Dave Brown <dave.brown@wandisco.com>

r880877, r880459, r879492, r878648, r878647

r878647 | markphip | 2009-08-05 19:34:29 +0000 (Wed, 05 Aug 2009)

JavaHL: Improve the existing revprop test to actually check that
the property is set.

[ in subversion/bindings/javahl/ ]

* tests/org/tigris/subversion/javahl/SVNAdminTests.java:
   (testSetRevProp): Improved test, made it actually verify the result.

Patch by: Dave Brown <dave.brown@wandisco.com>

r878648 | markphip | 2009-08-05 20:05:26 +0000 (Wed, 05 Aug 2009)

JavaHL: Fix issue #2979: javahl's SVNAdmin::load() sometimes passes a null
data buffer to the read() method of the caller's InputInterface.
JNIUtil::makeByteArray was returning NULL when the requested byte[] length
was 0.  This in turn was provoked by a read(length=0) on the svn_stream_t
loading a repository dump.  read(0) is perfectly valid (albeit superfluous),
such as when parsing a property value with length zero.

The fix just returns a byte[] with length 0 as requested from JNIUtil.cpp.

[ in subversion/bindings/javahl/ ]

* tests/org/tigris/subversion/javahl/SVNAdminTests.java:
  (testLoadRepo): Added test for issue 2979

* tests/org/tigris/subversion/javahl/SVNTests.java
  (OneTest): Tweaked inner class optionally not to load sample files
   into the test repository.  Keep behavior expected by existing tests.

* tests/data/issue2979.dump
  Data for reproducible test case of issue #2979.

* native/JNIUtil.cpp
  (makeJByteArray) Do not treat byte[0] and NULL as equivalent.

Patch by: Dave Brown <dave.brown@wandisco.com>
Reviewed by: hwright

r879492 | stsp | 2009-09-18 08:56:38 +0000 (Fri, 18 Sep 2009)

Raise an error in loggy_path() when given paths that are not in the WC,
rather than letting the code segfault.  (Absolute paths in /tmp seem to be
coming from merge operations).  Callers of loggy_path() are not prepared
to handle a NULL return anyway.  Even though this stuff is going away,
it is crashing midway through the javahl tests.  With this change,
the javahl tests complete, getting 1 error and 2 failures.

* subversion/libsvn_wc/log.c
    (loggy_path): Return an error if PATH is not a child of ADM_ABSPATH.
     The new behaviour is consistent with loggy_paths()'s docstring,
     which says that "PATH must not be outside that directory."

Patch by: Dave Brown <dave.brown@wandisco.com>
(patch and log message slightly tweaked by me)

r880459 | julianfoad | 2009-11-05 15:48:47 +0000 (Thu, 05 Nov 2009)

Improve the error condition encountered when trying to use a DAV
write-through proxy but not having mod_proxy loaded. See this email thread:

In mod_dav_svn, when processing an SVNMasterURI directive, check that
mod_proxy is available in the httpd runtime.  Forwarding writes to a master
from a slave requires the mod_proxy handler, and when it isn't present, the
failure is ugly & opaque.  Apache's core, default handler sends back a "405
Not Allowed," for non-GET which looks like an authz failure.

BTW, there's precedent for using ap_find_linked_module() to check that
module dependencies are present.  Namely, mod_rewrite looks for mod_proxy,
and mod_authnz_ldap looks for util_ldap.

* subversion/mod_dav_svn/mod_dav_svn.c
  (SVNMasterURI_cmd): Use ap_find_linked_module() to ensure that mod_proxy
    is available.

Patch by: Dave Brown <dave.brown{_AT_}wandisco.com>

r880877 | julianfoad | 2009-11-16 17:57:37 +0000 (Mon, 16 Nov 2009)

Update the BuildBot Centos slave configuration to test over RA Neon, and to
find Jave and SQLite dependencies better, and disable the Ruby tests as they
currently seg-fault on this platform.

Patch by: Dave Brown <dave.brown{_AT_}wandisco.com>

* tools/buildbot/slaves/centos/svnbuild.sh
  Update the configure script --with arguments for Java and SQLite.

* tools/buildbot/slaves/centos/svncheck-bindings.sh
  Comment out the Ruby test.

* tools/buildbot/slaves/centos/svncheck.sh
  Switch from RA-local ("make check") to RA-Neon ("make davautocheck").