On Sun, Aug 1, 2010 at 7:40 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> I tried to duplicate your results using what I believe to be the latest
> version of clang,
I'm using SVN tip of llvm+clang from ~one week ago.
>> (2) clang doesn't support (or require) "-no-cpp-precomp", which
>> src/template/darwin adds to $CC unconditionally. Adding the flag
>> unconditionally seems wrong regardless: e.g., -no-cpp-precomp isn't
>> supported by FSF GCC on OSX either. clang is happy to ignore the flag,
>> but it just emits countless "warning: argument unused during
>> compilation: '-no-cpp-precomp'"
>
> I do see that, but I also see it complaining about -fwrapv:
>
[...]
>
> We're certainly not going to just drop -fwrapv, as that would break the
> code on many modern versions of gcc. (I'm a bit surprised and concerned
> that clang hasn't got this flag, btw.)
Support for -fwrapv was apparently implemented recently:
https://llvm.org/viewvc/llvm-project?view=rev&sortby=date&revision=106956
FWIW, I think we should aim to eventually remove the dependency on
-fwrapv, and instead make the code correct under the semantics
guaranteed by the C spec. That will be hard to do without a tool that
checks for code that makes incorrect assumptions about integer
overflow behavior, but there seems to be progress in that area
recently:
http://blog.regehr.org/archives/226
(As it happens, their checker uses llvm, which is what motivated me to
start poking around in the first place...)
>> (3) There are countless warnings emitted during the compilation of
>> regcomp.c and related files, due to unused values returned by ERR(),
>> VERR(), FAILW(), and similar macros.
>
> I fixed this in HEAD, or at least my copy of clang doesn't complain
> anymore.
Yep, looks good here. Thanks!
Neil