Re: [coreutils] Re: Bug#609049: du: process_file: Assertion `level == prev_level - 1' failed

[Jim Meyering]

Pádraig Brady wrote:

> On 09/01/11 22:05, Jim Meyering wrote:
>> diff --git a/tests/du/move-dir-while-traversing
>> b/tests/du/move-dir-while-traversing
>
>> +# We use a python-inotify script, so...
>> +python -m pyinotify -h > /dev/null \
>> +  || skip_ 'python-inotify package not installed'
>
> A small point is that error is a bit fedora specific.
> The package is python-pyinotify on debian/ubuntu for example.

Good point.  I've dropped the "-":

  || skip_ 'python inotify package not installed'

>> +from pyinotify import *
>
> I generally don't include everything from a module,
> to keep the namespace clean.
> I'd do instead:
>
> import pyinotify as pn
> import os,sys
> ...

I prefer that, too.

>> +dir = sys.argv[1]
...
>> +print 'started'
>
> The above print is buffered by default.
> As we're using it for synchronization I'd
>
> sys.stdout.write('started\n')
> sys.stdout.flush()

Good catch!

...
>> +nonempty() { test -s start-msg && return 0; sleep $1; }
>> +retry_delay_ nonempty .1 5
>
> I think the above may iterate only once?
> How about:
>
> nonempty() {
>   test -s start-msg || { sleep $1; return 1; }
> }
> retry_delay_ nonempty .1 5 || framework_failure

Yes, that fixes another bug.

>> +# The above delay is insufficient in ~50% of my trials.
>> +# Sometimes, when under heavy load, a parallel "make check" would
>> +# fail this test when sleeping just 0.1 seconds here.
>> +sleep 1
>
> Hopefully this extra sleep is not required now

Bingo.  It does, so far.
Thanks for the advice and fixes.
Here's the incremental I'm testing with:

diff --git a/tests/du/move-dir-while-traversing 
b/tests/du/move-dir-while-traversing
index e42bc93..fe1615c 100755
--- a/tests/du/move-dir-while-traversing
+++ b/tests/du/move-dir-while-traversing
@@ -21,7 +21,7 @@ print_ver_ du

 # We use a python-inotify script, so...
 python -m pyinotify -h > /dev/null \
-  || skip_ 'python-inotify package not installed'
+  || skip_ 'python inotify package not installed'

 # Move a directory "up" while du is processing its sub-directories.
 # While du is processing a hierarchy .../B/C/D/... this script
@@ -33,12 +33,14 @@ python -m pyinotify -h > /dev/null \

 cat <<'EOF' > inotify-watch-for-dir-access.py
 #!/usr/bin/env python
-from pyinotify import *
+import pyinotify as pn
+import os,sys
+
 dir = sys.argv[1]
 dest_parent = os.path.dirname(os.path.dirname(dir))
 dest = os.path.join(dest_parent, os.path.basename(dir))

-class ProcessDir(ProcessEvent):
+class ProcessDir(pn.ProcessEvent):

     def process_IN_OPEN(self, event):
         os.rename(dir, dest)
@@ -47,10 +49,11 @@ class ProcessDir(ProcessEvent):
     def process_default(self, event):
         pass

-wm = WatchManager()
-notifier = Notifier(wm)
-wm.watch_transient_file(dir, IN_OPEN, ProcessDir)
-print 'started'
+wm = pn.WatchManager()
+notifier = pn.Notifier(wm)
+wm.watch_transient_file(dir, pn.IN_OPEN, ProcessDir)
+sys.stdout.write('started\n')
+sys.stdout.flush()
 notifier.loop()
 EOF
 chmod a+x inotify-watch-for-dir-access.py
@@ -61,14 +64,9 @@ mkdir -p $t/3/a/b/c/$long d2 || framework_failure
 timeout 6 ./inotify-watch-for-dir-access.py $t/3/a/b > start-msg &

 # Wait for the watcher to start...
-nonempty() { test -s start-msg && return 0; sleep $1; }
+nonempty() { test -s start-msg || { sleep $1; return 1; }; }
 retry_delay_ nonempty .1 5

-# The above delay is insufficient in ~50% of my trials.
-# Sometimes, when under heavy load, a parallel "make check" would
-# fail this test when sleeping just 0.1 seconds here.
-sleep 1
-
 # The above watches for an IN_OPEN event on $t/3/a/b,
 # and when it triggers, moves the parent, $t/3/a, up one level
 # so it's directly under $t.