tech-pkg archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: Selecting a C++ compiler



On 10/12/17 18:33, Greg Troxel wrote:
Jonathan Perkin <jperkin%joyent.com@localhost> writes:

* On 2017-10-11 at 01:34 BST, Greg Troxel wrote:

Or to fail hard if c++11 is in USE_LANGUAGES and < 5.1?
Not < 5.1, I've been very happily using 4.9 for our SmartOS builds for
a long time with no obvious issues, and this would screw me over.
I don't have any problem with considering 4.9 good enough, with the
notion that packages with C++11 that don't work with 4.9 will have
GCC_REQD=5.1, and thus fail non-mysteriously.

However, I think I'm of the opinion that we should just fail hard if
an insufficient compiler is detected.  If compiler/gcc.mk has taught
me anything it's that it is vastly over-complicated and a nightmare to
maintain.  Adding another layer of complexity and magic on top is just
going to make things worse.
I think pretty much everyone is coming around to that view and we are
now debating what the defaults should be.  Does anybody disagree?

I'd prefer we simply abort, tell the user to choose a decent compiler
for all C++ packages, then let them sort things out based on those
requirements and any local preferences they may have.
Agreed - that is basically what my recent complicated note intends.

I'll be attempting full bulk builds with gcc5 and gcc6 on CentOS in the shortly and the reports will be available on my web servers.  I think we'll have to move beyond 4.x before too long, but we need to know exactly what we're getting into before committing to anything.  For now, I think the issues with 4.8 or 4.9 are minimal and are better dealt with individually.

FYI, FreeBSD is in the process of switching to GCC 6 as the default for ports that need it (mostly those requiring Fortran or openmp).  There has been fallout that it still being cleaned up.

--
Earth is a beta site.



Home | Main Index | Thread Index | Old Index