Re: Auto update ChangeLog for binutils+gdb commits?

On 5/29/20 2:15 PM, Tom Tromey wrote:
> Martin> I'm the author of the scripts that are currently used in the GCC.
> Martin> As mentioned, we update ChangeLog files by a script that takes all
> Martin> the ChangeLog entries from git commit message. The supported ChangeLog
> Martin> format is documented here:
> 
> Martin> https://gcc.gnu.org/codingconventions.html#ChangeLogs
> 
> IIUC, we still have to write the ChangeLog entries by hand, just in the
> commit message, and using a much stricter format.  Is that accurate?

Yes, the entries should be places to git message.

> 
> If so, then this seems less convenient than the status quo.  At least
> for me, editing the files is simpler -- I use scripts to handle the
> merges, rebases, and date-updating; and Emacs provides direct support
> for writing the entries in the correct files and it usually gets the
> function name correct as well.

For generation of ChangeLog template, you can use:
https://gcc.gnu.org/git/?p=gcc.git;a=blob;f=contrib/mklog.py;h=243edbb15c522169709902b27a8558c6e0755107;hb=HEAD

It takes a diff a generates a template. Note that PR entries are auto-filled,
new and removed files are automatically added.

About the merges: it's a pain to make a backport (a.k.a) as you're very likely
conflict in ChangeLog entries. Having the ChangeLog entries in message one
can do simple git cherry-pick and it's done.

Note that quite all developers have self made scripts for these situations
and it seems to me an extra work everybody has to do.

> 
> It's the convenience of the last part that I'm most concerned about
> losing.  If there's a script that does as good a job, then maybe; but
> otherwise it is a -1 from me.  My reasoning here is that editing a
> ChangeLog entry is labor-intensive, and anything increasing the labor is
> a negative.  It's already a reasonably significant drag on submitting
> patches.
> 
> I'd be receptive to removing ChangeLog entirely.  I don't believe they
> provide much value.  However, I know others disagree on this point, so I
> don't plan to press it.
> 
> I'd also be open to changing the format to something requiring zero
> manual intervention.  For example, if we could have a commit script that
> simply listed the files and didn't require editing in the names of the
> functions.

Yes, mklog.py does that.

> 
> thanks,
> Tom
>