[Cvsnt] CVSNT error handling and Eclipse/WSAD branch browsing

[Schmidt, Martin]

We have been using IBM WebSphere Application Developer 4.02 / Eclipse R1.0
with CVSNT, and have not been able to immediately browse newly created
streams as can be done with unix versions of CVS.  After spending some time
working with the 57b source, I have hacked together a couple of fixes to
compensate for some divergent behavior in CVSNT error handling.  I have
noticed that the latest server sources have been drastically changed, but it
is not clear to me, however, that the new server code will address this
problem.  Consequently, I will describe my understanding of the problem and
appeal to the author(s) of the new server code to allay my fears and/or test
for and address this issue.

When a new stream/branch is "created" in Eclipse, Eclipse simply creates a
name for the stream/branch (which may be preexisting) and it shows up in the
repository tree view.  If the user then expands (+) that stream to see its
contents, they should see all current projects/modules in the repository.
With unix CVS, this occurs correctly, but with CVSNT (through 57c) no
expansion occurs.  This forces users to engage in cumbersome workaround
tricks to add a project to the workspace from the new stream (checkout).  By
using windump to examine the data exchange between Eclipse and CVSNT, and
between Eclipse and CVS running on unix, I have determined that CVSNT is not
returning certain error data that Eclipse depends upon.

When a new stream/branch is first expanded, Eclipse sends (via remote
protocol):

Global_option -n
Argument -r
Argument streamname
Argument -d
Directory .
cvsrootdirname
Argument .
update

Since a branch tag for streamname does not yet exist in the repository, CVS
(unix) replies:

E cvs [server aborted]: no such tag streamname

CVSNT will also send this reply, but ONLY on the very first execution of
cvs.exe, successive tries will fail to return any error.  Replacing the
cvs.exe with a copy or a new build seems to unlink you from whatever
external data is influencing this.  I never found what precisely was causing
this in the 57b source, but I "fixed" it by putting a sleep(1) at the
beginning of server_cleanup() in server.c.

After receiving the above error response from the CVS server, Eclipse
further waits for the response:

error

This is the seemingly redundant piece of output written from do_cvs_command
if resulting errs > 0 from command execution.  	/* We will have printed an
error message already.  */
	buf_output0 (buf_to_net, "error  \n");
It doesn't seem that this code is reachable for an abort severity error in
the 57b source.
Eclipse needs this response, however, before it will retry with the more
general request:

Argument -d
Directory .
cvsrootdirname
Argument .
update