Re: correctness and comprehensiveness of compilers [was: Re: C compiletime]

[Scott Robert Ladd <coyote at coyotegulch dot com>]

Robert Myers wrote:
> Scott Robert Ladd wrote:
> > That said, I've not found code that won't compile correctly with icc, 
> > but that may very well be the fact that I use it almost exclusively 
> > for code of my own invention, or standardized benchmarks that tend not 
> > to stress language features.
> The problem with icc is that (at least as far as I can tell) it gets out 
> of sync with the header files and library routines it needs to cooperate 
> with, and I have had to update the compiler just to get vanilla code 
> (not my own) to compile.  Maybe (probably?) not much of a problem if 
> you're paying for annual support and you keep your compiler up-to-date, 
> but it is a problem.

I'm excessively anal about keeping compilers up-to-date. One of my 
commercial clients provided the funding for ICC support.

In all fairness, GCC is not immune to such problems, as evidenced by the 
bug database. A primary advantage of GCC is that we can browse, extend, 
and fix those bugs, whereas with a commercial product options are 
somewhat limited.

> Relying on a proprietary compiler as the lynch-pin of (say) a 
> high-performance computing project disables much of what is appealing 
> about open source as a model for broad cooperation, because the cost of 
> the compiler and annual support becomes a barrier to entry.

For two of my paying projects, I need OpenMP support, which does not, as 
yet, exist in GCC. Thus I have little choice on Linux, and precious 
little time to devote to extending GCC via the GOMP project. Alas, none 
of my clients is willing to fund my time for working on GOMP, given the 
existence of Intel's compiler.

--
Scott Robert Ladd
Coyote Gulch Productions (http://www.coyotegulch.com)