James McCoy <james...@debian.org> writes:
> As an independent package or as part of the Ruby package?
The minitest available via yum is rubygem-minitest 4.3.2. I suppose
I could install a separate gem.
--
Philip Martin
WANdisco
st gem (of the 5.x series) is bundled. These
> changes broke check-swig-rb, since the test framework APIs behave
> differently.
This patch works on my Debian/stable system but not on CentOS 7 which
only has minitest 4.3.2.
--
Philip Martin
WANdisco
Branko Čibej <br...@apache.org> writes:
> On 16.10.2015 13:14, Philip Martin wrote:
>> Philip Martin <philip.mar...@wandisco.com> writes:
>>
>>> "Bert Huijben" <b...@qqmail.nl> writes:
>>>
>>>> I'm not able to answer all tha
the libsvn_client part, leaving the JNI part. See
r1710290.
--
Philip Martin
WANdisco
es.Version.isAtLeast(III)Z+1
j org.apache.subversion.javahl.NativeResources.init()V+17
j org.apache.subversion.javahl.NativeResources.loadNativeLibrary()V+71
j org.apache.subversion.javahl.types.Version.()V+0
--
Philip Martin
WANdisco
ry and that is not always true. For example pack on
a FSFS repository could remove revision files, a commit to a BDB
repository could delete a log file. Any such files will cause archive
creation to fail with your patch.
--
Philip Martin
WANdisco
y 1.8.15-dev (under development)
--
Philip Martin
WANdisco
Philip Martin <philip.mar...@wandisco.com> writes:
> "Bert Huijben" <b...@qqmail.nl> writes:
>
>> I'm not able to answer all that, but I do know that your change will
>> slow Subversion down on Samba shares as used from Windows systems with
>> th
time down to 69s, so only 5% slower
than 1.8 rather than 36%.
--
Philip Martin
WANdisco
ange r1659426 stated:
The TRUNCATE and DELETE journal modes are compatible, so different
Subversion clients with different journal mode should just work with same
working copy without problems.
--
Philip Martin
WANdisco
than a
local filesystem that uses LVM+mdadm+ZFS/BtrFS? Is NFS less reliable
than the firmware in a h/w RAID card?
--
Philip Martin
WANdisco
), but that obviously means
we better be *damned* sure it won't lose or corrupt the repo :-)
--
Philip Martin
WANdisco
failure(r);
- return HTTP_UNAUTHORIZED;
-}
+return DECLINED;
}
#else
if (!authn_required)
--
Philip Martin
WANdisco
src_lock = fs.get_lock(src_fs, dst_lock.path)
if not src_lock:
fs.unlock(dst_fs, dst_lock.path, dst_lock.token, True)
print "rm: " + dst_lock.path
access = fs.create_access("dummy")
fs.set_access(dst_fs, access)
fs.get_locks(dst_fs, "", rm_locks)
--
Philip Martin
WANdisco
Expires:
Comment (0 lines):
and I've just discovered that it is even possible to lock directories!
http://subversion.tigris.org/issues/show_bug.cgi?id=2507#desc28
--
Philip Martin
WANdisco
+OIkBVB50oV6eOZJ7JHhy9rJ5D3W9MaH+tiV3gBe
u5cFj6T4KUnj8gyNgpvzyBXWbCDhSHOkLdwvsg1PNjtLDkGiHkflpRKiMEJL6Umv
x8QNJGbmzC8hKhRtINK3mmUEBDTV3q9jUOcLmuqn7fHJ31RKl+Z73+H0tPNq5vM=
=irhF
-END PGP SIGNATURE-
--
Philip Martin
WANdisco
Detect the error from a broken RA
session and create another? Track the time when the session was last
used? Something else?
--
Philip Martin
WANdisco
nt - sum, ',',
pool),
249 (int)((stats->file_histogram.total.count - sum) * 100 /
250stats->file_histogram.total.count));
251 }
252
253 /* Print the (up to) 16 extensions in STATS with the largest total size
of
(gdb) p stats->file_histogram.total.count
$1 = 0
--
Philip Martin
WANdisco
messages from the command? Should they be dropped or
displayed? What about authentication prompts? Is --non-interactive
needed?
--
Philip Martin
WANdisco
is unfinished message and it is true
that support for 1.7 is not complete, e.g. no support for SQLite
indices. The script was an experiment that was never finished.
--
Philip Martin
WANdisco
is unfinished and not ready to be used on live data.
Trust us.
I don't really see how we could make it clearer.
--
Philip Martin
WANdisco
Stefan Sperling s...@elego.de writes:
On Mon, Aug 17, 2015 at 11:03:12AM +0100, Philip Martin wrote:
The script was an experiment that was never finished.
I was wrong, the script does handle SQLite indices as it copies the
entire schema. The main problem is we don't really know how well
descriptor for this process:
$ ls -l /proc/28574/fd/6
lr-x-- 1 pm pm 64 Aug 14 10:18 /proc/28574/fd/6 - pipe:[612259]
--
Philip Martin
WANdisco
Philip Martin philip.mar...@wandisco.com writes:
That's the apr_poll() call in data_available_handler_apr() failing, and
E09 could be EBADF.
Perhaps E09 is not EBADF on Windows? Looking on Google I find
Error Code 9: The storage control block address is invalid.
[ERROR_INVALID_BLOCK
.
--
Philip Martin
WANdisco
Stefan Fuhrmann stefan.fuhrm...@wandisco.com writes:
What does the version check look like on Windows?
Good point, I don't know how to handle that.
--
Philip Martin
WANdisco
Philip Martin philip.mar...@wandisco.com writes:
Stefan Fuhrmann stefan.fuhrm...@wandisco.com writes:
What does the version check look like on Windows?
Good point, I don't know how to handle that.
I suppose I have to put back the MMN checks in mod_authz_svn.c, retain
the new configure
Philip Martin philip.mar...@wandisco.com writes:
Writing autoconf code to run grep on the httpd header file might be an
option.
I worked up a patch. This has the added advantage the the detection now
happens earlier so broken httpd causes a configure error rather than a
compile error. It has
Philip Martin philip.mar...@wandisco.com writes:
I suppose I have to put back the MMN checks in mod_authz_svn.c, retain
the new configure checks, and have configure set some #define that
overides the checks in mod_authz_svn.c for httpd patched without MMN
bump. If I arrange it so
autoconf code to run grep on the httpd header file might be an
option.
We could add a configure option to use of the new API even when the MMN
has not been updated, the opposite of --enable-broken-httpd-auth.
Or we could leave users to edit the Subversion source code.
--
Philip Martin
WANdisco
to grep.
--
Philip Martin
WANdisco
/mO2J6mvjlgFE1xci7iycgkgUcBEUxEuVJVhrGGydDnJ35J0ORMM=
=JcMI
-END PGP SIGNATURE-
--
Philip Martin
WANdisco
-END PGP SIGNATURE-
--
Philip Martin
WANdisco
Philip Martin philip.mar...@wandisco.com writes:
svnsync has a simple algorithm that writes every revprop for each
revision.
If we put the write optimisation into svnsync, not writing unchanged
revprops, this would probably give most of the efficiency gain for the
case where the bulk
Philip Martin philip.mar...@wandisco.com writes:
RHEL 6 is on Python 2.6 and is unlikely to change. Python 2.7 is
available as a Red Hat Software Collection component.
The Software Collection python27 is sufficient to build Subversion and
run the regression tests but some of the tests FAIL
/jessie/stretch - currently
oldstable/stable/testing): py2.7
The very old enterprise versions are probably RHEL 6 and SuSE 10; both
are still in wide use.
RHEL 6 is on Python 2.6 and is unlikely to change. Python 2.7 is
available as a Red Hat Software Collection component.
--
Philip Martin
Philip Martin philip.mar...@wandisco.com writes:
svnadmin dump-revprops repo | svnadmin load-revprops repo
Oops! Load into a different repository:
svnadmin dump-revprops repo | svnadmin load-revprops repo2
--
Philip Martin
WANdisco
:
src_value = src_props[dst_name]
except:
fs.change_rev_prop(dst_fs, r, dst_name, None) # delete
--
Philip Martin
WANdisco
Philip Martin philip.mar...@wandisco.com writes:
Philip Martin philip.mar...@wandisco.com writes:
RHEL 6 is on Python 2.6 and is unlikely to change. Python 2.7 is
available as a Red Hat Software Collection component.
The Software Collection python27 is sufficient to build Subversion
to STMT_INSERT_DELETE_FROM_NODE_RECURSIVE is sufficient to PASS the
regression tests when reverse is enabled. Reverting r1687152 is the
most reliable way to avoid a regression but fixing
STMT_INSERT_DELETE_FROM_NODE_RECURSIVE is probably sufficient.
--
Philip Martin
WANdisco
for
STMT_INSERT_DELETE_LIST to be ordered. All that is necessary is that
the same rows are selected (so that the correct notifications are
generated).
--
Philip Martin
WANdisco
changed to not destroy the svnmover_wc_t pool.
--
Philip Martin
WANdisco
space for consistency with the rest of
the regex. I suppose we could switch [:space:] everwhere but I don't
think that is necessary. Comitted as r1689824. Thanks!
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
=
=Faco
-END PGP SIGNATURE-
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
violation of the
DAV rules. There are some definitions in mod_dav.h but callers can
pass other values.
- 'aprerr' is any underlying APR error, i.e. some sort of runtime
system error.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
://subversion.tigris.org/issues/show_bug.cgi?id=3750
It can lead to inconsistent locks:
- a lock that shows up on a file /A/f but doesn't show up when one
asks for all the locks in /A.
- a lock for a file that does not exist in HEAD
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
succeeding. I suppose we could
change svn_repos_notify_t.err to some other type, but svn_error_t is a
convenient way to package the information.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
Philip Martin philip.mar...@wandisco.com writes:
1.9 has a more accurate svn_fs_contents_changed() that doesn't report as
many false positives. This means that some (or all?) of the -g revisions
reported by svn_repos_get_file_revs2() do not include a textdelta that
was included by 1.8
Philip Martin philip.mar...@wandisco.com writes:
Philip Martin philip.mar...@wandisco.com writes:
1.9 has a more accurate svn_fs_contents_changed() that doesn't report as
many false positives. This means that some (or all?) of the -g revisions
reported by svn_repos_get_file_revs2() do
Bert Huijben b...@qqmail.nl writes:
-Original Message-
From: Philip Martin [mailto:philip.mar...@wandisco.com]
Sent: woensdag 10 juni 2015 14:05
To: Andreas Stieger
Cc: dev@subversion.apache.org
Subject: Re: [patch] Support modern network utilities for finding free
ports for
tests
Philip Martin philip.mar...@wandisco.com writes:
HTTPD_PORT=3691
-while netstat -an | grep $HTTPD_PORT | grep 'LISTEN' /dev/null; do
+while \
+ (ss -ltn sport = :$HTTP_PORT 21 | grep :$HTTP_PORT /dev/null ) \
+ || \
+ (netstat -an 21 | grep $HTTP_PORT | grep 'LISTEN' /dev/null
) \
+ do
You have HTTP_PORT where it should be HTTPD_PORT. I fixed that and
committed r1684649. Thanks!
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
Build 1.8 and 1.9 (I'm building 1684452). Run a 1.9 server pointing at
the 1.8 build directory; the server can be httpd or svnserve, but
svnserve is probably easiest:
../obj-1.9/subversion/svnserve/svnserve -Tdr
../obj-1.8/subversion/tests/cmdline
then run blame_tests.py 10:
cd
Philip Martin phi...@codematters.co.uk writes:
$ ../../svn/svn blame -g
svn-test-work/working_copies/blame_tests-10/trunk/iota
2jrandom This is the file 'iota'.
2jrandom 'A' has changed a bit, with 'upsilon', and 'xi'.
while the 1.9 client gives the correct output
Philip Martin philip.mar...@wandisco.com writes:
The 1.9 server is sending the same revisions as the 1.8 server but some
of the revisions from the 1.9 server do not have a textdelta. Both the
1.8 and 1.9 servers call svn_repos_get_file_revs2() which calls
send_path_revision(). Inside
=svn:authorpm/S:rev-prop
/S:file-rev
/S:file-revs-report
The significant difference is the missing txdeltas (ignore the hash
ordering of revprops).
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
7sYgyVuR3W6aJThFLPdToHT0D11OLezOPHTvahZJubM0EXPNnh7WK9WU9pE4wa80
pj1nOXSItmyoBYKzcNEfQQJE0bOCEOnKgK9Rl0aA7GEJx4f+TC6HWPo1wURiAVEn
3D1hqp35nayJa+/RQ2AXCnyVw1AoHVLlLOPEOY9wAxB6TWhdUMs0flp2HRn3I54=
=lFPC
-END PGP SIGNATURE-
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
Ivan Zhakov i...@visualsvn.com writes:
FWIW libsvn_fsfs uses dirent-name as key since it's already copied to
result_pool:
That was a FSFS change in r1572049; I think it is a better solution so I
have applied it to FSX.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
, entry.key, entry.keylen),
+ entry.keylen, dirent);
else
APR_ARRAY_PUSH(entries, svn_fs_x__dirent_t *) = dirent;
}
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
Stefan Fuhrmann stefan.fuhrm...@wandisco.com writes:
On Thu, May 28, 2015 at 9:54 PM, Philip Martin
philip.mar...@wandisco.com wrote:
fsync() works on file descriptors rather than files, do we need to keep
the original file descriptors open in order to fsync()?
We could b/c
to a temporary in
the destination directory and then calls apr_file_rename().
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
.
It probably is feasible to keep the file descriptors open, provided we
don't accumulate too many.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
fix svn_io_file_rename() that also make svn_io_file_move() work
properly on Windows. Then if we can get svn_fs_fs__move_into_place() to
call svn_fs_file_move() we move all the EXDEV stuff to io.c.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
requested Windows behaviour that does
not match Unix. Perhaps it was unintentional?
It could be nice
to fix in APR, but they seem to ignore our patches :(
Do they have a record of doing that? I've sent patches that have been
applied.
--
Philip Martin | Subversion Committer
WANdisco // *Non
think the file code is better
that is OK.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
with the stream approach.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
the opposite, the
code that opens the file/stream knows that the file/stream should be
flushed.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
() and it makes the stream code neater. We
could fix that by introducing svn_io_file_printf(). Either we add all
the neat stream features to the file code or we attempt to move to the
stream code.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
Ivan Zhakov i...@visualsvn.com writes:
I hope you're aware the original code has unbounded memory usage?
No, could you explain? Is it a problem with the underlying stream code,
or with the way it was used?
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
that is just
HEAD.
- We could avoid a public API and call some FSFS function from svnfsfs.
I'll probably go with the last option initially. Any comments?
I should note that WANdisco has an interest in this code being
developed.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
manifest file. That means replacing more, if
not all, of the stream code in revprops.c.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
Evgeny Kotkov evgeny.kot...@visualsvn.com writes:
I sketched this option in the attached patch. With this patch applied, we
could rework the r1680819 fix like below. What do you think?
Yes, that looks good.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
, svn_stream_mark_t **mark, apr_pool_t *pool)
{
struct baton_apr *btn = baton;
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
Branko Čibej br...@wandisco.com writes:
On 22.05.2015 10:50, Philip Martin wrote:
Another approach would be to have a function to enable flushing directly
on the stream created by svn_stream_from_aprfile2() without creating a
new stream. This is probably the smallest/simplest patch. After
Ivan Zhakov i...@visualsvn.com writes:
On 22 May 2015 at 11:50, Philip Martin philip.mar...@wandisco.com wrote:
Another approach would be to have a function to enable flushing directly
on the stream created by svn_stream_from_aprfile2() without creating a
new stream. This is probably
));
+
+ svn_stream_set_close(stream, close_handler_flush);
+
+ return SVN_NO_ERROR;
+}
+
+
+static svn_error_t *
mark_handler_apr(void *baton, svn_stream_mark_t **mark, apr_pool_t *pool)
{
struct baton_apr *btn = baton;
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
\n3.0\n3.0\n, 12) = 12
fsync(4)= 0
close(4)= 0
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
Evgeny Kotkov evgeny.kot...@visualsvn.com writes:
Or is it just a gut feeling that we should be using streams here?
We have been gradually moving our file-based code to stream-based code
for years.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
Philip Martin philip.mar...@wandisco.com writes:
So, following the approach of r876245, would something like this do the
trick?
It would help if I built the correct tree. No, that is not enough, the
regression tests fail.
--
Philip
moving away from
file code to stream code. Can we make the flush part of the stream
code? Perhaps we could create a flushing stream that just does flush
and then simply insert the new stream into the old code.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
(svn_stream_write(stream, compressed-data, compressed-len));
SVN_ERR(svn_stream_close(stream));
A patch like that is possibly easier to review than r1680819.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
[0] ^= data[1] APR_UINT64_C(0x);
I don't see why this needs to be reversible, and it's not clear it is
reversible.
The comment was added in r1458643 on the cache-server branch
http://svn.apache.org/viewvc?view=revisionrevision=r1458643
--
Philip Martin | Subversion Committer
Philip Martin philip.mar...@wandisco.com writes:
Is this comment in cache-membuffer.c:combine_key correct?
/* scramble key DATA. All of this must be reversible to prevent key
* collisions. So, we limit ourselves to xor and permutations. */
data[1] = (data[1] 27) | (data
21 pm wc/e
X2 ? ? wc/em
22 pm wc/s
X2 ? ? wc/s/es
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
large enough to trigger this will
probably fail memory allocation first.
In practice keys are short, a 6GB key probably indicates that a bug has
already occurred and that memory is corrupt. However, the code should be
more defensive.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop
Philip Martin philip.mar...@wandisco.com writes:
Prompted by the warnings I think there are some issues to fix. For
APR_HASH_KEY_STRING keys there is no protection against abnormally long
keys. combined_long_key() will allocate strlen() memory even if it is
many GB. The item will not get
Philip Martin philip.mar...@wandisco.com writes:
Prompted by the warnings I think there are some issues to fix. For
APR_HASH_KEY_STRING keys there is no protection against abnormally long
keys. combined_long_key() will allocate strlen() memory even if it is
many GB. The item will not get
Philip Martin philip.mar...@wandisco.com writes:
Philip Martin philip.mar...@wandisco.com writes:
Prompted by the warnings I think there are some issues to fix. For
APR_HASH_KEY_STRING keys there is no protection against abnormally long
keys. combined_long_key() will allocate strlen
or in addition.
Or perhaps trust should be applied to servers via some sort of naming
scheme like the options in the servers file?
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
-valid, other
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
/subversion-1.9.0-rc1.zip.sha1
I see 2443ccc26fd3280b24d38087c5f4291202787f1d as expected:
$ curl -s
https://dist.apache.org/repos/dist/dev/subversion/subversion-1.9.0-rc1.zip |
sha1sum
2443ccc26fd3280b24d38087c5f4291202787f1d -
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
is now a static .
* subversion/tests/libsvn_subr/path-test.c
(test_path_splitext): New case.
Patch by: eric{_AT_}lubin.us
me
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
to ever see this bit of code as svn_error.h has:
#ifdef SVN_DEBUG
#define SVN_ERR__TRACING
#endif
I can change that bit of JNIUtil.cpp code to junk and it still builds.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
max_threads = 1;
so that other test cannot interfere with the use of the cache by this
test.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
. */
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
Philip Martin philip.mar...@wandisco.com writes:
Index: ../src-1.9/subversion/libsvn_ra/ra_loader.c
===
--- ../src-1.9/subversion/libsvn_ra/ra_loader.c (revision 1677108)
+++ ../src-1.9/subversion/libsvn_ra/ra_loader.c
Philip Martin philip.mar...@wandisco.com writes:
When dynamic module loading is enabled I get a SEGV
svnadmin create repo
svnadmin create repo2
svnsync initialize file://`pwd`/repo2 file://`pwd`/repo
I found that because svnsync_authz_tests.py 7 fails over ra-svn as that
test also uses ra
Philip Martin philip.mar...@wandisco.com writes:
Philip Martin philip.mar...@wandisco.com writes:
Index: ../src-1.9/subversion/libsvn_ra/ra_loader.c
===
--- ../src-1.9/subversion/libsvn_ra/ra_loader.c (revision 1677108
==by 0x403F57: open_target_session (svnsync.c:996)
==25549==by 0x403BE6: initialize_cmd (svnsync.c:905)
==25549==by 0x407384: sub_main (svnsync.c:2380)
==25549==by 0x407463: main (svnsync.c:2414)
This is with 1.9.x.
--
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
Philip Martin philip.mar...@wandisco.com writes:
The problem is that the shared struct created in fs_serialized_init
doesn't live long enough:
==25549== Invalid read of size 8
==25549==at 0x8102166: init_lock_baton (fs_fs.c:265)
==25549==by 0x81022D4: create_lock_baton (fs_fs.c:317
201 - 300 of 2536 matches
Mail list logo