RE: property performance question

This is very helpful. Thank you.
About that rather silly last question, first: doh! I really meant exposing
property filtering to the javahl interface, which is what I'm using, though
I could just as easily implement your suggestion.
I think I will try storing children properties both as a namespaced list in
the parent dir-props, and as an actual file (either in xml or binary java
serialization format) in the parent directory (to take advantage of
deltafication) and compare throughput measurements.
Thanks again.

-----Original Message-----
From: Ben Collins-Sussman [mailto:sussman@collab.net]
Sent: Thursday, June 24, 2004 5:36 AM
To: Reid Hartenbower
Cc: users@subversion.tigris.org
Subject: Re: property performance question

On Wed, 2004-06-23 at 12:30, Reid Hartenbower wrote:

> Would it be significantly more efficient (in terms of network traffic
> minimization/ speed of child property change propagation) for me to just
have
> this structure

I imagine it would be more efficient, if only because libsvn_wc only
needs to read/write a single hash table, rather than N of them.

> Before I actually code the change and do comparison tests, I wondered if
any
> of you had advice. Is there a limit to the size of property files like
> dir-props?

Property files are just hash tables on disk. There's no limit to key or
value size.

> Are changes to property files sent as deltas during communication
> with the subversion server?

No, not at all. Only file contents are deltified over the network.

> Any chance of adding a filter argument to
> 'proplist' to help locate these 'flattened' properties more quickly, eg:
> svn proplist --filter child1.* parent

Um, pipe 'svn proplist' into 'grep'? :-)

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Thu Jun 24 19:41:25 2004