On Sat, 2008-09-27 at 17:53 -0700, Paul Charlton wrote:
> 1) ability for client to modify svn:date property is essential for the
> correct import of a repository from a different VCS --- there is not a sane
> person who wants all of the dates to be the import (transaction) date rather
> than the date stamp on the original VCS.
There are sane people who think you shouldn't be importing version
control history into an existing non-empty repository. It basically
means rewriting history; the question "what did I have in this
repository last Tuesday" changes.
> 2) the back-end code assuming its ability to perform a binary search *is
> broken* unless we take the time and effort to add a separate persistent
> index of the svn:date properties.
The specification of {date} makes no sense if revisions are not linearly
date-ordered. This is not an implementation issue.
(When this has come up in the past, brane has said that in an
Apache-style repository, {date} *probably* has a well-defined meaning
for a particular path, even if it doesn't for the repository root. But
that's hand-wavy, and our command-line and API specification does not
resolve {date} with reference to a particular path.)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe_at_subversion.tigris.org
For additional commands, e-mail: dev-help_at_subversion.tigris.org
Received on 2008-09-29 18:10:54 CEST