Nut upsdev - Apr 2012 - reposurgeon and svn:ignore on a newly added directory

Hi Eric,

One corner case I ran across when building from a reposurgeon-generated tree is
that .gitignore files do not seem to be generated when a directory is created at
the same time as its svn:ignore property is set.

In NUT, this manifests itself in r3109:
http://trac.networkupstools.org/projects/nut/changeset/3109

Trac doesn't show it there, but scripts/avahi was created in that commit,
and it had svn:ignore set:

http://trac.networkupstools.org/projects/nut/browser/trunk/scripts/avahi?rev=3109

With 'verbose 4', reposurgeon sees this in the import stream:

...
reposurgeon: node list parsing, line 11292079: 'Node-path:
trunk/scripts/avahi\n'
reposurgeon: node list parsing, line 11292080: 'Node-kind: dir\n'
reposurgeon: node list parsing, line 11292081: 'Node-action: add\n'
reposurgeon: node list parsing, line 11292082: 'Prop-content-length:
44\n'
reposurgeon: node list parsing, line 11292083: 'Content-length: 44\n'
reposurgeon: node list parsing, line 11292084: '\n'
reposurgeon: readprops, line 11292085: 'K 10\n'
reposurgeon: readprops: on trunk/scripts/avahi, setting svn:ignore =
'nut.service\n'
reposurgeon: readprops, line 11292090: 'PROPS-END\n'
...

But the .gitignore is not generated:

...
reposurgeon: Revision 3109:
reposurgeon: <NodeAction: r3109 change file 'trunk/configure.in'>
reposurgeon: <NodeAction: r3109 change file
'trunk/scripts/Makefile.am'>
reposurgeon: <NodeAction: r3109 add dir 'trunk/scripts/avahi'
properties=[('svn:ignore', 'nut.service\n')]>
reposurgeon: <NodeAction: r3109 add file
'trunk/scripts/avahi/nut.service.in' properties=>
reposurgeon: -> find_node(3108, trunk/configure.in)
reposurgeon: find_node(3108, trunk/configure.in) -> <NodeAction: r3022
change file 'trunk/configure.in'>
reposurgeon: -> find_node(3108, trunk/scripts/Makefile.am)
reposurgeon: find_node(3108, trunk/scripts/Makefile.am) -> <NodeAction:
r2836 change file 'trunk/scripts/Makefile.am'>
reposurgeon: r3109: 3 actions
reposurgeon: r3109 gets mark :11867
...

The .gitignore creation only seems to happen on line 2226 of reposurgeon (as of
a2ae20b - master on 2012-04-02), where it checks for a node.action of SD_CHANGE.

There's some logic in the SD_ADD case that I don't completely grok at
this late hour, so I haven't attempted a patch yet. If you would like me to
take a stab at it, let me know how what would be least intrusive. I'm
thinking that the property stuff moves up to where SD_ADD is, in a check for
SD_ADD || SD_CHANGE, then move the svn_nobranch logic into an 'if'
statement one level down from that.

Here's what I'm using:

https://gitorious.org/~clepple/reposurgeon/clepple-nut-conversion/commits/cfl-mods
(mostly debugging changes to the configuration, but some new items in nut.box)

https://gitorious.org/~clepple/reposurgeon/clepple-reposurgeon (tried with both
your latest pushed master branch as of this evening, and my nut-conversion
branch, which was based off of an older commit.)

-- 
Charles Lepple
clepple at gmail

On Apr 12, 2012, at 11:18 PM, Charles Lepple wrote:
> There's some logic in the SD_ADD case that I don't completely grok
at this late hour, so I haven't attempted a patch yet. If you would like me
to take a stab at it, let me know how what would be least intrusive. I'm
thinking that the property stuff moves up to where SD_ADD is, in a check for
SD_ADD || SD_CHANGE, then move the svn_nobranch logic into an 'if'
statement one level down from that.
I started to look into this, and noticed the 'compare' rule in the
Makefile in nut-conversion. I just pushed a commit to the cfl-mods branch to
compare .gitignore with what git-svn creates, and I ran across another corner
case where it looks like .gitignore files are not being deleted after a
directory is deleted.

I'm not entirely sure what's going on with
scripts/python/module/PyNUTClient.py or scripts/udev/nut-usbups.hotplug.in in
the attached diff, though.
>
https://gitorious.org/~clepple/reposurgeon/clepple-nut-conversion/commits/cfl-mods
(mostly debugging changes to the configuration, but some new items in nut.box)
-- 
Charles Lepple
clepple at gmail


-------------- next part --------------
A non-text attachment was scrubbed...
Name: compare.diff
Type: application/octet-stream
Size: 1714 bytes
Desc: not available
URL:
<http://lists.alioth.debian.org/pipermail/nut-upsdev/attachments/20120414/9fd5f7b0/attachment.obj>

On Apr 12, 2012, at 11:18 PM, Charles Lepple wrote:
> One corner case I ran across when building from a reposurgeon-generated
tree is that .gitignore files do not seem to be generated when a directory is
created at the same time as its svn:ignore property is set.

A quick update for the NUT list: I am still tracking this down, plus another
issue where .gitignore files linger after their parent directory has been
deleted. While these are not the only discrepancies between the reposurgeon and
git-svn conversions, they generate a lot of noise in the diffs, and I would like
to fix them so that they don't mask other differences.

Eric: do you have any local commits to reposurgeon that haven't been pushed
to gitorious? Fixing this will probably involve a good bit of code shuffling in
the vicinity of line 2216, and I don't even want to think what a merge would
look like.

Also, I am trying to work on a test case for this, but I am still trying to
figure out a good way to generate the "golden" files for the output
comparison. Suggestions are welcome.

-- 
Charles Lepple
clepple at gmail