From: Martin Langhoff <martin.langhoff@gmail.com>
Date: Mon, 08 Jan 2007 07:17:25 GMT
Subject: Re: [PATCH] cvsimport: skip commits that are too recent
Message-ID: <46a038f90701072317h9bede00o939d4c078ccd569c@mail.gmail.com>
In-Reply-To: <11682386193246-git-send-email-martin@catalyst.net.nz>

On 1/8/07, Martin Langhoff <martin@catalyst.net.nz> wrote:
> With this patch, cvsimport will skip commits made
> in the last 10 minutes. The recent-ness test is of
> 5 minutes + cvsps fuzz window (5 minutes default).

Here is the repost with appropriate doco and an override ;-)

In related news, I am trying to debug an import that consistently
skips over remote commits... which is bad, bad news. The culprit seems
to be cvsps -- it skips commits it clearly knows about, and I'm not
sure why. I do think those were commits that cvsps saw half-baked in
the first place.

Passing -x to cvsps does bring those commits back, cvsps with -x can
afford to rewrite history a little bit. As long as the history being
rewritten is not too old we are safe. So with this patch, passing -x
is safer, assuming that 10 minutes is enough of a time window for
cvsps to change opinion about the project history.

 (Before you ask: from a data correctness, this is a fine mess.)

For this repo, I'll start running cvsimport with -o ' -x ' and see how
it behaves. Time-wise, the bandwidth usage and cpu times are roughly
similar for me using --cvs-direct. The patch to do it by default in
cvsimport is trivial, but I'm not entirely happy with the concept just
now.

In any case -- this should be a bit of a warning. cvsps is not
particularly reliable (not that cvs data ever is!), and passing -o '
-x' may help.

cheers,


martin