Re: [PATCH] Supporting git-merge-changelog

[Akim Demaille]

Le 25 août 08 à 17:30, Paul Hilfinger a écrit :

> > Why do we still use the ChangeLogs?  git log is much
> > more pleasant to use.  The only situation where it
> > might be useful to have the ChangeLog is when we
> > only have the tarball.  So we could generate ChangeLog
> > from the git log at "make dist" time.  We would just
> > write git logs like we used to write ChangeLogs.
> >
> > ChangeLogs were great because they are local and don't
> > need to reach the repository.  Using git this is no longer
> > a problem.
>
> Not to be contrarian or anything (:->), but I'd be inclined to keep
> using ChangeLogs because they are so much more pleasant to use than
> git logs and their ilk.  I've never encountered the locality argument,
> in fact.  For me, the fact that a ChangeLog contains an easily
> searched change history AND allows the examination of individual
> changes without need for special tools, scripts, etc., makes it a
> pretty obvious technology for me.

I still browse the history, but instead of "less ChangeLog", I
use "git log" and voila! I'm already in less.  There are also
things that are so much easier: "git log -p" gives you the
ChangeLog plus the patches themselves, still in less(1), so you
can easily look for specific patterns in the changes instead of
only in the logs.

Not to mention that if you're interested in the history of
only some of the files "git log data" or "git log data/glr.c"
is more pleasant that "less ChangeLog".  At least I prefer :)

> I'm not sure why others prefer the metadata route provided  by most
> version-control systems. Perhaps it's because it is reasonably suited
> to the rather meager log entries we are inclined to make.

I agree that today I am much less interested in having a
very complete log on "what was changed" because precisely
"git log -p" etc. provide much better information.  But
I still expect much about the "why" of the change.  Often,
IMHO, ChangeLogs contain information that should be comment
in the code, but leave aside the motivations for the changes.
Often critical information is to be found in MLs.
(This is a general observation, not something precisely
about Bison).