> On Fri, Jan 15, 2010 at 04:09:01PM -0500, Greg Beaton wrote: > >> What is the point of updating the gcc you use? Don't you just run the >> risk of making incompatible with existing systems. For example, I just >> bought a MacBook Pro last summer. It was shipped with gcc 4.0.1. If >> its good enough for Apple, what is gained by 3rd party software using >> more recent versions. Don't you then have to bundle gcc with GHDL? >>
I think Greg Beaton makes a good talking point. Now I am thinking back to the time I had a very old version of gcc with gnat built in ... I incrementally used it to compile newer and newer versions of gcc until I had more or less caught up to a relatively recent release. I found that the most friendly gcc in terms of not requiring extraneous stuff was GCC 4.1.2 ... when I went to newer releases, it began whining about libraries it seemed to want... I think 4.3 versions of gcc began to whine like that, but since earlier releases did not have them, I know they were not needed to build GHDL. Also, coincidentally, when I unpacked sources for uCLinux, which is a stripped down Linux for embedded systems without an MMU, their standard gcc was ... you got it ... 4.1.2 ... This release seems to be a good one for general usability across many systems, and maybe thats why the embedded community chose it. I don't know if they are still using it now, but they tend to stay a few revisions back of the latest releases of gcc, because those newest ones are sometimes buggy and in flux as well. === I personally think that since GHDL piggybacks onto gcc and also uses GNAT, but that gcc appears to be more and more Linux-centric, making it a policy to always stay in lockstep with the latest gcc may hinder the general portability of GHDL on platforms other than Linux. I know I am in the minority, by doing much of my work on a Solaris 8, UltraSPARC system, but I think portability should be a key aspect of GHDL, even if the majority of GHDL users run Linux on x86. I think that GHDL should perhaps be more conservative and more or less track what the embedded systems developers use as their default gcc instead of trying to stay on the bleeding edge of gcc without any obvious performance gains, and a lot more headache. > > >> I don't know. It seems like every open source product has its own >> version of gcc. Whatever happened to backward compatibility. >> > > Backward compatibility is not really an issue. GHDL depends only on a > very few APIs. > > Of course I can stay with an old version of gcc, but the point is that this > is not possible for a long time. For example very old versions of gcc doesn't > compile anymore on current systems. The later I update the gcc I used the > harder it is. > I think its good for people to try to get GHDL going with later releases of gcc, but unless the backend code generator somehow produces a significant improvement in GHDL's simulation performance, I see little reason to make it a policy to tie GHDL distributions to the bleeding edge. Having said all this, perhaps its a good idea for there to be a GHDL distribution for a well-behaved release like 4.1.2 and also for those that want it ... GHDL compiled with the very latest, though I doubt there will be any performance differences. S. _______________________________________________ Ghdl-discuss mailing list [email protected] https://mail.gna.org/listinfo/ghdl-discuss
