disabled, so I thought it would be reasonable to start by adding most of
> the warnings that are currently being produced to that list.
>
>
>
> I triaged the warnings that I was seeing according to how certain I was
> that disabling them was a reasonable thing to do.  Here’s what I came up
> with (with a glaring bias toward just disabling common warnings and fixing
> uncommon ones):
>
>
>
>
>
> Warnings that should almost certainly be disabled
>
> -------------------------------------------------------------
>
> warning C4100: unreferenced formal parameter (263666 times)
>
> warning C4127: conditional expression is constant (101120 times)
>
> warning C4512: assignment operator could not be generated (40472 times)
>
> warning C4505: unreferenced local function has been removed (6606 times)
>
> warning C4610: [...] can never be instantiated (1714 times)
>
> warning C4510: default constructor could not be generated (1714 times)
>
> warning C4702: unreachable code (1343 times)
>
> warning C4706: assignment within conditional expression (296 times)
>
>
>
>
>
> Warnings that seem like they should be fixed but will take a little time
> to correct
>
>
> --------------------------------------------------------------------------------------------------
>
> warning C4245: signed/unsigned mismatch (962 times)
>
> warning C4310: cast truncates constant value (216 times)
>
> warning C4701: potentially uninitialized local variable (123 times)
>
> warning C4703: potentially uninitialized local pointer variable (40 times)
>
> warning C4389: signed/unsigned mismatch (28 times)
>
>
>
>
>
> Warnings that should probably be fixed
>
> -------------------------------------------------
>
> warning C4189: local variable is initialized but not referenced (6 times)
>
> warning C4204: nonstandard extension used : non-constant aggregate
> initializer (4 times)
>
> warning C4611: interaction between '_setjmp' and C++ object destruction is
> non-portable (2 times)
>
>
>
>
>
> As a caveat, I didn’t look at the code for any of these warnings -- not
> one -- so I don’t know anything about how benign any of them might be.  I
> also don’t know why there are two different “signed/unsigned mismatch”
> warnings (though I’m guessing one is assignments and the other is
> comparisons).
>
>
>
> I have a local build going right now with the first two groups of warnings
> disabled.  If it goes as expected, I’d like to commit this change along
> with fixes for the very small last group.  Then, after a reasonable
> adjustment period to let subprojects see how this looks, I’d like to turn
> on the “all warnings” option by default for Windows builds.
>
>
>
> Does anyone see any problems with the way I have categorized these
> warnings?
>
>
>
> Thanks,
>
> Andy
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20150320/eb3987cf/attachment.html>