Re: GCC Status (2004-08-09)

[David Edelsohn <dje at watson dot ibm dot com>]

>>>>> Steven Bosscher writes:

Steven> The question I think we need to answer first is, is mainline ready
Steven> to go in bug fix mode. IMVHO it's not.

Steven> As Kenner pointed out, there are still a lot of pending patches for new
Steven> passes that we really do need.  Also there are some existing passes that
Steven> may need some major surgery.
Steven> Then, we have have a very serious compile time problem, look at the SPEC
Steven> tester from Diego, for example.  So far tree-ssa has only added passes
Steven> and has not made it possible to obsolete existing compile time hogs in
Steven> the RTL optimizers.  Patches to improve this (like Dan Berlin's aliasing
Steven> work) are not ready yet.

Steven> And last time I tried, mainline can't even build a linux kernel with
Steven> 4k stacks (because BIND_EXPRs are gone, the whole stack slot allocation
Steven> mechanism was broken, and recently was ripped out).

Steven> With so many known issues and pending works, why rush going to stage3?

	This seems more like a semantic argument over what "Stage 3"
limitations mean.  If we decide that compile speed and stack usage and
performance are important enough "regressions" to block the release, we
can decide to perform those types of surgery in Stage 3.  Moving to Stage
3 is not equivalent to a runaway train leaving the station.

	Maybe Mark can provide a little more guidance about what he wants
to achieve and hopes to achieve for this release.

	IMHO, what we fix in Stage 3 depends on what we decide are
important bugs -- in the broadest sense.  If those bugs require all of LNO
and the aliasing improvements, then those will be candidates for Stage 3.

David