Re: testsuite: 143 failed on OS X 10.8.2

[Akim Demaille]

Reposted for the records.

Le 20 déc. 2012 à 10:18, Samuel John <address@hidden> a écrit :

> Hi Akim and thanks for your answer!
> 
> TL;DR. You are right, homebrew was interfering. Using homebrew's `--env=std` 
> switch yields:
> 350 tests were successful.
> 2 tests were skipped.
> 
> I should have checked this before reporting. Sorry.
> 
> 
> 
> 
> On 20.12.2012, at 08:48, Akim Demaille <address@hidden> wrote:
>> It seems that brew adds its own clutter to the process:
> 
> At homebrew we do some nasty stuff to compiler flags (unless --env=std is 
> used for brew), that is true. We provide a wrapper script for `cc` which 
> injects some flags and filters others and then forwards to clang (or 
> llvm-gcc). Similar for other tools. We do this for several reasons. One is 
> that Xcode ships includes and libs and tools in a non-standard location, 
> provides xcrun to call tools and we can control the libraries to link 
> against. So far we have great success in overall measured in the amount of 
> successful builds. We are fine tuning it still, though :-)
> 
> 
>> ./headers.at:274: $CXX $CXXFLAGS $CPPFLAGS -c -o cxx-only.o cxx-only.cc 
>> stderr:
>> brew: superenv removed: -g -O2
>> stdout:
>> ./headers.at:274: $CXX $CXXFLAGS $CPPFLAGS $LDFLAGS c-only.o cxx-only.o -o 
>> c-and-cxx ||
>>         exit 77
>> --- /dev/null        2012-12-19 11:02:32.000000000 +0100
>> +++ 
>> /private/tmp/bison-8C35/bison-2.7/tests/testsuite.dir/at-groups/143/stderr   
>>     2012-12-19 11:02:50.000000000 +0100
>> @@ -0,0 +1 @@
>> +brew: superenv removed: -g -O2
>> stdout:
>> 143. headers.at:125: 143. Several parsers (headers.at:125): FAILED 
>> (headers.at:274)
>> 
>> the failure is due to some interaction between your environment,
>> and our request to use the C++ compiler to link the application.
>> Could you investigate this issue?
> 
> I fail to understand the test code. What is tested exactly?
> Runnint the testsuite generates testsuite.dir/143. 
> There is a quite long input.c and a shorter input.y.
> 
> Can you give me a line to look at?
> 
> 
>> (So, yes, you can trust Bison 2.7).
> 
> I'll proceed with the update to 2.7 (disabling the `make check` step for 
> users on default).
> 
> Is there a way to tell `make check` to skip test 143?
> 
> Thanks,
> Samuel