| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| | |
|
| | |
|
| |
|
|
| |
Missed these earlier.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The actual fix was
@@ -339,7 +355,10 @@
fset = funcs.get("fset")
name = funcs.get("name", "<unknown>")
def _fget(self, new_value=None, from_fset=False): # only used if mutable == True
- value = fget(self)
+ if from_fset == True:
+ value = new_value # compare new value with cached
+ else:
+ value = fget(self) # compare current value with cached
if _cmp_cached_mutable_property(self, "change hook property", name, value) != 0:
# there has been a change, cache new value
old_value = _get_cached_mutable_property(self, "change hook property", name)
The reason for the double-save was:
>>> print t.settings["List-type"]==EMPTY
True
(the cached value here is EMPTY)
>>> t.list_type = []
(old fget compares cached EMPTY to current EMPTY, no change, so no
cache. fset notices change and saves EMPTY->[])
>>> t.list_type.append(5)
(now fget notices the change EMPTY->[], caches [], and calls extra save)
The new way:
>>> print t.settings["List-type"]==EMPTY
True
(the cached value here is EMPTY)
>>> t.list_type = []
(fget compares cached EMPTY to new [] and saves EMPTY->[])
>>> t.list_type.append(5)
(fget sees no change ([]->[]), which is correct)
In addition to the fix and the related corrections to
testChangeHookMutableProperty, I added details about mutables to all
relevant docstrings and stripped trailing whitespace from both files.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Got rid of a whole bunch of redundant .save() calls when
sync_with_disk==True.
Fixed up the "File-system access" portion of the BugDir docstring so
we can all remember how things are supposed to work ;).
Note that some .save() calls are still required. For example in
becommands/merge.py, the copied comments have their .bug changed, but
that is not a versioned property, so it doesn't trigger an automatic
save, and we have to force the .save() by hand.
libbe.rcs.RCS.mkdir() is now recursive by default, but you can set
check_parents==False if you want it to fail in the case of missing
parents. Because of the recursion, we removed the .update() call
on preexisting directories, since there will be at least one of
these occurrences for every .mkdir(check_parents=True) call, and
I don't know of any VCS that actually needs them...
Also stripped trailing whitespace from some files...
|
| |
|
|
|
|
|
|
| |
From back before commit
wking@drexel.edu-20090619184215-nfx205yaj02sqrqx
cleaned up the versioned_property implementation.
Also a few style fixes and typos.
|
| |
|
|
|
|
|
| |
Previously many backends would silently add an empty commit. Not very
useful. When the new --allow-empty flag and related allow_empty
options are false, every versioning backend is guaranteed to raise the
EmptyCommit exception in the case of an attempted empty commit.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Now we can commit changes from the command line with a unified
interface. The interface is much less flexible than using your
particular version control system's commit command directly, so this
command is mostly intended for user-interfaces and other tools that
don't want to be bothered with the extra flexibility.
Normalized spacing in rcs.RCS.commit to produce:
summary
<BLANKLINE>
body
<TRAILING-ENDLINE>
messages regardless of the input string format.
Also fixed a "--complete" handline bug in cmdutil, and some minor
docstring typos in libbe.rcs and .editor.
|
| |
|
|
|
|
| |
I'd forgotten to prefix the directory root, so calling
be show --only-raw-body COMMIT-ID
would fail if you weren't executing it in the repository root.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Fixes Ben's bug 00f26f04-9202-4288-8744-b29abc2342d6.
I also tweaked update_copyright.sh to make possible future
copyright-blurb revision easier. The new algorithm is greedier,
overwriting _all_ consecutive comments after a '^# Copyright' line, so
do
# Copyright
# GPL ... GPL ... GPL
# Your comment here...
not
# Copyright
# GPL ... GPL ... GPL
#
# Your comment here...
Without the blank line, your comment would get overwritten by the next
run of update_copyright.sh.
Note that catmutt is ignored by update_copyright.sh because Moritz
Barsnick has only licensed his grepm code under the GPLv2 (not
GPLv>=2). See the initial catmutt commit for details.
|
| | |
|
| |
|
|
|
|
| |
All the becommands have been using cmdutil CmdOptionParser for a long
time, but "be" parsed its options by hand. Now it used
CmdOptionParser, which makes adding new options much easier.
|
| |
|
|
| |
Now you don't have to edit them out by hand.
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reworked to allow "be comment" to handle unicode strings (see bug
e4ed63f6-9000-4d0b-98c3-487269140141). The solution was to escape all
the unicode to produce and ASCII string before calling
ElementTree.XML, and then converting back to unicode afterwards.
Added a unicode-containing comment to the end of bug
f7ccd916-b5c7-4890-a2e3-8c8ace17ae3a so that there's a handy unicode
comment for testing.
XML headers (e.g. '<?xml version="1.0" encoding="UTF-8" ?>') are
now added to all xml output from be.
Switched non-text/* encoding library to base64 instead of
email.encoders, which makes that code in libbe/comment.py simpler.
Changed libbe/mapfile.py error encoding from string_escape to
unicode_escape so it can handle unicode.
Everything's still untested, and be-xml-to-mbox doesn't handle unicode
yet, but I felt this commit was getting a bit unwieldy ;).
|
| |
|
|
|
|
|
|
|
| |
I don't really like the "defaults to None" for the settings that have
funky initialization procedures (most of them :p), but I'm not sure
how to handle that cleanly yet. Perhaps
be set --current
I also need to find a method of adding complicated settings like the
nested lists for severities, etc from the "be set" commandline.
|
| |
|
|
|
|
|
|
|
|
|
| |
These didn't work with my update_copyright.sh.
I went with
Aaron Bentley and Panometrics, Inc.
instead of
Aaron Bentley <abentley@panoramicfeedback.com> and Panometrics, Inc.
just because of line length, but I'm open to convincing if people
prefer the latter...
|
| | |
|
| |
|
|
|
| |
Also adjusted libbe/comment.py to move to user-specified alt_ids,
rather than uuids.
|
| |
|
|
|
|
| |
Now the behavior conforms to the docstring:
If both default and generator are None, then the property will be a
defaulting property which defaults to None.
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
| |
Removed superfluous nesting in ./be's error catching. Also replaced
KeyErrors due to unknown commands with the more specific
cmdutil.UnknownCommand, since all sorts of programming errors can
raise KeyErrors.
Untested, since my working tree is a mess at the moment, but what
could go wrong? ;)
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
| |
I don't know much darcs, so I make no claims about the beauty of my
implementation. It seems to get the job done though, until a darcs
guru comes along.
I also tweaked the libbe.git.Git._rcs_get_user_id to handle the case
where user.name or user.email are not set.
I also added the option to pass a stdin string into the
libbe.rcs.RCS._u_invoke* functions.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The main problem was the encoding/decoding that was happening to _all_
input/output. Now many I/O activities have a `binary' option to
disable any encoding/decoding. The `binary' flag is set whenever the
comment content-type is not a text/* type.
In order to print valid XML (and make life easy on xml/be-xml-to-mbox),
non text/* types are printed out as base64-encoded MIME messages, so
be list --xml | be-xml-to-mbox | catmutt
works as you'd expect.
With the standard (non-XML) output from `be show', we just print a
message telling the user that we can't reasonably display the MIME
type and that they should use the XML output if they want to see it.
|
| | |
|
| |
|
|
|
|
|
|
|
| |
This makes all the failed tests understandable, since they all crash
with strings like:
AssertionError: Arch RCS not found
Which makes more sense than spitting out the raw CommandError.
It also means that installed_rcs() actually works now ;).
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
| |
Just return an empty dict instead.
Steps to reproduce:
$ mkdir /tmp/BE-test
$ cd /tmp/BE-test
$ be set-root
$ be new 'having too much fun'
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
extra_strings returns to a defaulting property from a cached/generator
property, with the help of the new, mutable defaults. Lots of
deepcopies avoid mutable default uncertainty too ;). And
copy.deepcopy([]) should be pretty cheap.
tag --remove had previously left settings["extra_strings"] as [],
which polluted the bug's values file. Now the improved
defaulting_property notices a return to the default [], and sets the
internally stored value to EMPTY.
I struggled with creating a more intuitive way to notice changes to
extra_strings than the
tmp = bug.extra_strings
<work on tmp>
bug.extra_strings = tmp
but didn't have any luck. The problem seems to be that if you only
hand out copies of your default, you don't have any pointers to what
you handed out to check for changes. On the other hand, if you hand
out your original default, any external changes will _change_ your
original default. I suppose you could only hand out copies, but keep
a list of all copies handed out, but that sounds like a disaster.
Reassigning is easy enough.
|
| |
|
|
| |
This avoids the problems associated with mutable defaults.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
This continues the line of changes started in libbe/properties with
the last two commits. Also straightened up stranch double-default in
libbe.settings_object.versioned_property and moved the fn_checked
before checked, which shouldn't matter because I never use both at
once, and can't think of a case where you'd want to.
I've also added some docstrings to the settings_object unit tests,
since apparently docstrings get printed during the test if they exist,
and they look nicer than the name of the unittest itself. More like
./configure output ;).
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Now (except for a wimpy hash function) it's as good as it's going to
get for true mutables. Calls to change_hook occur for all changes,
sometime after the change-enducing action and before the next
attribute access. See testChangeHookMutableProperty for an example of
the expected behavior.
If you're doing some mutable-modification (e.g. t.x.append(5)) and you
want to `flush' the changes into a change_hook call, just assign t.x
to a dummy variable. e.g.
t.x.append(5)
dummy = t.x
If you _really_ need post-modification change_hook calls without such
a flush, you're on your own. Would you get the property-owning class
to poll for changes?
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Versioned properties whose data is a mutable type are tricky, since
the simple comparisons we'd been using in
libbe.properties.change_hook_property don't work for mutables. For
now, we avoid that problem by assuming a change happened whenever a
mutable property is set. change_hook_property is a bit untidy at the
moment while I work out how to deal with mutables.
As an example of using Bug.extra_strings to patch on some useful
functionality, I've written becommands/tag.py. I'd suggest future
add-ons (e.g. becommands/depend.py?) use the "<LABEL>:<value>" string
format to keep it easy to sort out which strings belong to which
add-ons. tag.py is still missing command line tag-removal and
tag-searching for `be list'. Perhaps something like
be list --extra-strings TAG:<your-tag>,TAG:<another-tag>,DEPEND:<bug-id>
would be good, although it would requre escaping commas from the tags,
or refusing to allow commas in the tags...
libbe.properties.ValueCheckError also got a minor update so the
printed error message makes sense when raised with allowed being an
iterable (i.e. check_property) or a function
(e.g. fn_checked_property).
All of this digging around turned up a really buggy
libbe.bugdir.MultipleBugMatches. Obviously I had never actually
called it before :p. Should be fixed now.
libbe.comment._set_comment_body has also been normalized to match the
suggested change_hook interface: change_hook(self, old, new).
Although, I'm not sure why it hadn't been causing obvious problems
before, so maybe I'm misunderstanding something about that.
|
| |
|
|
|
|
|
|
|
|
|
| |
Also added libbe.bug.cmp_last_modified, which handles part of
9ce2f015-8ea0-43a5-a03d-fc36f6d202fe. To do better we could extend
the RCS framework.
I also transcribed a few emails from the be-devel list onto their
relavent bugs and closed a few bugs.
Finally, I removed some left over InvalidValue cruft.
|
| | |
|
| | |
|
| |
|
|
|
|
| |
Since
<creator>John Doe <jdoe@example.com></creator>
is not valid XML.
|
| |
|
|
| |
The xml() method hadn't been updated since the settings_object revamp.
|
| |
|
|
|
| |
Renamed "name" -> "short-name" and "in_reply_to" -> "in-reply-to".
Reordered uuid before short-name.
|
| |
|
|
|
|
|
|
| |
Another bug introduced by James Rowe's user-config patch. Obviously
it's hard to read a file if there's no file there. I'm not sure how
it passed the unit tests earlier. Maybe I forgot to install the
pre-commit version before running the test suite... Anyhow, fixed
now.
|
| |\
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
This involved an `upgrade' of BE's bzr repo
Previous version (via `bzr info path/to/repo`): pack-0.92
Current version: rich-root-pack
The whole rich-root thing is a bzr features-vs-backwards-compatability
thing they've been wrestling with [1,2,3,4,...]. It seems that BE was
in some sort of unstable equilibrium [5], so I'll follow Ben's lead
and make the official switch. Note that you'll need to use bzr>=1.5
to make the shift [6]. For the sake of completeness, the whole
rich-root thing was introduced here [7], but I don't understand enough
of bzr to make sense of the diff. It just versions the repo's root
directory the same way it versions other directories [3]. The bzr
people seem to be planning to phase out non-rich-root formats in favor
of brisbane-core, aka 2.0beta [8], by bzr 2.0 [8], which is apparently
on the horizon [9,10,11]. What a headache.
Citations are all titles/X-List-Received-Date from
https://lists.ubuntu.com/archives/bazaar/
with the exception of the URL [11].
[1] [RFC] rich root pack as default in 1.8 ?
Sat, 06 Sep 2008 03:33:46 -0000
(conclusion: none)
[2] Re: 1.9rc1 countdown
Thu, 30 Oct 2008 08:44:53 -0000
(conclusion: "primary" format should be rich-root next time we make a new format)
[3] So many repo formats
Fri, 14 Nov 2008 08:41:33 -0000
Mon, 17 Nov 2008 07:37:47 -0000 (explains rich-root format)
Mon, 17 Nov 2008 22:37:39 -0000 (explains no-return policy)
Mon, 17 Nov 2008 20:57:08 -0000 (explicitly lists non-svn reasons for rich-root)
[4] Branch fails from 'pack-0.92' repo to 'rich-root-pack' repo.
Wed, 27 Aug 2008 11:31:11 -0000
(we're not sure again)
[5] Branch fails from 'pack-0.92' repo to 'rich-root-pack' repo.
Sun, 20 Apr 2008 12:58:09 -0000
[6] Branch fails from 'pack-0.92' repo to 'rich-root-pack' repo.
Fri, 29 Aug 2008 13:23:52 -0000
[7] [RFC] Knit format 2
Fri, 25 Aug 2006 22:55:36 -0000
[8] bazaar 2.0beta format for launchpad release
Fri, 29 May 2009 06:00:03 -0000
[9] Upgrading loggerhead to 1.9-rich-root
Mon, 11 May 2009 22:35:28 -0000
(mentions eventual switch to rich-root in 2.0)
[10] bzr 1.16rc1 released!
Fri, 12 Jun 2009 08:00:08 -0000
(confirms eventual switch to rich-root in 2.0)
[11] https://launchpad.net/bzr/+announcement/2733
(current outstanding releases: 1.17, 2.0)
|
| | | |
|
| | | |
|
| | | |
|
