[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: patch to speed up slow rcs patch processing.

From: msmith
Subject: Re: patch to speed up slow rcs patch processing.
Date: Fri, 7 Oct 2005 10:17:31 -0400 (EDT)
User-agent: SquirrelMail/1.4.3a-0.f0.9.6.legacy

> Cool!  Does your patch also speed up branch updates?
yes, from looking at the existing algorithm it should improve the
performance of branch updates because branches look like they patch from
head down to the branch point and then up to the branch tag.

> I'm now wondering if an O(1) algorithm is in sight.  Not really O(1) but
> rather O(n) with a comparatively miniscule constant of proportionality.
> By keeping not the actual file contents in the ,v repository file, but
> only a reference to a cached copy.  foo,v might point to foo-1.5,d, for
> example.  And from there, maybe to foo- too, but this is getting
> into wishful thinking territory.
I have considered this issue as well and I think that cvs should not try
to be all things and do this as well.  I do, however, have a pet project
which would provide a caching front-end for cvs.  My idea was a
bit-torrent'esqe implementation on top of cvs.  This would offload check
outs and updates to a set of clients instead of locked on a central server
-- the check out improvements and updates would be incredibly fast.  I'm
not sure what the typical performance of cvs is on large projects, but we
are having to lock access down to 5 concurrent users to prevent the server
from falling over.

> Would you please give a unified or context diff?  diff -u or diff -c
> works nicely; I don't know what the maintainers prefer

I have since fielded the patch locally and determined i missed a free on
the deleted lines removed.  So please note there is a memory leak in the
patch I previously provided.  I believe I have the correction of that item
currently in use (checkout expanded to 1GB at the end with the leak and no
more than 30-40MB after).  When I get back to the office on Monday I will
try and provide a diff -u on the module.  I apologize for not being more
clueful on what format the maintainers are looking for, but my primary
concern was allowing our developers to check out the baseline.  Providing
the patch back to the community was just an effort to show my thanks for
such a great overall produce.

> Do you have a copyright assignment filed?  Apropos, since CVS seems to
> be "nongnu", do contributors have to do all that paperwork?

I'm unsure of what you mean by a copyright assignments file.  Can you


reply via email to

[Prev in Thread] Current Thread [Next in Thread]