"J . A . Magallon" <[EMAIL PROTECTED]> writes:
> On 03.02 Mattias Eriksson wrote:
> >
> > For you who haven't read about the use of gcc 2.96 you can read the this
> > by Linus Torvalds.
> > http://boudicca.tux.org/hypermail/linux-kernel/2000week51/0868.html
> >
>
> Well, don't tell only the begin of the story, tell also the end.
> If people see the date of the post, it is of Dec 14 2000. Many things
> have happend since then.
>
> Everybody agrees that 2.96 was veru broken at his inital release. But now
> it is as good as 2.95.3, if not even better. I think it generates the
> best code in all gcc (pseudo)releases. And the initial problems with the
> optimizer are solved. I have been building kernels with 2.96 since many
[snip]
I was struck by a bug in which dependencies generated with -MM (etc)
were incorrect for cases where the build directory is separate from
the source directory. Basically the dependent was the source file with
the source extension replaced with the object extension. The correct
behavior is to use the basename of the source file as opposed to the
exact path. This incorrect behavior resulted in the object files being
placed in the source directory.
The bug was fixed in the gcc CVS just a week or so ago and on the 20th
I sent a patch for this to Chmouel. It doesn't seem to be applied to
the RPM yet though, but I hope it will be.
Although this is not related to the stability / correctness of the
generated code, it's an example of a bug that only exists in 2.96
(since it didn't exist in any prior versions and is fixed in 3.0).
--
[ Below is a random fortune, which is unrelated to the above message. ]
Life is a grand adventure -- or it is nothing.
-- Helen Keller
