On Friday 21 February 2014 23:56:47 chris beck wrote:
> Let me play Devil's advocate. If the number of irregularities is so small I
> can count them on my hand, and all are agreed we can fix them easily, then
> why not just do it and then not have any errata, or ever have to equivocate
> in saying that our history is well-formed and good.
As I have said in multiple occasions before, rewriting history is a highly
disruptive operation and should really be reserved for special situations like
the original SVN -> Git migration.
Rewriting any past Git commit changes the SHA1 hash for that commit and every
commit linked to it, resulting in a cascading effect spanning all history from
that point until the tip of the branch.
Issues (1) and (3) in ESR's post [1], and both of the issues pointed out by
AI0867 [2], are extremely minor and can be considered legitimate part of the
repository history. The "cat $i" file diff actually existed in SVN r14541 [3]
and r14542 [4], which is why it was carried over to the Git repository after
conversion.
1: https://mail.gna.org/public/wesnoth-dev/2013-04/msg00024.html
2: https://mail.gna.org/public/wesnoth-dev/2013-04/msg00025.html
3: http://svn.gna.org/viewcvs/wesnoth?view=revision&revision=14541
4: http://svn.gna.org/viewcvs/wesnoth?view=revision&revision=14542
Committer identities being bogus or incomplete is a thing that has happened in
the past, and *will* continue to happen in the present, for two reasons:
1) People make mistakes, much like the "cat $i" commit above;
2) We don't formally require people to make sure their Git configuration is
sane and their real name and email address are set correctly or are
meaningful (some committers don't even use their real names); and
3) We don't have a Linux-like structure where a very limited set of experts/
module maintainers have push access and review every single incoming patch
(even from established developers) for content and style -- that simply
wouldn't work well for us.
So, I find it preposterous to propose to rewrite history only because people
made such mistakes. I certainly have never been offered the option to rewrite
our history every time I have accidentally botched a commit with typos in the
summary, missing paths that end up becoming part of a separate commit, and so
on.
The most important thing to keep in mind here is that rewriting history
requires everyone afterwards to clone the repository again and export and re-
import any local branches or patches they may have had stashed. I for one
would prefer if I didn't have to download any content larger than 500 MiB over
mobile "broadband" if there is an alternative -- our repository is over 1.7
GiB in size.
Renaming the repository as-is, on the other hand, would only require people to
run a simple `git remote set-url` command at worst, taking up only a handful
of seconds of their time.
***
So, I covered the observed standard VCS mistakes above. The only commit to my
knowledge that constitutes an actual Git conversion bug is from point (2) in
ESR's post, commit 97d9aefdd166ab61260cbffa63ead04651e7c844 [5], which was
r56594 in SVN [6]. The commit message contains extraneous text, but it still
seems a minor non-issue to me. I asked in a previous email whether anyone is
aware of any other history problems in general, and I haven't seen a response
yet.
5: http://is.gd/c8trg2
6: http://svn.gna.org/viewcvs/wesnoth?view=revision&revision=56594
Finally, rewriting history is (as is evident by now) a time-consuming and
error-prone thing to do, especially with such a large repository as Wesnoth's,
comprising over 10 years of history.
--
Regards
Ignacio R. Morelle <shadowm>
_______________________________________________
Wesnoth-dev mailing list
Wesnoth-dev@gna.org
https://mail.gna.org/listinfo/wesnoth-dev
