Makefile.in | 8 3 + 5 - 0 !
build.conf | 6 3 + 3 - 0 !
2 files changed, 6 insertions(+), 8 deletions(-)

 build-fixes

Fix some build problems:
Fix some build problems:

- Add aprutil as a direct dependency for libsvn_client and libsvn_ra.

subversion/libsvn_subr/subst.c | 8 6 + 2 - 0 !
1 file changed, 6 insertions(+), 2 deletions(-)

 last-changed-date-charset

Bug #290774: Convert $LastChangedDate$ keyword to the local character
Bug #290774: Convert $LastChangedDate$ keyword to the local character
set, since it is already rendered in the local language.

Not accepted upstream: http://svn.haxx.se/dev/archive-2006-04/0730.shtml

subversion/libsvn_repos/repos.c | 28 18 + 10 - 0 !
1 file changed, 18 insertions(+), 10 deletions(-)

 repos-templates

Bug #210901: Add fully qualified paths to commands in example hook
Bug #210901: Add fully qualified paths to commands in example hook
scripts that get installed as .tmpl files in new repositories.  Use
$REPOS/hooks/ as the path, to encourage repository admins to copy the
hooks, not just use them in place.

build.conf | 4 2 + 2 - 0 !
build/generator/gen_base.py | 3 2 + 1 - 0 !
2 files changed, 4 insertions(+), 3 deletions(-)

 rpath

Prevent the linker from adding an rpath to shared libraries.  Original
patch by David Kimdon <dwhedon@debian.org>.  The basic theory is:

- Use libtool instead of apxs to install the apache modules.
  libtool relinks without rpath in this case, apxs obviously doesn't.

Makefile.in | 2 1 + 1 - 0 !
1 file changed, 1 insertion(+), 1 deletion(-)

 ruby-no-strict-aliasing

It seems swig is generating _really_ bad code here, or something.
Whatever it is, this line suppresses 1323 warnings, any of which might
be real bugs with real consequences.

subversion/bindings/swig/ruby/test/util.rb | 28 23 + 5 - 0 !
1 file changed, 23 insertions(+), 5 deletions(-)

 ruby-test-svnserve-race

Bug #378837: Ruby testsuite: wait for svnserve to start before
Bug #378837: Ruby testsuite: wait for svnserve to start before
connecting to it.  This mainly affects very slow machines - observed
on various arm and m68k builds.

Thanks to Roman Zippel, Kobayashi Noritada, Wouter Verhelst and Martin
Michlmayr.

subversion/libsvn_ra_svn/client.c | 2 1 + 1 - 0 !
subversion/libsvn_subr/config_file.c | 2 1 + 1 - 0 !
2 files changed, 2 insertions(+), 2 deletions(-)

 ssh-no-controlmaster

Bug #413102: Avoid accidentally becoming an ssh ControlMaster.
Bug #413102: Avoid accidentally becoming an ssh ControlMaster.

If you set ControlMaster via your ssh config:
.ssh/config:
Host *
    ControlMaster auto
    Controlpath ~/.ssh/master-%r@%h:%p

you can get 'svn: Can't write to connection: Broken pipe' with certain
ssh commands, because svn can open multiple ssh sessions and kill them
at unpredictable times.

I don't really think the problem is either svn's or openssh's fault,
but this is the obvious way to avoid it.  We will still use ssh
connection sharing, but only if a ControlMaster is already present.

tools/examples/svnshell.py | 8 1 + 7 - 0 !
1 file changed, 1 insertion(+), 7 deletions(-)

 svnshell

Bug #234462: The error messages that svnshell spits out for unknown
Bug #234462: The error messages that svnshell spits out for unknown
commands are confusing. They should be replaced by a "Unknown command."
error message.  Upstream says we have no sense of humor.  (:

Patch from Roland Illig <roland.illig@gmx.de>.

configure.ac | 2 1 + 1 - 0 !
1 file changed, 1 insertion(+), 1 deletion(-)

 apr-abi

Bug #266170: Use SOVERSION 1 for apr 1.x, because the apr 0.9.x -> 1.x
Bug #266170: Use SOVERSION 1 for apr 1.x, because the apr 0.9.x -> 1.x
ABI change leaks through the libsvn ABIs (because we use apr_off_t,
which is now 64 bits on all platforms).

Upstream refused this (http://svn.haxx.se/dev/archive-2006-03/0522.shtml
for thread), but fortunately when it came time to support apr 2.x in svn
1.7, they skipped SOVERSION 1 in consideration of Debian/Ubuntu, so we'll
be ABI-compatible again if and when we move to apr-2.x.

Makefile.in | 1 1 + 0 - 0 !
build.conf | 18 9 + 9 - 0 !
build/ac-macros/swig.m4 | 4 3 + 1 - 0 !
3 files changed, 13 insertions(+), 10 deletions(-)

 python-builds

Build for multiple python versions.  Patch from Matthias Klose
<doko@debian.org> via Ubuntu.  Mostly, rename libsvn_swig_py to
libsvn_swig_py2.x.

Submitted upstream: http://svn.haxx.se/dev/archive-2008-05/0361.shtml

tools/examples/getlocks_test.c | 8 3 + 5 - 0 !
tools/examples/headrev.c | 8 3 + 5 - 0 !
tools/examples/minimal_client.c | 6 2 + 4 - 0 !
tools/examples/svnput.c | 7 2 + 5 - 0 !
tools/examples/testwrite.c | 8 3 + 5 - 0 !
5 files changed, 13 insertions(+), 24 deletions(-)

 examples-compile-instructions

Fix build instructions for libsvn-doc/examples/*.c.

subversion/bindings/swig/ruby/test/util.rb | 2 2 + 0 - 0 !
1 file changed, 2 insertions(+)

 workaround_einval_on_kfreebsd

build/ac-macros/java.m4 | 2 1 + 1 - 0 !
build/generator/templates/build-outputs.mk.ezt | 12 4 + 8 - 0 !
subversion/bindings/javahl/src/org/apache/subversion/javahl/CommitItemStateFlags.java | 9 9 + 0 - 0 !
subversion/bindings/javahl/src/org/apache/subversion/javahl/callback/UserPasswordCallback.java | 5 5 + 0 - 0 !
subversion/bindings/javahl/src/org/apache/subversion/javahl/types/Revision.java | 2 2 + 0 - 0 !
5 files changed, 21 insertions(+), 9 deletions(-)

 java10-compatibility

subversion/libsvn_ra_svn/client.c | 17 15 + 2 - 0 !
subversion/svnserve/serve.c | 17 15 + 2 - 0 !
subversion/tests/libsvn_ra/ra-test.c | 89 89 + 0 - 0 !
3 files changed, 119 insertions(+), 4 deletions(-)

 provide a way for svnserve's 'get-deleted-rev' api to return 'not
 deleted'.

Previously the answer 'not deleted' was indistinguishable from an error when
using the svnserve protocol.

The standard 'svn' client software does not appear to invoke this case, but
it is meant to be a valid answer and is supported by the other repository
access protocols.

To work around the problem without changing the protocol, we re-purpose the
obsolete error code 'SVN_ERR_ENTRY_MISSING_REVISION' to communicate this
'not deleted' reply to the client.

(Re-purposing an old error code was not essential. An alternative would be
to create a new error code which would become part of the public API in the
next minor release, but would be deemed non-public when backported to older
release lines.)

Behaviour changes:

  - With a new client against a new server, such queries are now handled
    correctly.

  - With an old client against a new server, the client will report a more
    informative error message.

  - With a new client against an old server, there is no improvement.

* subversion/svnserve/serve.c
  (get_deleted_rev): If the answer is SVN_INVALID_REVNUM, return a
    SVN_ERR_ENTRY_MISSING_REVISION error to the client.

* subversion/libsvn_ra_svn/client.c
  (ra_svn_get_deleted_rev): Convert the error SVN_ERR_ENTRY_MISSING_REVISION
    back to a response of SVN_INVALID_REVNUM.

* subversion/tests/libsvn_ra/ra-test.c
  (commit_two_changes): New.
  (test_get_deleted_rev_no_delete,
   test_get_deleted_rev_errors): New tests.
  (test_funcs): Run them.

Signed-off-by: James McCoy <jamessan@debian.org>

subversion/tests/libsvn_ra/ra-test.c | 13 10 + 3 - 0 !
1 file changed, 10 insertions(+), 3 deletions(-)

 adjust expectations of a new test to account for differences between
 RA layers. A follow-up to r1860936.

* subversion/tests/libsvn_ra/ra-test.c
  (test_get_deleted_rev_errors): Expect a generic error code when running
    over HTTP, otherwise the specific error code for this error case.

Signed-off-by: James McCoy <jamessan@debian.org>

subversion/svnserve/serve.c | 13 7 + 6 - 0 !
1 file changed, 7 insertions(+), 6 deletions(-)

 in svnserve, consistently handle errors in opening a repository.

These errors are still logged and reported to the client, as they were
before, but now it is done in the same way as everywhere else. (The error
logging now happens higher up the call stack.)

* subversion/svnserve/serve.c
  After reporting an error to the client, don't log it explicity here and
  then clear it; instead return it so the caller can do so.

Signed-off-by: James McCoy <jamessan@debian.org>

subversion/libsvn_repos/config_file.c | 4 4 + 0 - 0 !
1 file changed, 4 insertions(+)

 fix a potential null dereference in the config file parser.

* subversion/libsvn_repos/config_file.c
  (get_repos_config): svn_repos_find_root_path() may return NULL.
   Check the return value accordingly.

subversion/libsvn_repos/log.c | 26 10 + 16 - 0 !
subversion/tests/cmdline/authz_tests.py | 55 55 + 0 - 0 !
2 files changed, 65 insertions(+), 16 deletions(-)

 don't show unreadable copyfrom paths in 'svn log -v'

* subversion/libsvn_repos/log.c:
  (detect_changed): In the add and replace cases, explicitly clear the
   svn_fs_path_change3_t::copyfrom_path pointer and rev when said path@rev is
   unreadable.

* subversion/tests/cmdline/authz_tests.py:
  (log_inaccessible_copyfrom): New regression test.
  (test_list): Run it.

Patch by: kotkov

git-svn-id: https://svn.apache.org/repos/asf/subversion/trunk@1899227 13f79535-47bb-0310-9956-ffa450edef68
Signed-off-by: James McCoy <jamessan@debian.org>

subversion/libsvn_repos/authz.c | 25 25 + 0 - 0 !
1 file changed, 25 insertions(+)

 fix issue #4880 "use-after-free of object-pools when used as httpd
 module"
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 8bit

Ensure that we initialize authz again if the pool which our authz
caches depend on is cleared. Apache HTTPD may run pre/post config
hooks multiple times and clear its global configuration pool which
our authz caching pools depend on.

Reported-by: Thomas Weischuh (thomas {at} t-8ch dot de)

Thomas has also confirmed that this patch fixes the problem.

* subversion/libsvn_repos/authz.c
  (deinit_authz): New pool cleanup handler which resets authz initialization
   in case the parent pool of our authz caches is cleared.
  (synchronized_authz_initialize): Register new pool cleanup handler.

git-svn-id: https://svn.apache.org/repos/asf/subversion/trunk@1894734 13f79535-47bb-0310-9956-ffa450edef68
Signed-off-by: James McCoy <jamessan@debian.org>