This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: [Proposed binutils PATCH] Re: Diagnosing an intricate C++ problem

[ binutils suppressed of the recipients ] said:
> Jeff Law tries to do a little bit of work on that branch, but the
> other GCC committers need to help him out to make any real branch
> maintenance take place.  C++ in 2.95.2 is so bad that I know many C++
> that have had to move the Russian Roulette game of 2.96 to get a
> [sometimes] working compiler. 

Well, as a C++ user I find that the comment on 2.95.2 is a bit harsh.
Certainly, it has some bugs and problems but for most programmers no 
real show-stoppers at least for my architectures and my code. OK,
I'm not doing a lot of exception, no threads (yet), and my use of the library is 
limited to the stl and to very basic io. On the other hand, I make an 
heavy use of templates (and sometimes I exercise some dark corners of 
the standard...). Of course, there were some very annoying things but 
not much that I was not able to deal with some ifdef or MACROS (in 
the worst case, yucks)

Then, if you have followed this list, you should certainly have 
noticed that a lot of stuff in 2.96 would/will not have made its way to
2.95.3 because it breaks compatibility or because it is a major 
change (eg function-at-a-time). So I'm not sure that a 2.95.3
would have helped much and I'm not sure to see what could have been 
done better with this respect.

Now, in addition to using 2.95.2, I'm also playing the "Russian Roulette game of 2.96"
for more than one year now. I would not call it a Russian Roulette 
game at all. Overall, I'm sure that the straight from cvs 2.96 compiler was up and 
working for me between 60% or 80% of the time during the last year 
and a lot of effort have been done to improve the situation lately.
I do this mostly to be sure to have little trouble with 3.0 (well I 
have not tested the new-abi yet...) and to report bugs ASAP so 
that what bothers me has a chance to be corrected (unfortunately, I'm 
not able to correct g++ bugs that often). said:
>  But we have not seen a release of FSF/GNU GCC in a year, and 3.0 is
> another ?6-9mo? off. 

Well, not yet a full year (2.95.2 was released october, 24, 1999) :-)
Certainly, the rythm of releases has decreased, maybe 2.95 --> 3.0 is too
big a step and a 2.97 would have been necessary, I do not know. I'm sure
it has been considered and discussed. From what I have seen, it looks like
gcc people have tried to minimize the number of compatibility breaks that
have caused some pain with previous releases, which is after all not so bad.

[ By the way, would a small explanation on the "longer time" between 
  2.95.2 and 3.0 than before (there was a 6 months rule during EGCS time IIRC)
  be a good thing ? ]
OTOH, go and look elsewhere. g++ is really doing well...
Some vendors have very bad C++ compilers and/or release them quite 
rarely, so criticizing g++ and gcc with this respect is IMHO inappropriate.
Theodore Papadopoulo
Email: Tel: (33) 04 92 38 76 01

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]