Strange error wih WinCVS client (and tigris cvs server)

[colm]

Hello all,

This is a bit of a strange one as I can only reproduce it using the
Windows CVS client. Also, I am using a patched version of cvs which is
part of the tigris suite which Collabnet used to host as an open
source project but which now is sadly only available to their extranet
clients. The last version of the tigris suite to retain its
open-source status was August 2001.

Basically their version was patched to call out to a seperate proces
to perform client authourisation. An external perl script is used to
query a database to find out if the combination of user, project (i.e.
repository) and action (e.g. commit, update) is permitted. While
debugging some strangeness with the Windows client I came across odd
behaviour which someone might be able to help me clear up. My
ignorance of cvs' internal workings means that I don't know whether
this is the fault of the windows client or something more subtle. This
is the behaviour:

If, from my working directory I co a source tree and then move down a
few levels, edit a file and attempt an update I get an error from
tigris' permission's system. Debugging the external script I realised
that the script was being told by cvs that the project/repo this file
belonged to was the  current directory (where the command was being
issued i.e. the first on the path to the file to be committed) instead
of the correct one (several levels above). If I go back to the working
dir and commit with the full path of the chaged file, it succeeds. A
similar co, edit, commit from a linux command line works with no
problem (even over the network) from the edited file's directory.

My questions are these: How does cvs know which repo a file belongs
to? I imagine its from the local CVS directory using the Repository
admin file? If this is the case then it suggests to me either that the
windows client is broken in this respect or that the tigris external
script is being passed an incorrect parameter for its project (i.e.
repo) criterion. The former seems hard to believe since it would
surely have been discovered before but I'd be pleased if someone could
confirm my assumptions. I don't have a vanilla cvs server to test
against.

The server is a patched 1.11 on RedHat 6.2 and client is 1.11 on Win2k
.

Many thanks for any assistance,
colm