[Gnu-arch-users] Re: Re: missing & star-merge

[Stig Brautaset]

On May 05 2004, Milan wrote:
> >>>I don't know in what case --new and --skip-present behave differently in
> >>>replay.
> >>
> >>My guess is that --new only applies the patches that are newer than the
> >>last patch in the current tree. So, if you've cherry-picked patch-17 but
> >>are missing patches 12-16 & 18-23, --new will replay only 18-23, but
> >>--skip-present will replay all missing patches.
> >
> >
> >skip-present and new should (I haven't checked :}) be orthogonal.
> >
> >i.e. if the above example is extend - say patch-20 is a merge from you, 
> >replay will replay 12-16,18-23
> >replay --skip-present will do 12-16,18-19,21-23
> >replay --new will do 18-23
> >replay --new --skip-present will do 18-19,21-23
> 
> Thanks for this explanation :-)
> 
> Going back to star-merge, I am still not sure what exactly it does.

> 
> In the above scenario, what would star-merge do, by applying delta?
> a) delta(12,16) + delta(18,23), or
> b) delta(18,23), or
> c) delta(12,23) - delta(17,18)

ISTR[*] it will take merge in anything after the last patchlog present
from my branch. So I think none of them are correct. Rather, I think
this will be it (in this particular case):

d) apply patches 21-23 to your branch 

I tend to regard star-merge as "a bit magic" and rather tend to use
replay with --skip-present (often followed by a sync-tree). Those
operations are easier to understand and thus also easier to get correct.

When I merge I usually want "gimme all patches from your tree I don't
have already". If cherry-picking takes place, that might not necessarily
be what you get with star-merge (as the above well demonstrates).

[*] the output of `tla star-merge -H' is not trivial to parse, so I rely
    on an random scraps of explanations and experience to reach the
    conclusions here ;)

Stig
-- 
brautaset.org