Re: 3.4 regressions: are 2.95 regressions still actual

[Steven Bosscher <s dot bosscher at student dot tudelft dot nl>]

On Monday 12 January 2004 06:21, Mark Mitchell wrote:
> On Sat, 2004-01-10 at 16:23, Giovanni Bajo wrote:
> > Mark,
> >
> > There have been some discussion about regressions from 2.95. Are they
> > still actual? That is, do we consider regressions bugs that were not
> > present in 2.95, but have been there for every 3.x release?
>
> Yes, such a thing is clearly a regression.
>
> A bug does not stop being a regression simply because we don't fix it
> for a long time.

There is no discussion about that, of course it is a regression.

But this came up because one such regression (an accepts-illegal bug)
was found about a week ago, and then targeted for 3.4.  Many people
were trying to reclassify and retarget bugs.  Giovanni had set the
target milestone for this bug to 3.4.0, and I moved it to 3.5 because
we want Bugzilla to reflect the quality of the compiler, and a bug
that requires four years to be discovered in a pity but not a reason
to delay branching 3.4.0.  That doesn't suddenly not make it a
regression -- just one that is not a show stopper for 3.4.0.

Gaby sent a good mail about this , this morning, and I agree with him.
However, there is a deeper problem here:  Many really minor bugs get
targeted to the first upcoming release so we appear to have many bugs.
Then you look at the number of bugs and say you will not branch.

All of this is just a result of our bug masters identifying things as
regressions much sooner, and key-wording/targeting everything.  Another
example is the 33 "wrong-code" bugs: Many of them were not wrong-code
at all, and after reviewing them all (and fixing some), we only have
16 left now (including the -fwritable-string and global register ones;
are we going to deprecate global registers for 3.4? It never worked
for any 3.x release anyway).

If you think that every regression should be targeted for 3.4.0, no
matter how relevant or serious it is, that is fine with me.  But then
you should announce this policy (perhaps even document it) not just
look at the number of bugs but also at the severity of the bugs.  Then
you'll see that 3.4.0 is actually in pretty good shape when we stamp
out the 20 worst bugs (many have patches pending or are being worked
on).

Gr.
Steven


P.S. It would help if people would review the patches mentioned in
http://gcc.gnu.org/ml/gcc-patches/2004-01/msg00891.html.