Vedr.: Re: Vedr.: Re: Vedr.: [Cvsnt] Shadow folder in cvs

Mon Mar 25 13:57:46 GMT 2002

I did not realize the significane of the sleep command in the loginfo!
However there IS in fact a sleep.exe file in the NT Resource Kit, so this
behaviour is surely possible on the NT-platform.  Another matter is that
having the script wait for 2 seconds is hardly fool-proof. The cvs-commit
may for whatever reason not have finished within the 2 seconds. I think it
would be more safe to make syncronization on fixed hours (e.g. with the AT
command on NT), and it will probably perform better, since the original
script (if I'm not mistaken) will do a complete update for each committed
file in a multi-file commit.

Regards

-----------------------------------------
Jørgen Møller Larsen, KMF
Nykredit Data
email: jml at nykredit.dk
Tlf. direkte 96 35 50 69
----------------------------------------

                    John Peacock
                    <jpeacock at row        Til:    jml at nykredit.dk
                    man.com>             cc:     Cvsnt at cvsnt.org
                    Sendt af:            Vedr.:  Re: Vedr.: Re: Vedr.: [Cvsnt] Shadow folder in cvs
                    cvsnt-admin at c
                    vsnt.org

                    2002-03-22
                    18:08
                    Besvar
                    venligst til
                    John Peacock

jml at nykredit.dk wrote:
> Dear John
>
> First: sometimes a message from somebody (say you) will cross a message
> from someone else (say me), and someone may not have had a chance to read
> the message from somebody!

I have been posting since yesterday that the proposed solution only works
under
*nix because of the sequence of of events that have to occur (see below for
even
more details).  I'm sorry, I was just feeling frustrated that the same
suggestions kept coming up, even though I thought I was explaining why they
won't work.  An additional problem is that half of the commands used in the
*nix
solution don't even exist under NT (and the most important one, sleep,
doesn't).

>
> Second: you declare "won't work" without any explanation why the
suggested
> solution won't work. That doesn't make any of us any wiser. I don't see
why
> this won't work. Please state your arguments, so we can learn from your
> expertize (which i do not challenge!).

Here is the sequence of events (simplified as I understand it) under a *nix
system:

1) user requests a file be committed (each file is handled individually
even if
the user tries to commit an entire directory).

2) CVS
     a) locks the file in the repository
     b) performs the necessary operations (updates the repository)
     c) checks to see if there are any loginfo events to fire

3) CVS tries to fire the loginfo event
     a) the loginfo even spawns a background shell process
     b) exits with no error

4) CVS
     a) unlocks the file
     b) exits with no error

5) The background shell sleeps for 2 seconds and then
     a) cd's to the sandbox
     b) runs a new CVS process to do the update
     c) exits with no error

The sandbox is now updated.

Under NT, step 3a is not possible (not really) and there is no way to use
START
to create a new process that waits for the first CVS to exit before
starting the
second.  The problem is that the loginfo line must exit with no error
*before*
the CVS commit process can unlock the and exit itself.  The inner CVS must
wait
until the outer CVS exits before it can sucessfully update the sandbox.
The
important sequence is that the repository copy of the file is locked in
step 2a
but not unlocked until 4a; it is not possible to do anything between those
two
points that involves accessing that file in the repository.

I know, I have tried everything I could think of, but I could not get
around
this.  My solution (based on suggestions that others gave me) was to use
the NT
AT program to schedule a new CVS instance as my step 3a.  Then this process
fires (something less than 1 minute later due to the limitations of the
Scheduler) and performs steps 5a-c.

HTH

John

--
John Peacock
Director of Information Research and Technology
Rowman & Littlefield Publishing Group
4720 Boston Way
Lanham, MD 20706
301-459-3366 x.5010
fax 301-429-5747

_______________________________________________
Cvsnt mailing list
Cvsnt at cvsnt.org
http://www.cvsnt.org/cgi-bin/mailman/listinfo/cvsnt https://www.march-hare.com/cvspro/en.asp#downcvs

_______________________________________________
Cvsnt mailing list
Cvsnt at cvsnt.org
http://www.cvsnt.org/cgi-bin/mailman/listinfo/cvsnt https://www.march-hare.com/cvspro/en.asp#downcvs