Re: GCC 4.0 vs. 3.5

[kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)]

    I think that tree-ssa is a big deal, but it's not a feature -- it's an 
    implementation detail.  New and/or better optimizations would be new 
    features.

Perhaps, but I'd argue that the change from GCC 1->2 and 2->3 were
similar internal changes, that only had noticable user effect down the road.

    So, if we're switching to 4.0 it should be because the code we
    generate is far better.  Not that some code is slightly better, or
    that we have the infrastructure we need to generate far better code --
    but that much actual code is actually far better.  Unless we meet that
    hurdle, a rational user's opinion of the next release is "well, that's
    nice, but I see no radical change here".  So, calling it 4.0 would be
    patting ourselves on the back, but miscommunicating with the users.

I don't see that.  The criteria of "far better" code is a fuzzy one.
We're not likely to achieve it in one jump anyway.  It's going to
come in small steps.  It may well be that none of those steps itself
may justify such a jump, but certainly the set of steps taken together
would, but that starts with the tree-ssa merge.

It would seem more peculiar to me to pick some arbitrary point in that
process and declare "this is now GCC 4.0" because the change from 3.x
to that point will be incremental from a performance point of view and
very small from an implementation point of view.

And there is a major user impact: the change of Fortran front end.  That
alone could justify the change of major version number.