I now have a rather complete conversion of the Pike repository to Subversion (some more copyfrom links could be added, but otherwise everything should be ok). It includes all checkins up to and including 2006-08-29, which when I'm writing this means "everything". The dump file is in pelix:/pike/home/marcus/pike_svn.dump.gz. My suggestion is that we create an accessible repository from this somewhere, to be able to work on tooling, and to check for glitches in the conversion. Once that is ready, I'll make a new conversion for the actual switch. Which means that everything checked into the temporary repository would be discarded.
I have not converted projects/ or Pike/ulpc or Pike/extra_tests, as these are not branches of the main Pike development tree. Any opinions on whether to convert some or all of these, and in that case should they be in the same repository or a different one?
Some statistics:
Conversion: ----------- Number of revisions: 31738 Conversion script run rime: 15h 15mins (on pelix) Dump size (uncompressed): 1.7Gb Dump size (compressed): 384Mb
Test import: ------------ Dump import time: 4h 24mins (on bhelliom) Reposiory size: 421Mb (with fsfs)
I have not converted projects/ or Pike/ulpc or Pike/extra_tests, as these are not branches of the main Pike development tree. Any opinions on whether to convert some or all of these, and in that case should they be in the same repository or a different one?
I'm in favour of different (as in "from the main pike repository"; no opinions expressed or implied about whether they, in turn, be joined) repositories for those.
Well, that one contains a total of 7 commits, and no tags or branches, so I think it can be arranged. :-)
Ok, a new version of the dump has been placed in the same location, now including extra_tests, as well as ulpc. projects/ is still excluded.
Updated statistics: ------------------- Number of revisions: 31896 Conversion script run rime: 17h 50mins Dump size (uncompressed): 1.7Gb Dump size (compressed): 391Mb
Marcus dump is now available for read-only inspection at https://svn.lysator.liu.se/repos/pike/ (WebDAV). All 31896 revisions of it. Marcus has done a monster job merging all repositories together with an impressive restoration of history.
This is not the final export, especially since the dump is a bit old by now, but you can inspect it and prod it with your favourite tools.
A conversion of the CVS as of this friday (32061 revisions) is now available on the same URL.
Tried it out a little and from the small bits I've so far it looks good. svn annotate is unfortunately a bit slow (tried it on src/program.c, takes 35 sec vs 7 sec with cvs), but I guess it can be lived with.
And why do I get this error?
kat:~/Pike/svn-test-7.7/src> svn log -r26702 program.c svn: REPORT request failed on '/repos/pike/!svn/bc/26702/Pike/7.7/src/program.c' svn: '/repos/pike/!svn/bc/26702/Pike/7.7/src/program.c' path not found
That revision has a change in that file if I look at a larger range (26000:27000), but even if no change existed I expected it to just not print anything.
On Sun, Nov 05, 2006 at 04:05:03PM +0000, Martin Stjernholm, Roxen IS @ Pike developers forum wrote:
svn annotate is unfortunately a bit slow (tried it on src/program.c, takes 35 sec vs 7 sec with cvs)
that is 5 times as much. checkouts will be slow (i saw someone complaining about it not being done after 24 hours (but maybe he was checking out all versions)
how about updates? if they are slow too that would be much more of a problem...
greetings, martin.
Yes, five times slower is annoying, especially since I do the annotate/diff dance fairly often to track down code history.
Updates are normally a lot faster since subversion caches more data locally than cvs. Diffs between old revisions take about the same time.
now,i wonder what is the cause for the slowdown. i don't think it could be a general svn problem. people would have noticed that by now.
what kind of machine is svn running on? maybe the slowdown is just do to weaker hardware, and would go away once moved to pelix?
greetings, martin.
svn annotate is unfortunately a bit slow (tried it on src/program.c, takes 35 sec vs 7 sec with cvs)
that is 5 times as much. checkouts will be slow (i saw someone complaining about it not being done after 24 hours (but maybe he was checking out all versions)
Annotate behaviour will probably not affect checkout behaviour, but we will probably notice many operations taking vastly different order of magnitude time to complete. I expect many (particularly tag/branch related) operations will speed up significantly instead.
I would not expect checkout to be one of the operations that take much more time than with cvs (based mostly on the subversion team's goals).
On Sun, Nov 05, 2006 at 09:20:01PM +0000, Johan Sundstr�m (Achtung Liebe!) @ Pike (-) developers forum wrote:
I would not expect checkout to be one of the operations that take much more time than with cvs (based mostly on the subversion team's goals).
exactly. the problem must be elsewhere.
greetings, martin.
And why do I get this error?
kat:~/Pike/svn-test-7.7/src> svn log -r26702 program.c svn: REPORT request failed on '/repos/pike/!svn/bc/26702/Pike/7.7/src/program.c' svn: '/repos/pike/!svn/bc/26702/Pike/7.7/src/program.c' path not found
The reason is that the name "program.c" is expanded to the path "Pike/7.7/src/program.c", and in version 26702 that path does not exist. The path that you _want_ to log in that version is "Pike/7.5/src/program.c". If you have svn 1.4, you can use a "peg revision" to say "the path which I want to log in -r26702 is the one which corresponds to the one which is called Pike/7.7/src/program.c (abbreviated as program.c) _now_".
See http://subversion.tigris.org/issues/show_bug.cgi?id=2460
pike-devel@lists.lysator.liu.se