CVS issues

CVS issues

[JD Smith]

With a recent build of CVS Emacs, I've experienced problems relating to
the "cvs" command in VC mode.  E.g. C-x v l often results in:

Process cvs killed

or sometimes "No differences" on a file which has changed.  It is
intermittent, and only about 50% of the time does the command succeed. 
Emacs 21.4 does not have this problem, succeeding every time.  These issue
also cause the log buffer window to be much smaller than it normally
would (like 3 lines high), which persists even when the command later
succeeds.

In case it matters, I have:

Concurrent Versions System (CVS) 1.11.19 (client/server)

JD

Re: CVS issues

[JD Smith]

On Mon, 17 Apr 2006 12:34:19 -0700, JD Smith wrote:

> 
> With a recent build of CVS Emacs, I've experienced problems relating to
> the "cvs" command in VC mode.  E.g. C-x v l often results in:

I should mention the same happens with C-x v = (cvs diff), which
results in "No differences" and "Process cvs killed" more commonly.

JD

Re: CVS issues

[Stefan Monnier]

>> With a recent build of CVS Emacs, I've experienced problems relating to
>> the "cvs" command in VC mode.  E.g. C-x v l often results in:
> I should mention the same happens with C-x v = (cvs diff), which
> results in "No differences" and "Process cvs killed" more commonly.

We've seen some reports of such a problem and I've experienced it as well.
I can't reproduce it, but I have the impression that it's timing related.
In my case, the *vc-diff* buffer is shown in a separate frame, and if that
frame is visible when I hit C-x v = the command normally suceeds, whereas if
the frame needs to be deiconified (or worse, created) as part of the C-x v =
command, then the risk seems to be higher that I get "No differences".
I've never experienced the "Proces killed" problem, OTOH.

My impression until now was that if some X window event comes in just at the
wrong time, Emacs would notice that the cvs (or tla in my case) process
exited without reading all the pending output from that process.
Whether that's really what's happening or not, I don't know.

This was all on GNU/Linux systems.


        Stefan

Re: CVS issues

[JD Smith]

On Mon, 17 Apr 2006 16:58:48 -0400, Stefan Monnier wrote:
 
> We've seen some reports of such a problem and I've experienced it as well.
> I can't reproduce it, but I have the impression that it's timing related.

When I encounter it again, can you suggest an effective way to debug the
problem?  It seems to be related to a particular Emacs session: some
sessions have no problems, others are hit and miss.

JD

Re: CVS issues

[Richard Stallman]

    With a recent build of CVS Emacs, I've experienced problems relating to
    the "cvs" command in VC mode.  E.g. C-x v l often results in:

    Process cvs killed

Has there been any progress debugging this?

Since this is not reproducible, I see only two ways to debug it:

* Add code to record some history information that you could
look at after a failure.

* Try building checkouts from various dates, hoping thus to determine
which change caused the problem.

You recall approximately when you made a build that did not have this
problem?

Re: CVS issues

[JD Smith]

On Tue, 2006-04-25 at 12:48 -0400, Richard Stallman wrote:
>     With a recent build of CVS Emacs, I've experienced problems relating to
>     the "cvs" command in VC mode.  E.g. C-x v l often results in:
> 
>     Process cvs killed
> 
> Has there been any progress debugging this?
> 
> Since this is not reproducible, I see only two ways to debug it:
> 
> * Add code to record some history information that you could
> look at after a failure.
> 
> * Try building checkouts from various dates, hoping thus to determine
> which change caused the problem.
> 
> You recall approximately when you made a build that did not have this
> problem?

No progress here.  I can only say for certain that 21.X did not show the
problem (i.e., I'd experience the problem, run an old Emacs 21.4 on the
same files, repeat the same commands, and have no problem).  As far as I
know, all versions since then are affected (though I have only
experienced it in the last several months).  I have not successfully
reproduced it with Emacs from CVS since I reported it.  

The one potential silver lining is it seems to depend somehow on the
session, such that once a cvs command fails, it is likely to fail again
in that same Emacs session.  This may in fact result from persistent
external conditions, like the difficulty of SSH to connect to a CVS
server, etc.  If I encounter it again, I'll leave that session running
and then we can consider how best to get to the bottom of it.

Re: CVS issues

[Richard Stallman]

How often does this problem occur?
Is it days apart?  Weeks?  Months?

Since the bug is not reproducible, I think the only way to fix it
is to add more instrumentation code.

Re: CVS issues

[JD Smith]

On Apr 28, 2006, at 8:50 PM, Richard Stallman wrote:

> How often does this problem occur?
> Is it days apart?  Weeks?  Months?
>
> Since the bug is not reproducible, I think the only way to fix it
> is to add more instrumentation code.

To be honest, I'm not sure.  I've had it happen at least three  
separate times, and typically it persists for a day or more,  
sporadically occurring, approximately half the time.  But it hasn't  
happened in several weeks (though my CVS usage varies considerably).   
My strong suspicion is that this relates to a persistent behavior of  
the CVS host, timing out or otherwise generating some situation which  
CVS Emacs cannot cope with well, but it hasn't repeated enough to  
allow honing in on the problem.  I've tried disconnecting the network  
to set cvs up to fail, but now all I get is the correct standard  
error message.  Since people reading this list no doubt exercise cvs  
via Emacs far more than I do, the fact that many others haven't  
echoed the problem makes me think it might be incredibly rare, or  
pertain to my specific setup in some unknown way.  I'll stay on the  
lookout for it.