[Bug c++/11489] [3.4 Regression] ICE/segfault on legal code

["numerical dot simulation at web dot de" <gcc-bugzilla at gcc dot gnu dot org>]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11489



------- Additional Comments From numerical dot simulation at web dot de  2003-07-11 08:58 -------
Subject: Re:  [3.4 Regression] ICE/segfault on legal code

bangerth at dealii dot org wrote:

> I said it before, but I'm going to repeat it:
> > No, I only ask the C++ parser guys to run a decent subset,
> > namely some 50000 C++ lines, which cannot take too long.
> The situation is that gcc is largely a volunteer project and can be
> characterized as severely starved of people who understand the internals
> of the compiler. [...]

> it takes people who are willing
> to do some work that we don't want to put onto our precious developers.

Well, the precious developer for sure creates a compiler with his
code before checkin, right? Or do you want to tell me some guys
check in code that was not compiled before?

So if the precious developer has built a compiler with his
precious changes, how about asking him to run some _subset_ of the
tests, namely those which are likely to be affected by this change.

I mean, if someone changes the c++ parser code it is a good idea
to let the compiler munch some template code on it.
If it compiles e.g. the daixtrose (sf.net) testsuite, which takes about 10
minutes on a P4, he can be sure the parser is rather OK.

So this little burden, namely find some hard, but quickly compiling
mini testsuite for the C++ part may well be put on the back of our
precious ones.

And if you do not have the time for this:
It took me 5 years to build Daixtrose, the hottest C++ compiler testsuite.
Use it.

> PS: The most important piece of C++ is certainly not boost, although I
> appreciate that it is a nice library. The most important piece definitely
> is 1) KDE, and 2) Mozilla.

:-) Your mileage may vary.
Regarding gcc testing: KDE and Mozilla both only use gcc-2.95 features
of C++, which is  IMHO 30% of what the language offers.
So they are very important testcases, but inappropritae for C++ parser
testing.

Also if Qt, KDE and Mozilla used boost and STL they would still run on my
PI 200 Mhz, without constant swapping (but that's another topic).


Markus