Re: moving to git

[Ralf Wildenhues]

Hi Gary,

* Gary V. Vaughan wrote on Sun, Apr 13, 2008 at 02:09:58PM CEST:
>
> I see some anomalies in the commit log, though I don't know whether they
> are correctly ported CVS anomalies, or indicative of actual errors in  
> the conversion.  These lists are not exhaustive, just selected
> examples of the classes of problems I see.

Thanks for bringing them up!

> 1. Thomas Tanner has many many commits with no message.  If we ever
>    decide to move to generating a ChangeLog from the commit log, we'll
>    lose a lot of information:

Hmm.  Well, I don't think we should ever try to generate ChangeLog for
past commits.  Besides practical hassles, this can also cause legal
ones.  We may however want to do this for new commits at some point in
the future.  While my opinion about is that I would rather get people to
use a script like ./commit to ensure ChangeLog and commit message are
synchronized, this part of the discussion isn't important ATM: it's
independently useful to have good commit messages for old log entries,
we can discuss the rest later.

I can try to adjust my filter-branch script to rewrite those commits
that have an empty CVS log message, to use the ChangeLog diff.
That should help, so I'll try this.

> commit a4f2b1f84691970ff2ff3917dbade735f897c4a8
> Author: Thomas Tanner <address@hidden>
> Date:   Mon Nov 23 21:26:38 1998 +0000
>
>     *** empty log message ***

> 2. Alexandre Oliva seems to have many commits that cover several patches
>    and/or commit authors:

I don't know how to fix this in any automatized way.  It looks like he
just committed them to CVS all at once.  My take is that we have to live
with it, there is no information lost in the CVS -> git conversion.

> 3. There seem to be several commit logs by other authors that suffer the
>    same problem (though not as drastically as Alexandre Oliva):

Same reason, I looked up the examples you gave in the RCS ,v files.

> 4. For a time we put the committer of a change in the ChangeLog entry
>    and credited the author on a separate line, especially if they
>    submitted a patch with no ChangeLog entry.  Maybe we'd like to
>    fix that in the commit history?  (We stopped doing it when RMS
>    told us it was legally wrong)

Oops, I didn't know that.  Hmm, there is several commits matching
  grep '^       From .*@' ChangeLog*

(note the TAB), but there are several more matching
  grep 'From ' ChangeLog

and here the credits are not necessarily on a separate line, often do
not have an email address, may span more than one line, may be at the
beginning or at the end of a ChangeLog entry.  Also, there are some that
cover only part of one commit, for example this one:

> commit 1d8f975749e5c9b75becca2f8a5f365dba834f10
> Author: Gary V. Vaughan <address@hidden>
> Date:   Sat Dec 23 16:30:43 2000 +0000
>
>     From address@hidden:
>     * doc/fdl.texi (GNU Free Documentation License): contained @bye
>     command which prevented part of document to be generated (indices,
>     etc).

>     * doc/libtool.texi (Dlpreopening): the @deftypevar did not contain
>     a space after a type.

So effectively I have two problems with this: I don't see an easy way to
generate a robust matching rule, and I'm not sure a rewrite would only
improve some and not regress others, unless I go and check each and
every such revision manually.

Unless you press hard, I'm inclined not to work on this one.

> 5. There are some commit logs that imply actual code was commited
>    by known lurkers:

;-)

Cheers,
Ralf