[m-rev.] for post-commit review: rbmm disjunct commit fix

[Quan Phan]

Hi,

I resend this email because it seemed not get through the first time.

Quan.

On Sun, Dec 30, 2007 at 03:08:53PM +1100, Zoltan Somogyi wrote:
> For post-commit review by Quan.
> 
> Zoltan.
> 
> Fix a problem in region handling spotted by Quan Phan: when a model_semi
> disjunction commits to a non-last disjunct, we need to clean up the disj frame
> it allocated.
> 
In terms of adding the extra cleanup instruction it looks fine.

We still need to save the regions which need to be protected for the semidet
disjunction so that when cleaning up we can reclaim them. We only need
this saving for semidet disjunctions, not for nondet ones. This saving
is the same as what we do for if-then-else and commit. The generation of
MR_region_fill_disj_protect instruction is not needed any more because
now we use a different mechanism to protect backward live for nondet
disjunctions (i.e., saving the sequence number). Currently, in
mercury_region.h, I just made the instruction empty.

Regards,
Quan.
--------------------------------------------------------------------------
mercury-reviews mailing list
Post messages to:       mercury-reviews at csse.unimelb.edu.au
Administrative Queries: owner-mercury-reviews at csse.unimelb.edu.au
Subscriptions:          mercury-reviews-request at csse.unimelb.edu.au
--------------------------------------------------------------------------