[PROPOSAL] Move sleeping for timestamps to libsvn_wc

Working on the sleep_for_timestamps stuff, Peter Lundblad and I found
out that currently all sleep operations are built into libsvn_client.

I think this means they are misplaced: we have the sleep_for_timestamp
calls to increase the chances for the change-detection-heuristic which
is built into libsvn_wc. And solely for that purpose.

I think for that reason, libsvn_wc is therefore the best part of the
code to decide whether we should sleep or not.

Imagine an application built on top of libsvn_wc (but not
libsvn_client) which would in rapid succession update, modify and
commit working copies. An application like that would entirely legally
be using our APIs, but not get the same guarantees we'd want them to
be given.

So what now?

Well, in lock.c in libsvn_wc, we have a number of routines which could
sleep for us, for example when closing an adm_access or when dropping
a write-lock. I'd like to build sleeping into those routines and drop
it from libsvn_client.

bye,

Erik.

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Mon Mar 12 22:03:29 2007