Re: [PATCH] cvsimport: skip commits that are too recent

From: Martin Langhoff <martin@catalyst.net.nz>
Date: Jan 11, 2007, 20:18 UTC
Message-ID: <45A69B81.40306@catalyst.net.nz>
In-Reply-To: <200701110922.07997.robin.rosenberg.lists@dewire.com>

Robin Rosenberg wrote:

> The idea is nice,  but the downside of this patch is that I (and presumably 
> others) have to rewrite the scripts to invoke cvsps explicitly now.

This patch did _not_ change how we invoke cvsps at all. It did change that we now ignore the very recent commits (and pick them up in the next run), unless you pass -a.

Show 6 quoted lines

> The fix
> should really be in cvsps, not git-cvsimport (which is the reason I haven't 
> fixed this). Running a full cvsps takes two hours and consumes more than a 
> gigabyte of memory for me, which makes it impossible to run on all but one 
> machine, wheras the incremental import runs in less than five minutes on any 
> machine.

Many things would need fixing in cvsps. This aspect [that commits we do not know if recent activty belongs to a finished commit or a commit that is still happening], is not cvsps' fault. It is due to the lack of atomicity in CVS, combined with its rather bad network protocol.

> Add to that the risk that the buggy nature of cvsps probably increases the 
> risk of errors, so please make the old behaviour the default (import all, 
> retain cvsps cache) and make the changed behaviour the result of an explicit 
> switch.

What seems to concern you is the "retain cvsps cache" -- which we do.

I did comment later in the thread that we should consider rebuilding the cvsps cache. The reason for that is that I am seeing LESS breakage than maintaining the cache. Significantly less.

As you say, however, it is a major change, so I'm still evaluating options.

cheers

martin