Still broken in 6dab29f6df37 after mpm's recent changes. No "abort: outstanding
uncommitted changes" now, but final output is still "a c" and stat in
hgtest-fetch-a still shows "M b; R a". As before, replacing fetch with
pull;merge;commit produces the correct result.


Note that Hg now also prints

failed to contact inotify server: No such file or directory

after each clone, which is annoying. Improving inotify/__init__.py a bit to read:

                    ui.warn(_('failed to contact inotify server from %s: %s\n')
                             % (self._rootdir, err[-1]))

produces

failed to contact inotify server from /tmp/hgtest-fetch-a/: No such file or
directory
...
failed to contact inotify server from /tmp/hgtest-fetch-b/: No such file or
directory

This seems a gratuitous warning message - the clone has just been created, so of
course there is no server active in it yet.

djc: yes

Anyway I need to understand better what the problems are.

Is the print statement here for debugging?

ok, it seems inotify doesn't handle negative size and time (they have a special
meaning in the dirstate).

diff --git a/hgext/inotify/server.py b/hgext/inotify/server.py
--- a/hgext/inotify/server.py
+++ b/hgext/inotify/server.py
@@ -248,14 +248,15 @@
         try:
             type_, mode, size, time = self.repo.dirstate._map[fn][:4]
         except KeyError:
-            type_ = '?'
+            type_ = st and '?' or None
+        print type_, mode, size, time, fn, st
         if type_ == 'n':
             if not st:
                 return '!'
             st_mode, st_size, st_mtime = st
-            if size and (size != st_size or (mode ^ st_mode) & 0100):
+            if size >= 0 and (size != st_size or (mode ^ st_mode) & 0100):
                 return 'm'
-            if time != int(st_mtime):
+            if time != int(st_mtime) or size < 0:
                 return 'l'
             return 'n'
         if type_ in 'ma' and not st:

Forget what I said. Hg 1.0+ apparently does not activate the inotify extension
on demand, you need to run inserve. Given that, I make up a new test script:


#!/bin/bash
killall hg
set -e -x
hg version
cd /tmp
rm -rf hgtest-fetch*
hg init hgtest-fetch
hg -R hgtest-fetch inserve &
hg clone hgtest-fetch hgtest-fetch-a
hg -R hgtest-fetch-a inserve &
hg clone hgtest-fetch hgtest-fetch-b
hg -R hgtest-fetch-b inserve &
cd hgtest-fetch-a
touch a
hg add a
hg ci -m.
hg push
cd ../hgtest-fetch-b
hg pull -u
ls
touch b
hg add b
hg rem a
hg ci -m.
ls
hg push
cd ../hgtest-fetch-a
touch c
hg add c
hg ci -m.
hg fetch --switch-parent
#hg pull
#hg merge
#hg ci -m.
ls
hg push
cd ../hgtest-fetch-b
ls
hg pull -u
ls


which gives me a different problem:


hg: no process killed
+ hg version
Mercurial Distributed SCM (version 1.0.2)

Copyright (C) 2005-2008 Matt Mackall <mpm@selenic.com> and others
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ cd /tmp
+ rm -rf hgtest-fetch hgtest-fetch-a hgtest-fetch-b
+ hg init hgtest-fetch
+ hg -R hgtest-fetch inserve
+ hg clone hgtest-fetch hgtest-fetch-a
updating working directory
0 files updated, 0 files merged, 0 files removed, 0 files unresolved
+ hg -R hgtest-fetch-a inserve
+ hg clone hgtest-fetch hgtest-fetch-b
updating working directory
0 files updated, 0 files merged, 0 files removed, 0 files unresolved
+ hg -R hgtest-fetch-b inserve
+ cd hgtest-fetch-a
+ touch a
+ hg add a
+ hg ci -m.
+ hg push
pushing to /tmp/hgtest-fetch
searching for changes
adding changesets
adding manifests
adding file changes
added 1 changesets with 1 changes to 1 files
+ cd ../hgtest-fetch-b
+ hg pull -u
pulling from /tmp/hgtest-fetch
requesting all changes
adding changesets
adding manifests
adding file changes
added 1 changesets with 1 changes to 1 files
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
+ ls
a
+ touch b
+ hg add b
+ hg rem a
not removing a: file is modified (use -f to force removal)
+ hg ci -m.
+ ls
a  b
+ hg push
pushing to /tmp/hgtest-fetch
searching for changes
adding changesets
adding manifests
adding file changes
added 1 changesets with 1 changes to 1 files
+ cd ../hgtest-fetch-a
+ touch c
+ hg add c
+ hg ci -m.
+ hg fetch --switch-parent
abort: outstanding uncommitted changes


For some reason, 'hg rem a' claims 'a' is modified when it clearly is not. Look at:

tmp$ hg -R hgtest-fetch-b st
M a
M b
tmp$ hg -R hgtest-fetch-b di
tmp$ 

I have seen this also in my own repo when trying to use inotify in Hg 1.0.2;
incorrectly says things are modified even right after I 'hg revert' them. Until
this more obvious bug is fixed, I have to assume the bug originally reported
here could still be broken. (Using 'hg rem -f a' in the test case does not help,
because it just dies later on in 'hg fetch' claiming outstanding changes.)

I cannot reproduce this bug in Hg 1.0.2. (Note that the behavior of 'hg fetch'
changed in 1.0, so I tried also with --switch-parent to better mimic the 0.9.5
merge order.)

I am unfortunately unable to confirm that the bug is still reproducible in
0.9.5, because I cannot use the tip of
http://hg.kublai.com/mercurial/extensions/inotify with Hg 0.9.5 built from source:

....
  File "/space/src/hg/build/lib.linux-i686-2.5/mercurial/commands.py", line
2571, in status
    list_clean=all or opts['clean'])]
  File "/space/src/hg/build/lib.linux-i686-2.5/mercurial/localrepo.py", line
909, in status
    list_ignored, list_clean)
  File "/space/src/hg-inotify/__init__.py", line 83, in status
    list_unknown)
TypeError: status() takes exactly 5 arguments (6 given)

@jglick: can you try a more recent version ? can you still reproduce it ?

I can't reproduce here (with inotify):

+ hg pull -u
pulling from /tmp/hgtest-fetch
searching for changes
adding changesets
adding manifests
adding file changes
added 2 changesets with 1 changes to 1 files
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
+ ls
b  c

Degrading to bug, as disabling the extension will fix it.

Ha, got it: bug only occurs when the inotify extension is enabled.

I am attaching a version of your initial script usable as a mercurial test. Can
you reproduce it when run as a test ?

I cannot reproduce it with your script and 0.9.5 on macosx 10.4, the last pull
gives me:
"""
+ cd ../hgtest-fetch-b
+ ls
b
+ hg pull -u
pulling from /private/tmp/hgtest-fetch
searching for changes
adding changesets
adding manifests
adding file changes
added 2 changesets with 1 changes to 1 files
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
+ ls
b  c
"""

Run:

---%<---
#!/bin/bash
set -e -x
hg version
cd /tmp
rm -rf hgtest-fetch*
hg init hgtest-fetch
hg clone hgtest-fetch hgtest-fetch-a
hg clone hgtest-fetch hgtest-fetch-b
cd hgtest-fetch-a
touch a
hg add a
hg ci -m.
hg push
cd ../hgtest-fetch-b
hg pull -u
ls
touch b
hg add b
hg rem a
hg ci -m.
ls
hg push
cd ../hgtest-fetch-a
touch c
hg add c
hg ci -m.
hg fetch
#hg pull
#hg merge
#hg ci -m.
ls
hg push
cd ../hgtest-fetch-b
ls
hg pull -u
ls
---%<---

I get:

---%<---
+ hg version
Mercurial Distributed SCM (version 0.9.5)

Copyright (C) 2005-2007 Matt Mackall <mpm@selenic.com> and others
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ cd /tmp
+ rm -rf hgtest-fetch hgtest-fetch-a hgtest-fetch-b
+ hg init hgtest-fetch
+ hg clone hgtest-fetch hgtest-fetch-a
0 files updated, 0 files merged, 0 files removed, 0 files unresolved
+ hg clone hgtest-fetch hgtest-fetch-b
0 files updated, 0 files merged, 0 files removed, 0 files unresolved
+ cd hgtest-fetch-a
+ touch a
+ hg add a
+ hg ci -m.
+ hg push
pushing to /tmp/hgtest-fetch
searching for changes
adding changesets
adding manifests
adding file changes
added 1 changesets with 1 changes to 1 files
+ cd ../hgtest-fetch-b
+ hg pull -u
pulling from /tmp/hgtest-fetch
requesting all changes
adding changesets
adding manifests
adding file changes
added 1 changesets with 1 changes to 1 files
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
+ ls
a
+ touch b
+ hg add b
+ hg rem a
+ hg ci -m.
+ ls
b
+ hg push
pushing to /tmp/hgtest-fetch
searching for changes
adding changesets
adding manifests
adding file changes
added 1 changesets with 1 changes to 1 files
+ cd ../hgtest-fetch-a
+ touch c
+ hg add c
+ hg ci -m.
+ hg fetch
pulling from /tmp/hgtest-fetch
searching for changes
adding changesets
adding manifests
adding file changes
added 1 changesets with 1 changes to 1 files (+1 heads)
merging with new head 2:25015baca70d
1 files updated, 0 files merged, 1 files removed, 0 files unresolved
new changeset 3:2e03af81f688 merges remote changes with local
+ ls
b  c
+ hg push
pushing to /tmp/hgtest-fetch
searching for changes
adding changesets
adding manifests
adding file changes
added 2 changesets with 1 changes to 1 files
+ cd ../hgtest-fetch-b
+ ls
b
+ hg pull -u
pulling from /tmp/hgtest-fetch
searching for changes
adding changesets
adding manifests
adding file changes
added 2 changesets with 1 changes to 1 files
2 files updated, 0 files merged, 1 files removed, 0 files unresolved
+ ls
a  c
---%<---

For some reason the merge done by the Fetch extension affects the wrong files.
In fact the correct merge is left in the working dir but not committed:

---%<---
$ hg -R hgtest-fetch-a stat
M b
R a
---%<---

If you uncomment the three lines to use regular pull+merge+ci instead of fetch,
it works as expected; the last line shows 'b c'.