This fixes races in thread-local and global tagging in notmuch-search
(e.g., "+", "-", "a", "*", etc.). Previously, these would modify tags
of new messages that arrived after the search. Now they only operate
on the messages that were in the threads when the search was
performed. This prevents surprises like archiving messages that
arrived in a thread after the search results were shown.
This eliminates `notmuch-search-find-thread-id-region(-search)'
because these functions strongly encouraged racy usage.
This fixes the two broken tests added by the previous patch.
These tests check that both thread-local and global search tagging
operations are race-free. They are currently known-broken because
they aren't race-free.
These queries will match exactly the set of messages currently in the
thread, even if more messages later arrive. Two queries are provided:
one for matched messages and one for unmatched messages.
This can be used to fix race conditions with tagging threads from
search results. While tagging based on a thread: query can affect
messages that arrived after the search, tagging based on stable
queries affects only the messages the user was shown in the search UI.
Since we want clients to be able to depend on the presence of these
queries, this ushers in schema version 2.
(Unfortunately, it's difficult to first demonstrate this problem with
a known-broken test because modern Linux kernels have argument length
limits in the megabytes, which makes Emacs really slow!)
It was looking in completely the wrong place for the backup and the
(test) xapian database. Unfortunately test_begin_subtest hides the
relevant errors.
Currently notmuch-show looks at the prefix-arg directly via
current-prefix-arg. This changes it to use the interactive
specification.
One test (for elide-toggle functionality) set the prefix arg
directly. Update this test to set the new argument directly.
One test (reply to encrypted message in the crypto test) recently
started failing on some systems. The failure I saw were two extra
lines of the form
<87d2nbc5xg.fsf@host.i-did-not-set--mail-host-address--so-tickle-me>
The test pipes the output through
grep -v -e '^In-Reply-To:' -e '^References:'
which would normally these two ids but it does not, in this case,
because they are so long they get put on a separate line in the output.
To fix this we set mail-host-address for emacs deliver. example.com
seems a sensible address to use. This is short enough that we don't
get the line breaks above and the tests then all pass.
As explained by Jeffrey Stedfast, the author of GMime, quoted in [1]:
> Passing the GMIME_ENABLE_RFC2047_WORKAROUNDS flag to g_mime_init()
> *should* solve the decoding problem mentioned in the thread. This
> flag should be safe to pass into g_mime_init() without any bad side
> effects and my unit tests do test that code-path.
The thread being referred to is [2].
[1] id:87bo56viyo.fsf@nikula.org
[2] id:08cb1dcd-c5db-4e33-8b09-7730cb3d59a2@gmail.com
Some common broken RFC 2047 encodings that we currently let gmime
parse strictly. We could tell gmime to be forgiving in what it accepts
as RFC 2047 encoding, making these tests pass.
When 'xpg_echo' bash shell option is unset (usually the default)
echo builtin does not expand backslash-escape sequences by default
(i.e. '\n' is echoed as '\n' instead of newline). Not all bash
installations have this feature we depend on activated by default.
Note that the feature is bash (and GNU /bin/echo) specific. It is used
as it is convenient. If portability is needed (elsewhere) use printf(1)
(also often available as a shell builtin).
If any of the tests in our test system is not passing the execution
of the test suite completes with nonzero exit value.
It is better to rely on the exit value of the test system instead
of some arbitrary strings in test output (or use both).
notmuch_message_tags_to_maildir_flags() unconditionally moves messages from
maildir directory "new/" to maildir directory "cur/", which makes messages lose
their "new" status in the MUA. However some users want to keep this "new"
status after, for instance, an auto-tagging of new messages.
However, as Austin mentioned and according to the maildir specification,
messages living in "new/" are not allowed to have flags, even if mutt allows it
to happen. For this reason, this patch prevents moving messages from "new/" to
"cur/", only if no flags have to be changed. It's hopefully enough to satisfy
mutt (and maybe other MUAs showing the "new" status) users checking the "new"
status.
Changelog:
* v2: Fix bool type as well as NULL returned despite having no errors (Austin
Clements)
* v4: Tag the related test (contributed by Michal Sojka) as working
Signed-off-by: Louis Rilling <l.rilling@av7.net>
[Condition for keeping messages in new/ was extended to satisfy all
tests from the previous patch. -Michal Sojka]
[Added by David Bremner, to keep the tests passing at each commit]
update insert tests for new maildir synchronization rules
As of id:1355952747-27350-4-git-send-email-sojkam1@fel.cvut.cz
we are more conservative about moving messages from ./new to ./cur.
This updates the insert tests to match
As mentioned by Jani Nikula in id:87vcccp4y3.fsf@nikula.org, some cases
of maildir synchronization are not covered by our tests. Let's add the
missing tests.
Some MUA's like mutt show the difference between "new" emails living in maildir
directory new/, and "old" emails living in maildir directory cur/. However
notmuch tag unconditionally moves selected messages from new/ to cur/, even if
no maildir synchronized tag is changed.
While maildir specification forbids messages with tags living in new/, there is
no need to move messages to cur/ when no maildir synchronized tag is changed.
Thus notmuch can remain transparent with respect to other MUA's.
[ Edited commit log to better describe the intended changes, and tag the
test as broken until the actual changes are implemented -- Louis Rilling ]
Signed-off-by: Louis Rilling <l.rilling@av7.net>
[ Converted to use test_subtest_known_broken, David Bremner ]
RFC 2047 states that the encoding and charset in an encoded word are
case-insensitive, so force them to lower case in the reply test. This
fixes an issue caused by GMime versions (somewhere between 2.6.10 and
2.6.16), which changed the capitalization of the encoding.
Previously, reply's default text format used an odd mix of RFC 2045
MIME encoding for the reply template's body and some made-up RFC
2822-like UTF-8 format for the headers. The intent was to present the
headers to the user in a nice, un-encoded format, but this assumed
that whatever ultimately sent the email would RFC 2047-encode the
headers, while at the same time the body was already RFC 2045 encoded,
so it assumed that whatever sent the email would *not* re-encode the
body.
This can be fixed by either producing a fully decoded UTF-8 reply
template, or a fully encoded MIME-compliant RFC 2822 message. This
patch does the latter because it is
a) Well-defined by RFC 2822 and MIME (while any UTF-8 format would be
ad hoc).
b) Ready to be piped to sendmail. The point of the text format is to
be minimal, so a user should be able to pop up the template in
whatever editor they want, edit it, and push it to sendmail.
c) Consistent with frontend capabilities. If a frontend has the
smarts to RFC 2047 encode the headers before sending the mail, it
probably has the smarts to RFC 2047 decode them before presenting
the template to a user for editing.
Also, as far as I know, nothing automated consumes the reply text
format, so changing this should not cause serious problems. (And if
anything does still consume this format, it probably gets these
encoding issues wrong anyway.)
Previously, the References header code seemed to assume
notmuch_message_get_header would return NULL if the header was not
present, but it actually returns "". As a result of this, it was
inserting an unnecessary space when concatenating an empty or missing
original references header with the new reference.
This shows up in only two tests because the text reply format later
passes the whole reply template through g_mime_filter_headers, which
has the side effect of stripping out this extra space.
This switches `notmuch-mua-reply' and `notmuch-query-get-threads' to
the S-exp format. These were the last two uses of the JSON format in
the Emacs frontend.
While looked good on paper, its attempted use caused confusion, complexity,
and potential for information leak when passed through wrapper scripts.
For slimmer code and to lessen demand for maintenance/support the set of
commits which added top level --stderr= option is now reverted.
The find option syntax `-perm +111` is deprecated gnu find feature.
The replacement `( -perm -100 -o -perm -10 -o -perm 1 )` should also
work outside of the GNU domain.
In addition to being the Right Thing to do, this noticeably improves
the time taken to display the first page of search results, since it's
roughly an order of magnitude faster than the JSON parser.
Interestingly, it does *not* significantly improve the time to
completely fill a large search buffer because for large search
buffers, the cost of creating author invisibility overlays and
inserting text (which slows down with more overlays) dominates.
However, the time required to display the first page of results is
generally more important to the user experience.
Previously, search started the async notmuch process directly. Now,
it uses `notmuch-start-notmuch'. This simplifies the process sentinel
a bit and means that we no longer have to worry about errors
interleaved with the JSON output.
We also update the tests of Emacs error handling, since the error
output is now separated from the search results buffer.
We now check error handling more carefully in the last test in
test/emacs and we're about to add more error handling tests. (This
was also a strange place for this test, since it had nothing to do
with large search buffers.)
This unifies the part button actions and the underlying part action
functions into single interactive command that simply applies to the
part containing point using the just-added part p-list text property
instead of button properties. Since all part actions can be performed
by applying the appropriate mm function to an mm-handle, this patch
abstracts out the creation of mm handles, making the implementations
of the part commands trivial. This also eliminates our special
handling for part save in favor of using the appropriate mm function.
This necessarily modifies the way we handle the default part button
action, but in a way that does not change the meaning of the
notmuch-show-part-button-default-action defcustom.
Since these commands are no longer specific to buttons, this patch
eliminates the extra metadata stored with each button. This also
eliminates one rather special-purpose macro for a collection of
general purpose part handling utilities.
--stderr=FILE tests were added to test/help-test as it is the one
doing most global option testing. Also, it was simplest to test
this new option using `notmuch help` command.
Presently, the code which finds the parent of a message as it is being
added to the database assumes that the first Message-ID-like substring
of the In-Reply-To header is the parent Message ID. Some mail clients,
however, put stuff other than the Message-ID of the parent in the
In-Reply-To header, such as the email address of the sender of the
parent. This can fool notmuch.
The updated algorithm prefers the last Message ID in the References
header. The References header lists messages oldest-first, so the last
Message ID is the parent (RFC2822, p. 24). The References header is
also less likely to be in a non-standard
syntax (http://cr.yp.to/immhf/thread.html,
http://www.jwz.org/doc/threading.html). In case the References header
is not to be found, fall back to the old behavior.
V2 of this patch, incorporating feedback from Jani and (indirectly)
Austin.
The use of realpath(3) in
commit 58ed67992d
Author: Jani Nikula <jani@nikula.org>
Date: Sun Apr 7 20:15:03 2013 +0300
cli: config: do not overwrite symlinks when saving config file
broke config file save when the file does not exist, which results in
'notmuch setup' always failing to create a new config file.
Fix by checking ENOENT from realpath(3).
Use realpath to canonicalize the config path before writing.
Previously 'notmuch setup' and 'notmuch config set' overwrote the
config file even if it was a symbolic link.
We now have a notmuch_config_is_new() function to query whether a
config was created or not. Change the notmuch_config_open() is_new
parameter into boolean create_new to determine whether the function
should create a new config if one doesn't exist. This reduces the
complexity of the API.
When execution of tests is interrupted by signal coming outside of the
test system itself, output just one line "interrupted by signal <num>"
message to standard output. This distinguishes the case from internal
exit and reduces noise.
Set the variable '$test_subtest_name' in all functions which starts
a new test and use that variable in all functions that output
test results.
Additionally output the latest '$test_subtest_name' in case of
abnormal exit, to avoid confusion.
Instead of checking immediately for the watched process, delay a
minute, or in the case that process-attributes returns nil, for two
minutes. This is intended to cope with the case that
process-attributes is unimplimented, and returns always returns nil.
In this case, the watchdog check is the same as the two minute limit
imposed by timeout.
The TERM environment variable is set to 'dumb' when running tests, but
the original value of it is stored for echoing colors and running emacs
(somewhat interactively) in detached session. Emacs requires some
terminal control sequences to be available for interactive operation.
In case original TERM is (also) 'dumb' (or unset/empty) emacs cannot
run interactively. To fix this problem dtach (and emacs as it's child
process) is run with TERM=vt100 in case original TERM was unset, empty
or 'dumb'. This way there is a chance to run emacs tests with different
user terminals and potentially find problems there.
This test also serves as documentation of the quoting
requirements. The comment lines are so that it exactly matches the man
page. Nothing more embarrassing than having an example in the man page
fail.
The (now fixed) bug that this test revealed is that unquoted
message-ids with whitespace or other control characters in them are
split into several tokens by the Xapian query parser.
This is based on the similar test for notmuch restore, but the parser
in batch tagging mode is less tolerant of a few cases, in particular
those tested by illegal_tag.
We recently switched to popping up a buffer to report CLI errors, but
this was too intrusive, especially for transient errors and especially
since we made fewer things ignore errors. This patch changes this to
display a basic error message in the minibuffer (using Emacs' usual
error handling path) and, if there are additional details, to log
these to a separate error buffer and reference the error buffer from
the minibuffer message. This is more in line with how Emacs typically
handles errors, but makes the details available to the user without
flooding them with the details.
Given this split, we pare down the basic message and make it more
user-friendly, and also make the verbose message even more detailed
(and more debugging-oriented).
This switches the new batch-tag format away from using a home-grown
hex-encoding scheme for message IDs in the dump to simply using Xapian
queries with Xapian quoting syntax.
This has a variety of advantages beyond presenting a cleaner and more
consistent interface. Foremost is that it will dramatically simplify
the quoting for batch tagging, which shares the same input format.
While the hex-encoding is no better or worse for the simple ID queries
used by dump/restore, it becomes onerous for general-purpose queries
used in batch tagging. It also better handles strange cases like
"id:foo and bar", since this is no longer syntactically valid.
When we switch to using regular Xapian queries in the dump format, \n
will cause problems, so we disallow it. Specially, while Xapian can
quote and parse queries containing \n without difficultly, quoted
queries containing \n still span multiple lines, which breaks the
line-orientedness of the dump format. Strictly speaking, we could
still round-trip these, but it would significantly complicate restore
as well as scripts that deal with tag dumps. This complexity would
come at absolutely no benefit: because of the RFC 2822 unfolding
rules, no amount of standards negligence can produce a message with a
message ID containing a line break (not even Outlook can do it!).
Hence, we simply disallow it.
This patch corrects several undesirable behaviours:
1) Empty files were not detected, leading to buffer read overrun.
2) An initial blank line cause restore to silently abort
3) Initial comment line caused format detection to fail
notmuch_json_show_sanitize replaced "filename" field values even in part
structures, where the value is predictable. Make it only normalize the
filename value if it is an absolute path (begins with slash), which is
true of the Maildir filenames that were intended to be normalized away.
This slightly changes the output of an existing test since we now
report non-zero exits with a pop-up buffer instead of at the end of
the search results.
* test/emacs:
- Rename subtests "{Add,Remove} tag from notmuch-show view" to
"notmuch-show: {add,remove} single tag {to,from} single message"
to be consistent with the following tests.
- New subtest "notmuch-show: add multiple tags to single message":
`notmuch-show-add-tag' ("+") can add multiple tags to a message.
- New subtest "notmuch-show: remove multiple tags from single message":
`notmuch-show-remove-tag' ("-") can remove multiple tags from a message.
We want to test both that error/warning messages are generated when
they should be, and not generated when they should not be. This varies
between restore and batch tagging.
These one need the completed functionality in notmuch-restore. Fairly
exotic tags are tested, but no weird message id's.
We test each possible input to autodetection, both explicit (with
--format=auto) and implicit (without --format).
The quoting for ${SEARCH} is broken when it's supposed to be '*', and
it seems tricky to get it right. Just drop the variable and use '*'
directly. Before this, none of the messages ever matched, and the test
was comparing zeros.
test_expect_equal_json uses json.tool from the system Python. While
Python 2 wasn't picky about the encoding of stdin, Python 3 decodes
stdin strictly according to the environment. Since we set LC_ALL=C
for the tests, Python 3's json.tool was assuming stdin would be in
ASCII and aborting when it couldn't decode the UTF-8 characters from
some of the JSON tests. This patch sets the PYTHONIOENCODING
environment variable to utf-8 when invoking json.tool to override
Python's default encoding choice.
Without this change, GCC complains as follows:
gcc test/random-corpus.o test/database-test.o notmuch-config.o command-line-arguments.o lib/libnotmuch.a util/libutil.a parse-time-string/libparse-time-string.a -o test/random-corpus -lgmime-2.6 -lgio-2.0 -lgobject-2.0 -lglib-2.0 -Wl,-rpath,/usr/lib -ltalloc -lxapian
/usr/bin/ld: lib/libnotmuch.a(database.o): undefined reference to symbol '_ZNSs4_Rep10_M_destroyERKSaIcE@@GLIBCXX_3.4'
/usr/bin/ld: note: '_ZNSs4_Rep10_M_destroyERKSaIcE@@GLIBCXX_3.4' is defined in DSO /usr/lib/libstdc++.so.6 so try adding it to the linker command line
/usr/lib/libstdc++.so.6: could not read symbols: Invalid operation
collect2: error: ld returned 1 exit status
make: *** [test/random-corpus] Error 1
We demonstrate the current notmuch restore parser being confused by
message-id's and tags containing non alpha numeric characters
(particularly space and parentheses are problematic because they are
not escaped by notmuch dump).
We save the files as hex escaped on disk so that terminal emulators
will not get confused if the test fails (as we mostly expect it to do).
Initial use case is testing dump and restore, so we only have
message-ids and tags.
The message ID's are nothing like RFC compliant, but it doesn't seem
any harder to roundtrip random UTF-8 strings than RFC-compliant ones.
Tags are UTF-8, even though notmuch is in principle more generous than
that.
updated for id:m2wr04ocro.fsf@guru.guru-group.fi
- talk about Unicode value rather some specific encoding
- call talloc_realloc less times
Initially, provide a way to create "stub" messages in the notmuch
database without corresponding files. This is essentially cut and
paste from lib/database.cc. This is a seperate file since we don't
want to export these symbols from libnotmuch or bloat the library with
non-exported code.
Previously, the test framework generated a variable name for each
external prereq as a poor man's associative array. Unfortunately,
prereqs names may not be legal variable names, leading to
unintelligible bash errors like
test_missing_external_prereq_emacsclient.emacs24_=t: command not found
Using proper associative arrays to track prereqs, in addition to being
much cleaner than generating variable names and using grep to
carefully construct unique string lists, removes restrictions on
prereq names.
Previously, if a test script aborted (e.g., because it passed too few
arguments to a test function), the test driver loop would simply
continue on to the next test script and the final results would
declare that everything passed (except that the test count would look
suspiciously low, but maybe you just misremembered how many tests
there were).
Now, if a test script exits with a non-zero status and did not produce
a final results file, we propagate that failure out of the driver loop
immediately.
To keep this simple, this patch removes the PID from the test-results
file name. This PID was inherited from the git test system and seems
unnecessary, since the file name already includes the name of the test
script and the test-results directory is created anew for each run.
And require that if TEST_EMACS is specified, so is TEST_EMACSCLIENT.
Previously, the test framework always used "emacsclient", even if the
Emacs in use was overridden by TEST_EMACS. This causes problems if
both Emacs 23 and Emacs 24 are installed, the Emacs 23 emacsclient is
the system default, but TEST_EMACS is set to emacs24. Specifically,
with an Emacs 24 server and an Emacs 23 client, emacs tests that run
very quickly may produce no output from emacsclient, causing the test
to fail.
The Emacs server uses a very simple line-oriented protocol in which
the client sends a request to evaluate an expression and the server
sends a request to print the result of evaluation. Prior to Emacs bzr
commit 107565 on March 11th, 2012 (released in Emacs 24.1), if
multiple commands were sent to the emacsclient between when it sent
the evaluation command and when it entered its receive loop, it would
only process the first response command, ignoring the rest of the
received buffer. This wasn't a problem with the Emacs 23 server
because it sent only the command to print the evaluation result.
However, the Emacs 24 server first sends an unprompted command
specifying the PID of the Emacs server, then processes the evaluation
request, then sends the command to print the result. If the
evaluation is fast enough, it can send both of these commands before
emacsclient enters the receive loop. Hence, if an Emacs 24 server is
used with an Emacs 23 emacsclient, it may miss the response printing
command, ultimately causing intermittent notmuch test failures.
Previously, many tests in emacs-subject-to-filename didn't quote the
$output argument to test_expect_equal. As a result, if $output was
empty, test_expect_equal would be passed only one argument and would
abort the entire test script. By quoting the argument, we ensure
test_expect_equal will always receive two arguments.
We now test for user ignore patterns before attempting to determine if
a directory entry is itself a directory. As a result, we no longer
abort for broken symlinks if the user has explicitly ignored them.
This fixes the test added in the previous patch. It also slightly
changes the debug output checked by another test of ignores.
