Community
Participate
Working Groups
Doing a cvs update -dP in the root of a project (e.g. hibernate2 on sf.net) takes a considerable amount inside eclipse compared to "raw" commandline, winCVS, emacs PCVS etc. And this happens both on remote repositories (e.g. sf.net) and local repositories (at work via 100 MBit net) - so it has nothing to do with the net latency. Why is that ? Shouldn't it be possible to do almost as fast as the "native" ones ?
It should be possible in most cases. Given that we cache sync info, we should even be able to be faster. What Eclipse build are you using? What connection method (pserver, ext, extssh)? What JVM? How big is the project involved?
Also, make sure you compare with auto-build turned off in Eclipse since a checkout will result in a full rebuild of your project if auto-build is turned on.
In 2.0 we ran benchmarks and for update, commit, and checkout we were as fast if not faster. We recently found a bug a regression in 2.1 that slowed down CVS operations by 25%, that will be fixed in RC2. I'm going to take ownership and re-run the benchmarks for 2.1 against the command line.
We've fixed the regressions that affected performance. - console output - isIgnored - and some other String optimizations Result, stopwatch benchmarks show that command line versus eclipse are close and our benchmarks test cases confirm. Please re-open if you don't see an improvement in 2.1 RC2.
Just wanting to say that in RC2 CVS operations is much faster then before - so know I can actually use the synchronization command without waiting forever :) Thanks!