Re: [eigen] Failing tests in 3.0.2 release

[Benoit Jacob <jacob.benoit.1@xxxxxxxxx>]

2011/9/21 Gael Guennebaud <gael.guennebaud@xxxxxxxxx>:
> Yes that's sad but I've already wasted days on these crappy compilers,
> and when GCC 4.2.1 has a code generation bug that affects us but fixed
> in let's say 4.2.2, then there is nothing we can do to support 4.2.1.
> Same for Mac' Clang. I wasted 3 hours on it yesterday with no sense
> segfaults.

The thought of you losing hours or days on bugs in Mac's versions of
Clang or GCC is absolutely disheartening to me!

Just to confirm, these segfaults are specific to Mac's Clang (the one
that ships in the new XCode) ? I.e. can't be reproduced in plain
Clang?

Benoit

>
> PS: I've a macbook pro now....
>
> gael
>
> On Wed, Sep 21, 2011 at 10:18 AM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx> wrote:
>> 2011/9/20 Gael Guennebaud <gael.guennebaud@xxxxxxxxx>:
>>> Default MacOSX's compilers sucks, _a lot_, I'll mark them as
>>> unsupported and recommend to use macports' gcc. We could even try to
>>> detect them via predefined macros and issue an error or maybe just a
>>> big warning. Eigen is not the only C++ projects to have troubles with
>>> these compiler.
>>
>> It's sad if we have to drop support for Apple's GCC 4.2.1 while we
>> otherwise still support GCC 3.4 but if you say it's a too big
>> maintainance burden, and nobody else steps up to be the Apple GCC
>> savior, then so be it.
>>
>> If you're going to issue a warning or an error, please only make it an
>> error if you know for sure that it's going to fail to compile anyway.
>> Let's not make the problem worse...
>>
>> Apple decided to stay for a very long time with GCC 4.2.1 just out of
>> irrational anti-GPL3 ideology (GCC 4.2.1 was the last pre-GPL3
>> version). The FSF on its side should really have waited for the next
>> major release to do the license change!! When you're stuck between the
>> FSF and Apple, I guess you're in trouble.
>>
>> Disappointing to hear that there also are problems with Apple's Clang.
>>
>> Benoit
>>
>>
>>>
>>>
>>>>
>>>> Less reproducible test errors (usually do not appear within default 10
>>>> repetitions but within e.g. 100 or 1000 repetitions):
>>>> ============
>>>>
>>>> Eigensolver_complex_4:    MacOSX GCC 4.2.1 64bit Release
>>>>
>>>> geo_hyperplane_1:         clang 3.0 64bit Release/Debug, VC10 32bit Debug,
>>>> VC8 64bit Debug
>>>>
>>>> geo_quaternion_1:         Linux GCC 4.4.3 64bit Release/Debug
>>>>
>>>> geo_quaternion_2:         Linux GCC 4.4.3 64bit Release/Debug, Mac GCC 4.2.1
>>>> 64bit Release
>>>>
>>>> jacobisvd_3:              MacOSX GCC 4.2.1 64bit Release
>>>>
>>>> qr_colpivoting_1:         MacOSX GCC 4.2.1 64bit Release
>>>>
>>>> qr_colpivoting_3:         MacOSX clang 3.0 64bit Release
>>>>
>>>> qr_colpivoting_5:         Linux GCC 4.4.3 64bit Release
>>>>
>>>> umeyama_5:                MacOSX GCC 4.2..1 64bit Release, VC9 64bit Debug
>>>
>>> Probably nothing really serious, but it makes sense to look more
>>> closely into them. Feel free to report a bug about them.
>>>
>>> Gael.
>>>
>>>
>>>>
>>>>
>>>> I guess the compile errors for VC8/9 32bit as well as the well reproducible
>>>> test errors with clang and GCC Debug on Mac may be considered as bugs and I
>>>> should post bugs for them? However, I am not sure about the less
>>>> reproducible errors. Should I post bugs for them as well or is it to be
>>>> expected that the tests may fail from time to time?
>>>>
>>>>
>>>>
>>>> Regards, Michael
>>>
>>>
>>>
>>
>>
>>
>
>
>