keyword: restore kwtemplater.restrict at the end of wrapped patch.diff
Before this patch, kwdiff doesn't restore kwtemplater.restrict after
invocation of wrapped patch.diff(). This suppresses keyword expansion
at subsequent filelog.read().
Typical usecase of this issue is "hg cat" after "hg diff" with command
server. In this case, kwtemplater.restrict=True is kept in command
server process even after "hg diff".
To ensure kwtemplater.restrict=True while original patch.diff()
running, this patch makes kwdiff() yield values returned by it,
because it returns generator object.
Strictly speaking, if filelog.read() is invoked before completely
evaluating the result of previous patch.diff(), keyword expansion is
still suppressed, because kwtemplater.restrict isn't restored yet.
But this fixing should be reasonable enough, because patch.diff() is
consumed immediately, AFAIK.
$ hg init outer
$ cd outer
$ echo '[paths]' >> .hg/hgrc
$ echo 'default = http://example.net/' >> .hg/hgrc
hg debugsub with no remapping
$ echo 'sub = libfoo' > .hgsub
$ hg add .hgsub
$ hg debugsub
path sub
source libfoo
revision
hg debugsub with remapping
$ echo '[subpaths]' >> .hg/hgrc
$ printf 'http://example.net/lib(.*) = C:\\libs\\\\1-lib\\\n' >> .hg/hgrc
$ hg debugsub
path sub
source C:\libs\foo-lib\
revision
test cumulative remapping, the $HGRCPATH file is loaded first
$ echo '[subpaths]' >> $HGRCPATH
$ echo 'libfoo = libbar' >> $HGRCPATH
$ hg debugsub
path sub
source C:\libs\bar-lib\
revision
test absolute source path -- testing with a URL is important since
standard os.path.join wont treat that as an absolute path
$ echo 'abs = http://example.net/abs' > .hgsub
$ hg debugsub
path abs
source http://example.net/abs
revision
$ echo 'abs = /abs' > .hgsub
$ hg debugsub
path abs
source /abs
revision
test bad subpaths pattern
$ cat > .hg/hgrc <<EOF
> [subpaths]
> .* = \1
> EOF
$ hg debugsub
abort: bad subrepository pattern in $TESTTMP/outer/.hg/hgrc:2: invalid group reference (glob)
[255]
$ cd ..