Radical proposal: skip 3.4

[Nathanael Nerode <neroden at twcny dot rr dot com>]

So, stage 3 is taking too long.  But it's not reasonable to release 3.4 
with this number of regressions, as nobody will use it.

So let's just skip 3.4.  Fix as many regressions in 3.3.3 as is 
reasonable and release it.  Meanwhile, start stage 1 of 3.5, merging in 
only apparently regression-free patches.  Hopefully some of the 
unsuitable-for-stage-3 patches will fix the remaining 3.4 regressions as 
an incidental matter.

OK, so this is an absurd proposal, and I don't even really support it 
actually, but it's food for thought.

Now for a more serious analysis.

Currently we have 132 bugs targeted for 3.4.0:
ada: 2
bootstrap: 8
c: 9
c++: 31
debug: 9
driver: 3
fortran: 2
java: 5
libgcj: 5
libobjc: 1
libstdc++: 8
middle-end: 1
optimization: 21
other: 4
pch: 1
preprocessor: 3
target: 19

More disturbingly, 33 of these are wrong-code bugs.  I really don't like 
releasing with wrong-code bugs.

We also have 42 bugs targeted for 3.3.3, of which 19 are still bugs in 
3.4.0, including 6 wrong-code bugs.

This is really pretty bad shape.  It's bad enough it's probably worth 
completely avoiding new development work until this gets cut down 
somewhat.  On the other hand, if the new development work is going to 
incidentally fix these bugs, it's probably worth going with it rather 
than trying to come up with temporary fixes.

So I think people doing development work should identify which 
regressions are already fixed by their development work.  This would 
help identify which regressions are going to be "incidentally" fixed in 
3.5, and which aren't.  If lots of bugs will be incidentally fixed, 
we're in one situation; otherwise we're in quite a different situation.