keyword expansion collapses non-keywords

[Matthew Fuller]

When expansion is enabled, non-keywords are in danger of getting collapsed.

This is visible several places in the bzr source, for instance. To reproduce, add a rules file like

[name *.txt]
keywords = on

and then cd bzr.dev ; rm doc/en/admin-guide/hooks-plugins.txt ; bzr revert ; bzr diff

[Tom_OConnor (tom-twinhelix)]

I have a test case:
file "test.php" contains :

// Simplest case that breaks things:
$a-b::c
$d

// Why we'd do this:
echo "There are ".$thispage-SearchResults::NumPages()." more pages of results";
$nextpage = $thispage + 1;

[tom@turmeric test]$ bzr ci test.php
then becomes

[tom@turmeric test]$ bzr cat test.php
// Simplest case that breaks things:
$a-b$d

// Why we'd do this:
echo "There are ".$thispage-SearchResults$nextpage = $thispage + 1;

Using latest bzr-keywords as produced with `bzr co lp:bzr-keywords`

[Ian Clatworthy (ian-clatworthy)]

This is a bad bug. Branch specific rules will help this by isolating the working trees the expansion is applied to. Even without that though, I think we need to err on the conservative side here and be stricter about what's considered a keyword vs otherwise.

[Matthew Fuller (fullermd)]

Yeah. It's bad in general, but in languages with sigils, it's a
showstopper :|

[Tom_OConnor (tom-twinhelix)]

I created a local branch, and was trying to experiment with only matching known keywords thusly:

known_tag_list = ['Date','Committer','Authors','Revision-Id','Path','Directory','Filename','File-Id','Now','User','Comm
itter-Name','Committer-Email','Author1-Name','Author1-Email','Author2-Name','Author2-Email','Author3-Name','Author3-Ema
il','User-Name','User-Email']

_KW_COOKED_RE_DICT = {}
_KW_RAW_RE_DICT= {}

for known_tag in known_tag_list:
        _KW_COOKED_RE_DICT[known_tag] = re.compile(r"\$(%s)(:[^$]*)?\$" % known_tag)
        _KW_RAW_RE_DICT[known_tag] = re.compile(r"\$(%s)(:[^$]*)?\$" % known_tag)

...
for regex in _KW_RAW_RE_DICT.keys():
                match = _KW_RAW_RE_DICT[regex].search(rest)

But didn't manage to get it working.

[Andrew Johnson (anj)]

I've just pushed a branch that fixes this by not compressing keywords that are not registered (I'll attach the patch here too for completeness). However if the keyword compression has already happened then your repository has lost the original cooked value and you'll have to fix and commit any instances manually.

[Andrew Johnson (anj)]

See my previous comment...