Re: Will tree-ssa be GCC 3.5?

["Zack Weinberg" <zack at codesourcery dot com>]

Jim Wilson <wilson@specifixinc.com> writes:

> On Sat, 2003-11-29 at 15:58, Zack Weinberg wrote:
>> I don't agree.  Major version number bumps should be reserved for
>> changes which are guaranteed to break compatibility for everyone.
>
> We will never be able to do that, and hence by this criteria, we will
> never be able to change the major version number again.

I admit that the second half of this policy is "... never do that."

I also admit that this is too hyperbolic, as was the original
statement.  I can think of cases that seem like they might happen,
where I would support a change in the major version number.  For
instance, if we merged development of GCC with development of
binutils, such that they weren't separate things anymore.  You'd get
GNU Toolchain version X, and use the cc1, as, and ld from that
package; further, use of a different as/ld would no longer be
supported.  That would be sufficient disruption to the way our users
do things, to justify calling it 4.0.

But tree-ssa isn't that.  tree-ssa is an internal implementation
change.  It happens to be a huge one, but I do not think it will
cause massive disruption to our users.  Of your list of potential
problems:

> 1) Loss of Ada support, until such time as the Ada front end is
> gimplified.  This will be a major problem for all Ada users.
> 2) Replacing the f77 front end with the f95 front end.  This will be
> noticed by many Fortran users.
> 3) Loss of all externally maintained front ends, such as Pascal and
> Mercury, until such time that they are all gimplified.
> 4) The addition of new infrastructure, which will require all gcc
> developers to re-educate themselves.
> 5) The addition of new optimization passes which are likely to cause new
> classes of bugs for users, and also probably new classes of features.

4 is of interest only to developers; that's not a reason to change
the major version number, which is an advertisement to end users.

1,3,5 are adequately addressed by warning people that version 3.5 will
be more buggy and support fewer languages than version 3.4, and people
may want to hold off on using it in production immediately.  I expect
that Ada will be gimplified by the time 3.5 is out, anyway.

2 if people have trouble compiling their old Fortran with g95 then
it's buggy, and that can be addressed in the normal fashion.  The new
feature - support for f90 and f95 - is hardly disruptive; it becomes
possible to use GCC for code that it didn't formerly accept.

> If we maintain gcc3 in parallel with gcc4 for a while, then this
> gives us a chance to add in the tree-ssa stuff now even if it isn't
> perfect yet.  People can continue to use gcc3 if necessary until
> gcc4 stabilizes.

How is that different from continuing to maintain the 3.4.x patch
series until 3.5 stabilizes?  It's just a matter of labeling.  Perhaps
we would want to have slightly less strict rules for fixing bugs on
that branch (any bug, not just regressions?)

zw