IRA for GCC 4.4

[Mark Mitchell <mark at codesourcery dot com>]

Vladimir --

As I mentioned earlier today, theRMs would like IRA to be included in 
GCC 4.4.  We feel this is an important improvement with great potential 
benefit.  However, we also feel that it is very important that, by the 
time 4.4 is released, we have only one allocator.  We do not want to 
risk an incomplete transition.  And, while there may of course be some 
hiccups when including a major new feature, we want to make sure that 
developers working on other aspects of GCC are not impacted more than 
necessary.

Therefore, we suggest the following strategy:

* IRA patch should be reviewed by one or more middle-end maintainers for 
technical design, documentation, and coding style.  I know that you have 
already been discussing some issues with Kenny, Steven, and others.  I 
hope that you will continue to work to address issues raised.  I also 
hope that others will help you to do that.

* Ensure that IRA works for all primary and secondary targets.  You 
certainly do not need to do this personally, and I hope and expect that 
the target maintainers will continue to work with you to validate their 
targets.

* Ensure that IRA performance on SPEC is no worse than old-RA 
performance on primary targets.  Again, I hope and expect that the 
target maintainers will work with you to gather data and investigate any 
issues.

Once these goals have been met:

* Merge the IRA branch to the mainline and make it the default, 
temporarily leaving the old RA in place for the benefit of any 
unconverted targets.

* After 30 days, remove the old RA.  During the 30-day period, please 
work with any target maintainers who are having trouble converting.  We 
will try to remember to warn the maintainers of unconverted targets once 
or twice during the 30-day period.

Is this strategy acceptable to you?

Thanks,

--
Mark Mitchell
CodeSourcery
mark@codesourcery.com
(650) 331-3385 x713