Revert commit 8370e3cfe2, and remark the
corresponding test as broken. Also update the expected output of the
broken test to show excludes active in the user defined saved searches.
Commit [0] left the stemmer object accessible, but did not add
de-allocation code to notmuch_database_destroy. This commit corrects
that oversight.
Leak originally reported by Austin Ray [1].
[0]: 3202e0d1fe
[1]: id:20220105224538.m36lnjn7rf3ieonc@athena
In [1] Austin Ray reported some memory leaks in
notmuch_database_open. One of those leaks is caused by jumping to the
next key without freeing val. This change avoids that leak.
[1]: id:20220105224538.m36lnjn7rf3ieonc@athena
In particular tags that only occur on otherwise excluded messages do
not show up in "All tags" without this change. The reported numbers
_do_ take excludes into account, so it is possible that
searches (e.g. the aformentioned tags) will show up with a count of
zero.
Gregor Zattler observed that tags could be unintentionally hidden in
the "All tags" view, and Tomi Ollia worked out [2] that the issue was tags
that only occured on excluded messages. This test reproduces that bug.
[1]: id:87wox1vovj.fsf@len.workgroup
[2]: id:m28t9faaim.fsf@guru.guru-group.fi
Although it makes sense for the extra headers to be added to the copy
of the message headers included in the sexp/json, it is a bit
surprising for them to show in the new message constructed for the
reply, especially when, as here, they are always missing/empty.
This is based on a patch from Johan Parin [1], which is in turn
responding to a bug report / feature requiest from Jan Malkhovski.
The update to the structured output documented in schemata is intended
to be upward compatible, so the format version stays the same
[1]: id:20191116162723.18343-1-johan.parin@gmail.com
[2]: id:87h8sdemnr.fsf@oxij.org
We already use an allocated (and presumably open) database as a talloc
context. Keeping the pointer in the allocated struct will allow us to
e.g. interrogate the configuration in a sprinter function without
threading the database all the way through the various levels of function.
Since the bug was first reported in [1], notmuch has gained the
ability to have the database located outside the mail root, hence this
this change differs slightly from Jani's proposed solution [2] in not
using notmuch_database_get_path, but rather the already retrieved
mail_root.
[1]: id:87mwhifu9a.fsf@trouble.defaultvalue.org
[2]: id:87ios5v59p.fsf@nikula.org
In [1] Rob observed that notmuch new ignored directories called
.notmuch everywhere in the tree, where they should only (and now, with
split configs, at most) be ignored at the top level. Add a test to
demonstrate the problem.
[1]: id:87mwhifu9a.fsf@trouble.defaultvalue.org
If we know the configuration is split, but there is no mail root
defined, this indicates a (lack of) configuration error. Currently
this can only arise in XDG configurations.
The extra talloc struct "local" was left over from before the notmuch
struct was allocated earlier. Having the notmuch struct available in
this function will allow more flexibility to track the configuration
variations (e.g. split vs. non-split).
If notmuch_database_open_with_config finds a database, but that
database is not in a legacy, non-split configuration, then it
currently incorrectly deduces the mail root and returns SUCCESS. Add
to two tests to demonstrate this bug.
emacs-mac that compiled with OS X system API for image display
support cannot correctly render svg without xmlns parameter [1].
[1]: id:tencent_127AA231767438AC66FEE4DDB4BBF51DF909@qq.com
As stressed by the gpg documentation, the non-'with-colons' output
format is subject to change, and indeed it did in 2.3.x (x<=3). This
should make the the test suite more robust against such changes.
New notmuch-show-header-line customizable boolean to allow inhibiting
a header line in notmuch-show-mode buffers (for instance, because one
prefers to just include Subject in notmuch-message-headers).
The previous (pre-0.34.2) constructor searched for a config file but
only if the database path was not specified, and only to retrieve
database.path. Neither of the available options (CONFIG.SEARCH or
CONFIG.NONE) matches this semantics exactly, but CONFIG.SEARCH causes
less breakage for people who relied on the old behaviour to set their
database.path [1]. Since it also seems like the friendlier option in
the long run, this commit switches to CONFIG.SEARCH as default.
This requires a certain amount of updating the pytest tests, but most
users will actually have a config file, unlike the test environment.
[1]: id:87fsqijx7u.fsf@metapensiero.it
If we return regular Message objects, python will try to destroy them,
and the underlying notmuch object, causing e.g. the crash [1].
[1]: id:87sfu6utxg.fsf@tethera.net
This reproduces the bug reported at [1]. The second test hints at the
solution, making reply return OwnedMessage objects.
[1]: id:87sfu6utxg.fsf@tethera.net
Since mail store and database directory can be different,
path and folder completions must use the directory given by
database.mail_root, not by database.path.
The mail store directory is given by database.mail_root,
which can be different from database.path.
However, notmuch-insert documentation was still referencing the latter
as the provider of the maildir directory instead of the former.
Existing users of the legacy python bindings use
message.get_flags(Message.FLAG.MATCH) to determine which messages in a
thread matched. Since the bindings don't provide get_flags anymore,
they should provide a property analogous to the existing "excluded"
property.
In addition to avoiding problems with user specified include paths
picking up an installed version of notmuch.h, this should also enable
users to override more options (in particular they could override
warning options since the last one takes effect).
[1]: id:7851CAB5-4556-4931-A0A2-37003E56C927@ryandesign.com
In [1] Ryan Schmidt reported a problem on macports [2] with notmuch
finding an existing installed version of notmuch.h during the build.
This is a partial fix; the user might also specify -I in CFLAGS.
[1]: id:7851CAB5-4556-4931-A0A2-37003E56C927@ryandesign.com
[2]: The problem does not arise when notmuch.h is installed into a
"system" include path that the compiler searches by default.
In [1] Ryan Schmidt reported a problem on macports [2] with notmuch
finding an existing installed version of libnotmuch during the build
when the user specified LDFLAGS including the libnotmuch install
directory.
This change should prevent that.
LDFLAGS also occurs in FINAL_LIBNOTMUCH_LDFLAGS. The only built
library linked to that is util/libtnotmuch_util.a, and that passed as
explicit (relative) path, and is thus not affected by -L.
[1]: id:7851CAB5-4556-4931-A0A2-37003E56C927@ryandesign.com
[2]: The problem does not arise when libnotmuch is installed into a
"system" library path that the compiler/linker searches by default.
Jani reported that lines prefixed with '>' (as opposed to '> ') are
not highlighted properly [1]. David E responded with a updated regex
[2]. This change implements David E's suggestion.
[1]: id:87a8b5pcky.fsf@nikula.org
[2]: id:m2pok1e3gv.fsf@dme.org
This will correct the current use of an undefined variable when
setting LD_LIBRARY_PATH in doc/Makefile.local
It is tempting to try to replace the use of test/export-dirs.sh, but
this is not as simple as it looks, as NOTMUCH_BUILDDIR is used to
locate sh.config, so probably cannot also sensibly be used to define
it.
It makes perfect sense for users to want to pre-create .notmuch,
e.g. to install hooks, so we should handle the case of a .notmuch
directory without an actual xapian database more gracefully.
`notmuch new' should go ahead and create the xapian database if it is
missing, even in the case where the parent .notmuch (or equivalent)
directory exists.
In [1] Mark Walters reported a problem with messages being removed
from the database when the parent directory was renamed. Jani Nikula
proposed [2] these tests but observed
This test is not suitable for merging since it's not deterministic.
After applying Jani's patch [3], the tests now pass deterministically,
and could usefully act as regression tests.
[1]: id:87siray6th.fsf@qmul.ac.uk
[2]: id:1393191650-28333-1-git-send-email-jani@nikula.org
[3]: id:1441445731-4362-2-git-send-email-jani@nikula.org
In [1] Daniel observed that the gzipped man pages were only being
rebuild every second time when building with `make -j4'. This may be
caused by a race condition between sphinx-build rebuilding the roff
files and the recipe to gzip them. This commit sequentializes these
two steps by making the stamp file a prerequisite for (all of) the
gzip files.
[1]: id:87tveotn1g.fsf@fifthhorseman.net
