Re: bootstrap problems

[Andreas Jaeger <aj at suse dot de>]

Jan Hubicka <jh@suse.cz> writes:

> Hi,
> Just to introduce - my recent changes to subreg handling has added some
> extra sanity checking code, that found some latent bugs around the code
> (the paradoxical subreg combining problems for instance). Currently the
> PPC/i386 bootstrap passes fluently, but mips bootstrap still rings.
> 
> This is definitly problem for people doing development on the mips machine
> and is really unfortunate.  Jeffrey has suggested to revert the patch,
> but I believe this is not best solution for few purposes:
> 1) the patch is not buggy by itself in some point of view - just traps latent
>    bug and as it is IMO good.
> 2) similar situation has happened about the conditional code revertion code.
>    The patch has been reverted, the latent problem remains to my best knowledge
>    (see PR 3053)
> 
> So I am thinking about alternate way around this.  One way is to branch between
> stable and development tree, but we already do have such branch.
> Other way that goes to my mind is to add tags "target-bootstrap" and
> use scripts to move them each time the bootstrap on given target suceeds
> for some write-access maitainer.

I don't think this is really feasible.

>
> Then people not interested in having bootstrap problem all the time will just
> use the tag. Maitainers can always try to update to latest and see if there
> is problem or not.
>
> What do you think about that?

Since your code is correct what about either of the following:
- #if 0 the code and add a comment documenting this.
- enable the checks only with one of the --enable-checking flags (we
  could even invent a new flag and name it breaksBootstrapButIsCorrect
  ;-)

Either way gives people the chance to fix the problem while other can
continue to bootstrap.

You could also create temporarily a branch with your patch in it, fix
all platforms (I'm sure others will help with the testing) and then
merge again.  The branch should be just for this task and short-lived.

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj