[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: wcprops and wc-ng

From: Greg Stein <gstein_at_gmail.com>
Date: Thu, 19 Mar 2009 10:20:39 +0100

On Thu, Mar 19, 2009 at 06:19, C. Michael Pilato <cmpilato_at_collab.net> wrote:
> Hyrum K. Wright wrote:
>> Greg,
>> It appears that the currently failing tests on trunk over ra_neon are
>> due to various issues running the wcprops log.  It's the same set of
>> tests which fails when do_sync is removed from svn_wc__entry_modify(),
>> and I think it has to do with the fact that read_entries() now returns
>> a unique hash from prune_deleted().
>>
>> I'm not sure what the Grand Plan for wcprops in wc-ng is.
>> Unfortunately, we still need to store them for compat reasons with the
>> older http protocol.  Are they just part of the properties hash in
>> BASE, or should there be another column for wcprops?  (Thankfully,
>> unlike "real" properties, wcprops only have a BASE version, not
>> WORKING or ACTUAL.)
>>
>> I'd like to get these wcprops issues ironed our fairly quickly.  I
>> think that we can implement simple wc-ng prop handling for wcprops,
>> switch over to that, and fix the test failures on trunk.  This should
>> be relatively straight-forward *and* give us some context for use in
>> props.c.  Thoughts?
>
> Not sure if this helps you to decide or not, but the current WC
> implementation has two different property stores, one for regular versioned
> properties (which has both BASE and WORKING), one for wcprops (which has ...
> neither, really).  The wc-props abstraction isn't strictly required going
> forward, but when talking to pre-HTTP-v2 servers, the absence of wc-props
> has the potential to vastly slow things down.

Yeah. We want to keep them. But it sounds like some kind of Bad
Interaction is happening to create these bugs right now. Sigh.

> (There's a third concept called "entry props", but these find their storage
> as fields in the svn_wc_entry_t.)

The entry props are the Last-Changed properties, right? In wc-ng,
their naming precedent is: changed_rev, changed_date, and
changed_author. The changed_rev value is a MUST, and changed_date and
changed_author are MAY be present (somebody could remove the revprop
that sources those properties).

Ah. From svn_props.h, I also see that the UUID and lock tokens are
passed as "entry" properties. Those also have distinct fields/tables.

Cheers,
-g

------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=1354359
Received on 2009-03-19 10:20:55 CET

This is an archived mail posted to the Subversion Dev mailing list.