I am sorry to have to make this not-very-useful bug report.  For a few
years now, I have been using NTFS-3G to provide Linux access to Windows
files allowing me to easily switch back to Windows.  This has been quite
trouble free, except for one important exception.  One of the
filesystems I access from Linux via NTFS-3G holds my local CVS checkout
sandbox.  What I have seen is intermittent problems with the cvs update
command (cvs version 1.12.13) when run on Linux (Ubuntu 8.04).

The symptoms are that the source file is properly updated, but the
CVS/Entries files is not updated.  The effect is that while the source
code is up to date, the checked out version still refers to the
pre-update version.  After a successful update "cvs status" should
report that the Working and Repository versions are the same.  But
intermittently I find that some or all of the updated files still show
the old Working revision number.  This failure is silent in the sense
that "cvs update" does not complain or generate a non-zero exit status.

Sometimes I can just reissue the cvs update command and it succeeds.
Other times, I can run the command over and over and it persistently
fails to update the Working revision (usually, it is will work the next
day or on the next update).  I have found it more often fails
persistently when updating an explicit list of files.  Updates of
directories are more often successful.

I have tried debugging this by capturing the strace output, but cvs is
sufficiently idiosyncratic that I have not been able to understand from
this how a working and non-working version differ in their system call
level behavior.

As a test, I recently copied my source tree to an ext3 partition and the
problem completely disappeared.  I've been running this way for several
weeks have have not seen a single problem with cvs update.  This makes a
pretty strong case that the problem is directly related to using NTFS-3G
to store the checked-out files and CVS metadata files.

I realize this report doesn't give NTFS-3G developers much to go on.
However, I am hoping that other users will have noticed similar behavior
and have additional useful details or anecdotes to report.

Ted Anderson

------------------------------------------------------------------------------
Stay on top of everything new and different, both inside and 
around Java (TM) technology - register by April 22, and save
$200 on the JavaOne (SM) conference, June 2-5, 2009, San Francisco.
300 plus technical and hands-on sessions. Register today. 
Use priority code J9JMT32. http://p.sf.net/sfu/p
_______________________________________________
ntfs-3g-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/ntfs-3g-devel

Reply via email to