Explaining switch_tests.py 19 failure over ra-serf (as seen on 1.6.x, and present but not typically seen on trunk)

I've been looking into switch_test.py 19 failures tonight. For me, it fails
on trunk using ra_neon, but passes using ra_serf; on the 1.6.x branch it
fails using ra_serf, but passes using ra_neon. At the moment my attention
is one the 1.6.x branch.

I've stared at this thing for hours, and finally (at 4:30am in my current
timezone) I see something that looks fishy. Note this request/response pair:

<<<<<<<<<
PROPFIND /svn-test-work/repositories/switch_tests-19.other/A/D HTTP/1.1
Transfer-Encoding: chunked
Label: 1
Depth: 0
Accept-Encoding: gzip
Authorization: Basic anJhbmRvbTpyYXlqYW5kb20=
DAV:
http://subversion.tigris.org/xmlns/dav/svn/depth,http://subversion.tigris.org/xmlns/dav/svn/mergeinfo,http://subversion.tigris.org/xmlns/dav/svn/log-revprops

Content-Type: text/xml
User-Agent: SVN/1.6.7 (dev build) serf/0.3.0
Host: localhost

<?xml version="1.0" encoding="utf-8"?><propfind
xmlns="DAV:"><prop><checked-in xmlns="DAV:"/></prop></propfind>
 ---------
HTTP/1.1 207 Multi-Status
Date: Thu, 29 Oct 2009 03:22:17 GMT
Server: Apache/2.2.11 (Unix) DAV/2 mod_python/3.3.2-dev-20080819
Python/2.6.2 mod_wsgi/3.0c5 SVN/1.6.7-dev
Vary: Label
Content-Length: 470
Content-Type: text/xml; charset="utf-8"

<?xml version="1.0" encoding="utf-8"?>
<D:multistatus xmlns:D="DAV:" xmlns:ns0="DAV:">
<D:response xmlns:lp1="DAV:"
xmlns:lp3="http://subversion.tigris.org/xmlns/dav/">
<D:href>/svn-test-work/repositories/switch_tests-19.other/A/D/</D:href>
<D:propstat>
<D:prop>
<lp1:checked-in><D:href>/svn-test-work/repositories/switch_tests-19.other/!svn/ver/2/A/D</D:href></lp1:checked-in>
</D:prop>
<D:status>HTTP/1.1 200 OK</D:status>
</D:propstat>
</D:response>
</D:multistatus>
>>>>>>>>>

ra_serf make a request for information about a path with the "Label: 1"
header (that is, version 1 of that thing). But we return information that
claims that the checked-in URL for that item is ".../ver/2/..." My guess
was that the mod_dav_svn codebase seems to indicate that the Label header is
ignored altogether. Sure enough, if you look at mod_dav_svn/repos.c, there
are comments in a whole bunch of URI-parsing functions which read:

  /* ### what to do with LABEL and USE_CHECKED_IN ?? */

How did this ra_serf code ever work? More importantly, how do we fix it?

This problem probably exists on trunk, too, but the test suite doesn't catch
it without extra effort because HTTPv2 changes the way ra_serf queries a
trunk-era server altogether. My guess is that the bug shows up here when
talking to a pre-1.7 server (or a 1.7 server that has HTTP v2 disabled).
I'm too exhausted to test that right now.

Note that ra_neon doesn't have this problem because it takes a different
route (with more PROPFINDs that don't try to use "Label:") to ask the same
question of the server. Maybe that's the fix for ra_serf, too -- using
ra_neon-like query semantics in this case?

-- 
C. Michael Pilato <cmpilato_at_collab.net>
CollabNet   <>   www.collab.net   <>   Distributed Development On Demand
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=2412375