Re: [PATCH 4/3] Fold get_merge_bases_clean() into get_merge_bases()

[Rene Scharfe <rene.scharfe@lsrfire.ath.cx>]

Linus Torvalds schrieb:
> 
> On Sun, 2 Jul 2006, Rene Scharfe wrote:
>> Due to popular request ;-), change get_merge_bases() to be able to
>> clean up after itself if needed by adding a cleanup parameter.
> 
> Btw, I think the symmetric thing is still wrong.
> 
> Try this:
> 
> 	git rev-list ^HEAD~1 HEAD...HEAD~2
> 
> which _should_ return just HEAD ("HEAD...HEAD~2" should basically expand 
> into "HEAD HEAD~2 ^HEAD~2")
> 
> I haven't actually tested your patch, but I think it returns HEAD and 
> HEAD~1.
> 
> The reason? You clear UNINTERESTING as part of clear_commit_marks(), so 
> HEAD~1, which was marked thus by the user, gets cleared of its mark as 
> part of the get_merge_bases() invocation.
> 
> I suspect the only way to fix that is to make "get_merge_bases()" not use 
> UNINTERESTING at all, but instead just explicitly use something like

No and yes.  Patch 1 in the 3+1 series changes the flags used in
commit.c to not conflict with the ones in revision.h[*].  So we have two
different UNINTERESTINGs, and get_merge_bases() doesn't mess up the
show/no-show markings.

René


[*] That fix was the one which reportedly made Dscho break a table.
Sorry for that by the way. :-P