Re: Tag locking change

[Paul Sander]

>--- Forwarded mail from address@hidden

>[ On Sunday, October 6, 2002 at 18:01:01 (-0400), Larry Jones wrote: ]
>> Subject: Tag locking change
>>
>> I've checked in a change to the [r]tag command to go back to locking
>> one directory at a time (correctly, this time) rather than locking the
>> entire tree for the whole time.

>Is this really a good idea?  Do people who start a 'cvs co -r' or some
>other command using the new tag too soon before an [r]tag is finished
>deserve to lose (assuming by some strange quirk of concurrency that
>their command catches up to the tag)?

Another serious issue is when someone commits while an rtag is in
progress, and the new data are erroneously tagged.

You could get away without using two-phase locking if you can identify
the applicable version of each file in advance.  This is easily done
by the "cvs tag" command, but not with "cvs rtag".  To make it work
with "rtag" you must rely on a branch/timestamp pair or an existing
version tag.  The user must specifiy the branch or version tag, and
the timestamp can be either given on the command line or assumed from
time of day clock at the moment the command is initiated.

But this method still suffers the same problem that Greg points out.
Using two-phase locking is way better from a correctness perspective.

>--- End of forwarded message from address@hidden