Posted Mar 26, 2011 22:59 UTC (Sat) by jd (guest, #26381)
Parent article: GCC 4.6.0 released
I'd like to see GCC live up to the modular concept a bit better by including more modules as standard rather than allowing them to decay on the sidelines or having GCC developers re-invent wheels that already work damn-well in other projects.
The Fortran 2003 and 2008 standards should be fully implemented. Right or wrong, I'm going to blame the delays in politics and bickering, since I know damn well gfortran was limited to Fortran90 specs long after G95 became the preferred front-end. NASA's chief complaint was of gfortran core-dumping when compiling their CFD codes and little evidence anyone was doing much about it. Would it not have been quicker to port anything Fortran90 missing from G95 into G95, replace gfortran with that, and eliminate the need for time-consuming bugfixes and implementation of an already-implemented standard?
Then there's a D front-end - D is proving to be a better choice than C++ in some situations and I'd like to see it become a part of the mainstream compiler collection. Ok, the maintenance of it isn't so hot right now, and I sympathise with those who don't want stale code in GCC, but in the end staleness is a problem of attitide. "Not Invented Here" is not a phrase that is acceptable in Open Source.
There's also a Unified Parallel C front-end, which has a lot of potential. It's a nice supercomputer language and GCC is popular in that market. At SC|05, most compiler vendors considered the GCC benchmark as the only one that really mattered. That should tell you something.
The most absurd example, in my mind, is the Pascal front-end for GCC that is already provided by many distributions -- but not by the GCC maintainers. Didn't we go through that with ADA, and wasn't it blatantly obvious to everyone that the longer it took to get integrated, the more teething problems there would be?
The great thing about GCC is that you needn't compile components you don't use. Pretty much the same as X, in fact, but with a better configuration mechanism than X11R4 had. (That was NASTY!) Ok, so the source would be a bit bigger. But I wonder by how much. There's got to be a lot of duplication in the isolated projects, so integration would mean a more comprehensive common library - something that can't happen when you have isolated projects.
It wouldn't impact binary distros at all, since you only download the packages you use. (Right?) All it means for those users is more choice.
What it would do is increase the number of ways you can hammer the rest of GCC, which means you get the opportunity to do better-quality testing.