Hi tree-conflict folks,
I'm looking at the problem of what to say when an update is attempted of a)
a tree conflicted item or of b) an item somewhere within a tree-conflicted
The current stance is that conflicted nodes should be 'Skipped'. This sort
of applies to a). We reach a tree-conflicted node by recursion:
$ svn st
$ svn up dir
Now we've got case b), where an *ancestor* dir is in conflict.
$ svn st
$ svn up dir/sub/node
That's how it is currently specified to look. But, for example, if `dir' was
removed in the repository, that's pretty much failing to let the user know
why the update skipped everything.
Then, there's case c). First cd into the conflicted subtree and then update
something. Suddenly, when I look through the code, there's no place to
"skip" such an update. The failure information is available before any
update stuff even kicked off. When we let the update stuff kick off, it
issues another failure, namely that we're sort of not in a proper working
$ svn st
$ cd dir/sub
$ svn up
svn: Target path '/.../dir/sub' does not exist
(depending on the type of tree-conflict?)
In case a), where we reach a tree-conflicted item by recursion, it's obvious
to do the 'Skipped' thing as usual.
But in case c), where we're *in* a directory that is *completely*
tree-conflicted by an ancestor, it makes sense to me to issue an error
instead of a 'Skipped':
svn: Aborting update: '/.../dir' remains in tree-conflict.
In case b), I'm undecided really. We're currently still *outside* a
tree-conflicted subtree, but some of the *explicit* targets reach into a
tree-conflicted subtree. Intuitively, I want to issue an error on targets
within a conflicted subtree, like c). But we probably want to carry out the
updates on any other, non-conflicted targets issued on the same commandline.
Maybe an "augmented skip" notification would do for both b) and c), saying
"Skipped target 'dir/sub/node' because 'dir' remains in tree-conflict".
A code implication: The place to catch both b) and c) is in an entirely
different code layer than all other skip notifications. It feels like
fabricating a skip and having problems with not running whole chunks of code
although we exit with "success" (haven't tried that yet).
If c) were implemented to bail, that would also bail on b) cases, and it
would be some complex fiddling to allow any other targets to go through.
Do you guys have an opinion on this? That might shed some light.
The place to catch c) would be for example here:
subversion/libsvn_client/update.c:210, in svn_client__update_internal().
or here, where the currently seen error is produced:
subversion/libsvn_repos/reporter.c:1162 in drive().
Attached is a trial patch that catches b) and c) by issuing an error in
svn_client__update_internal(). The patch also re-enables update_tests.py 50
to show this case.
Alternatively, find it here:
The place where tree-conflicted ancestry is currently detected, i.e. cases
a) and b), is deeper within the onion, here:
subversion/libsvn_wc/update_editor.c:1600, by already_in_a_tree_conflict()
Received on 2008-12-12 05:58:10 CET