advogato diary, homepage at W3C.

/yes, I should send all this stuff to the MailingList... but I still feel I should do some more study before I oblige the mailing list subscribers to consider these issues./

Trying to test hg on an svn project; I tried these commands:

source_repository  svn+ssh://dirk/home/connolly/webtechs-svn/
source_module bottles
source_kind svn
target_kind hg
logfile scm-convert-log
state_file scm-map
print_executed_commands 1
bootstrap 1

but doesn't seem to use hg addremove, so it's not working: files don't seem to get added, and a lot of SVN revisions turn into noops.

Also thinking about hg and distributed version control in the context of RDF diff and paper trail.

The cryptographic history style of git and hg is cool in a lot of ways, but I wonder about this 23 Apr 2003 msg from Tom Lord:

/However, in a distributed system -- whether that system is a distributed SCM or just a bunch of programmers trading trees and patches by hand -- it is simply not realistic to assume that the unbounded amount of history needed for your merge technique will be reliably available./

/You have, in effect, a tree of revisions. The loss of any one node partitions that tree into two trees of revisions and makes your style of merging between those two revision trees, thereafter, impossible./

/Loss of a network site? Disk crash and a bad back-up tape? Node has to be taken (or kept) off-line because some NDA-data slipped into it? End of the world./

It seems that CVS is more fractal in this way: I sometimes start with a local RCS repository and copy the ,v files into a CVS repostitory. I suppose hg handles that case, and better. But what about splitting a module out of a CVS repostory into its own repository, and preserving the history?

And what about checking out just part of a huge project? The entire W3C web site, going back to 1995, is one huge CVS repostory. I know the hg repository format is compact, but even so, I don't want to check out gigabytes worth of history just to edit the minutes of a meeting.

Also... shouldn't hgweb export the actual contents of the tip into HTTP space? Navigating the history is all well and good, but how about the actual state, without the pretty-printing etc. I think SVN does this.


DanConnolly (last edited 2009-05-19 19:31:06 by localhost)