I see.
(1) Are there any specific cases that this scan solves? (I know there's one
problematic case that I encountered in the LyX repo, where a whole bunch of
branches appear under branches/personal/ rather than directly under branches/,
and then they were all getting imported into mercurial as a single branch, with
a wrong file layout; and this was wreaking havoc with the repository's size.
However, I'm not sure that the scan would solves this, since by the time it
happens (after having imported the changesets) it's too late to decide what is a
branch and what is not?).
(2) Perhaps the information gleaned from such a scan could be cached, so that
incremental imports would not take as long? The initial convert takes a long
time on a large repository as it is, so the additional overhead of the scan, is,
I think, acceptable. But taking so long for incremental imports is not.
(3) Or perhaps there could be some other solution?
But the current situation means that incrementally tracking LyX's subversion
repository is basically not workable, at least not automatically.
Thanks for your feedback!
|