Re: [PATCH] reset cached ident date before creating objects

[Jeff King <peff@peff.net>]

On Fri, Jul 29, 2016 at 11:15:39AM -0700, Junio C Hamano wrote:

> > It does feel a little backwards to cache by default, and then try to
> > catch all the places that want to reset. Another way of thinking about
> > it would be to almost _never_ cache, but let a few callsites like (the
> > commit object creation) explicitly ask for a stable timestamp between
> > the author and committer.
> 
> ... and the reflog?

I left that part out. I can't decide if it is a bug or a feature that
the reflog retains the same timestamp.

I.e., I'm not sure it would be wrong to do:

diff --git a/refs/files-backend.c b/refs/files-backend.c
index 12290d2..465cfc5 100644
--- a/refs/files-backend.c
+++ b/refs/files-backend.c
@@ -2886,6 +2886,8 @@ static int log_ref_write_1(const char *refname, const unsigned char *old_sha1,
 	logfd = open(logfile->buf, oflags);
 	if (logfd < 0)
 		return 0;
+
+	reset_ident_date();
 	result = log_ref_write_fd(logfd, old_sha1, new_sha1,
 				  git_committer_info(0), msg);
 	if (result) {

on top. If somebody writes a lot of refs without updating any commit
objects, should those all have a stable timestamp or not?

On the one hand, moving the timestamp reflects reality.

On the other, I have done dirty things in the past like "undoing" the
results of somebody's mistaken:

  git clone ...
  git push --mirror ;# oops, this deletes everything except master!

by grouping ref updates according to their reflog mtimes.

So I kind of punted by not changing it but also not making any claims
either way. :)

-Peff