Re: [eigen] buildtests fail with gcc, icc, suncc and mingw (3.0.4/Debian/amd64)

[ Thread Index | Date Index | More Archives ]

2011/12/18 Nicolas Limare <nicolas.limare@xxxxxxxxxxxxxxxxxx>:
>> > * how do I disable all non-standard features if I only want to use
>> >  C++98 code, compatible in theory with any C++98 compiler (or C++03
>> >  to be exact) ?
>> The only non-standard features we are using are to control the
>> alignment of small objects that you can disable using
>> EIGEN_DONT_ALIGN_STATICALLY, in which case you also lost vectorization
>> for small fixed-size objects. Then we also use some compiler dependent
>> directives to improve inlining. Besides, our code should be C++03
>> compliant. Do you have something particular in mind regarding non
>> standard features?
> I have nothing in mind, I thought the compiler problems I experienced
> were due to some non-standard code written with gcc/msvc in mind.
> I expected sunCC to be a standard-compliant C++03 compiler, and I
> thought that if Eigen could be reduced to this standard, there was no
> reason why Eigen couldn't be compiled in sunCC. But I have not much
> experience with sunCC and I didn't analyze the reasons of the
> Eigen/sunCC clash.

I guess that no compiler is 100% standards compliant, due to various
bugs, especially with the handling of templates. SunCC is a
combination of buggier than average, and less worth making
work-arounds for than others due to very small user base. Old versions
of Eigen did support SunCC but it eventually appeared not to be worth


> --
> Nicolas LIMARE
>                         pgp:0xFA423F4F
> Version: GnuPG v1.4.10 (GNU/Linux)
> iEYEARECAAYFAk7tsqoACgkQvviFAPpCP0+SiwCeJoVQ98ecd8r5nLT9XSycuowM
> ao8AoKAK8sjgW0rl+sHC8MrJFvhkQyrr
> =R1DS

Mail converted by MHonArc 2.6.19+