notmuch.git
13 years agoNew part output handling as option to notmuch-show.
Jameson Graef Rollins [Mon, 23 May 2011 22:31:32 +0000 (15:31 -0700)]
New part output handling as option to notmuch-show.

Outputting of single MIME parts is moved to an option of notmuch show,
instead of being handled in it's own sub-command.  The recent rework
of multipart mime allowed for this change but consolidating part
handling into a single recursive function (show_message_part) that
includes formatting.  This allows for far simpler handling single
output of a single part, including formatting.

13 years agoNormalize part counting and formatting in show_message_part function.
Jameson Graef Rollins [Mon, 23 May 2011 05:08:08 +0000 (22:08 -0700)]
Normalize part counting and formatting in show_message_part function.

Simplify the function by moving part counting and formatting outside
of conditionals, thereby eliminating redundant code.  This also wraps
message part output handling with proper part formatting.

13 years agorename do_show_raw to do_show_single, and create params.raw for raw message output
Jameson Graef Rollins [Mon, 23 May 2011 03:04:02 +0000 (20:04 -0700)]
rename do_show_raw to do_show_single, and create params.raw for raw message output

We rename here in order to make do_show_single into a generic function
for handling output of just a single message, or which format=raw is a
special case.  The raw case is handled by setting a new parameter,
params.raw, which is used to tell do_show_single to output a single
message as a raw file.

This is mostly in preparation for much improved part handling to
follow imminently.

13 years agoadd part_sep formatter to replace "first" argument to part format functions
Jameson Graef Rollins [Mon, 23 May 2011 02:56:53 +0000 (19:56 -0700)]
add part_sep formatter to replace "first" argument to part format functions

A new field "part_sep" is added to the notmuch_show_format structure,
to be used for part separation.  This is cleaner than the "first"
argument that was being passed around to the part arguments, and
allows the function that handles overall part output formatting
(show_message_part) to directly handle when outputting the separator.

13 years agocreate notmuch_show_params_t structure for holding parameters passed to show functions.
Jameson Graef Rollins [Fri, 20 May 2011 22:01:52 +0000 (15:01 -0700)]
create notmuch_show_params_t structure for holding parameters passed to show functions.

This simplifies the passing of arguments to the show functions.  This
will be very useful as we accumulate more parameters that will need to
be passed.  Currently only the entire_thread parameter is passed this
way.

13 years agotest: force deletion of test remnants
Jameson Graef Rollins [Sat, 7 May 2011 19:10:04 +0000 (12:10 -0700)]
test: force deletion of test remnants

This keeps the test from failing if only a subset of the remnants were
available for deletion, because e.g. only a subset of the tests were
run.

13 years agotest: allow specifying tests to run with NOTMUCH_TESTS env var
Jameson Graef Rollins [Sat, 7 May 2011 19:08:34 +0000 (12:08 -0700)]
test: allow specifying tests to run with NOTMUCH_TESTS env var

This is useful for just running a specific subset of tests, ie:

NOTMUCH_TESTS=crypto make test

13 years agoTODO: Add note for bug with message with References but no In-Reply-To
Carl Worth [Mon, 23 May 2011 21:53:52 +0000 (14:53 -0700)]
TODO: Add note for bug with message with References but no In-Reply-To

We recently noticed on the mailing list a case where a message wasn't
properly connected to its parent. We should fix that of course.

13 years agonotmuch part: Fix part numbering to match what's reported by "notmuch show"
Carl Worth [Mon, 23 May 2011 21:27:23 +0000 (14:27 -0700)]
notmuch part: Fix part numbering to match what's reported by "notmuch show"

Since commit c51d5b3cdb5ca0816816e88ca6f7136a24e74eee we are counting
multipart containers when emitting part numbers in the "notmuch show"
output. Unfortunately, "notmuch part" wasn't updated with the same
numbering, (and the test suite is inadequate to catch this).

Fix this by adding a similar part-numbering change to "notmuch part" here.

13 years agopass entire format structure to various show_message functions
Jameson Graef Rollins [Fri, 20 May 2011 18:45:33 +0000 (11:45 -0700)]
pass entire format structure to various show_message functions

Various show_message* functions require formatting functions, which
were previously being passed individually as arguments.  Since we will
need to be needing to passing in more formatting function in the
future (ie. for crypto support), we here modify things so that we just
pass in the entire format structure.  This will make things much
simpler down the line as we need to pass in new format functions.

We move the show_format structure into notmuch-client.c as
notmuch_show_format.  This also affects notmuch-reply.c, so we create
a mostly-empty format_reply to pass the reply_part function to
show_message_body.

13 years agotest: remove hard-coded paths from multipart test
Jameson Graef Rollins [Thu, 19 May 2011 13:04:06 +0000 (06:04 -0700)]
test: remove hard-coded paths from multipart test

Small oversite, easily corrected.

13 years agoAdd a .dir-locals.el file for the benefit of emacs users
Daniel Kahn Gillmor [Wed, 24 Nov 2010 05:45:43 +0000 (00:45 -0500)]
Add a .dir-locals.el file for the benefit of emacs users

This file causes emacs to automatically set the correct style
variables to control indentation, etc. One more thing to make it
easier for everyone to collaborate with a consistent coding style.

Signed-off-by: Jameson Rollins <jrollins@finestructure.net>
13 years agoemacs: Show cleaner `From:' addresses in the summary line.
David Edmondson [Wed, 19 May 2010 10:21:05 +0000 (11:21 +0100)]
emacs: Show cleaner `From:' addresses in the summary line.

Remove double quotes and flatten "foo@bar.com <foo@bar.com>" to
"foo@bar.com".

Edited-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net> (clean up
expected output for emacs tests).

Signed-off-by: Jameson Rollins <jrollins@finestructure.net>
13 years agoemacs: Add custom `notmuch-show-elide-same-subject'
David Edmondson [Wed, 19 May 2010 09:41:08 +0000 (10:41 +0100)]
emacs: Add custom `notmuch-show-elide-same-subject'

This controls the appearance of collapsed messages in notmuch-show
mode, avoiding redundancy for repeated subject).

Remove `notmuch-show-always-show-subject'.

Signed-off-by: Jameson Rollins <jrollins@finestructure.net>
13 years agoemacs: Add `notmuch-show-always-show-subject', allowing control over
David Edmondson [Tue, 18 May 2010 08:36:58 +0000 (09:36 +0100)]
emacs: Add `notmuch-show-always-show-subject', allowing control over
the display of collapsed messages.

Signed-off-by: Jameson Rollins <jrollins@finestructure.net>
13 years agoemacs: Allow renderer of multipart/related parts access to non-primary parts.
David Edmondson [Mon, 17 May 2010 12:43:51 +0000 (13:43 +0100)]
emacs: Allow renderer of multipart/related parts access to non-primary parts.

Typically used to allow a `text/html' renderer access to images which
are sent along with the HTML.

This is not enabled by default, instead the user must execute
`notmuch-show-setup-w3m' for it to take effect.

Edited-by: Carl Worth <cworth@cworth.org> Add documentation string for
notmuch-show-setup-23m and clean up warning about reference/assignment
of free variable.

Signed-off-by: Jameson Rollins <jrollins@finestructure.net>
13 years agoemacs: Optionally show all parts in multipart/alternative.
David Edmondson [Mon, 17 May 2010 12:41:12 +0000 (13:41 +0100)]
emacs: Optionally show all parts in multipart/alternative.

Add a variable `notmuch-show-all-multipart/alternative-parts' that
allows the user to indicate that all candidate sub-parts of a
multipart/alternative part should be shown rather than just the
preferred part. The default is `nil', showing only the preferred part.

This is mostly a debugging aid.

Signed-off-by: Jameson Rollins <jrollins@finestructure.net>
13 years agonotmuch: Add the content-id of a part to the JSON output if it is known.
David Edmondson [Mon, 17 May 2010 12:38:06 +0000 (13:38 +0100)]
notmuch: Add the content-id of a part to the JSON output if it is known.

It's simply one more property of a MIME part that might be useful, and
json makes it so easy to add additional properties.

Signed-off-by: Jameson Rollins <jrollins@finestructure.net>
13 years agotest: Link to compat files when building program during "make test"
Carl Worth [Wed, 18 May 2011 20:15:46 +0000 (13:15 -0700)]
test: Link to compat files when building program during "make test"

The compilation of the smtp-dummy program would fail if a build was
attempted on a system without getline. Fix this by simply including
the existing notmuch_compat_srcs variable when constructing the list
of source files for compiling smtp-dummy.

13 years agoemacs: Render text/x-vcalendar parts.
David Edmondson [Wed, 12 May 2010 08:07:49 +0000 (09:07 +0100)]
emacs: Render text/x-vcalendar parts.

Use code from icalendar.el to convert text/x-vcalendar parts to
something suitable for use with the Emacs diary.

Signed-off-by: Jameson Rollins <jrollins@finestructure.net>
13 years agoemacs: Allow indentation of multipart children.
David Edmondson [Mon, 10 May 2010 10:15:30 +0000 (11:15 +0100)]
emacs: Allow indentation of multipart children.

Signed-off-by: Jameson Rollins <jrollins@finestructure.net>
13 years agoemacs: Add `notmuch-show-multipart/alternative-discouraged'.
David Edmondson [Mon, 10 May 2010 10:03:08 +0000 (11:03 +0100)]
emacs: Add `notmuch-show-multipart/alternative-discouraged'.

Also improved implementation of indication of which parts are
not shown.

Signed-off-by: Jameson Rollins <jrollins@finestructure.net>
13 years agoemacs: add more part handling functions
David Edmondson [Fri, 13 May 2011 06:09:30 +0000 (23:09 -0700)]
emacs: add more part handling functions

This adds new notmuch-show-insert-part functions to handle
multipart/alternative and message/rfc822 parts.

13 years agonotmuch show: Properly nest MIME parts within mulipart parts
Carl Worth [Tue, 17 May 2011 22:34:57 +0000 (15:34 -0700)]
notmuch show: Properly nest MIME parts within mulipart parts

Previously, notmuch show flattened all output, losing information
about the nesting of the MIME hierarchy. Now, the output is properly
nested, (both in the --format=text and --format=json output), so that
clients can analyze the original MIME structure.

Internally, this required splitting the final closing delimiter out of
the various show_part functions and putting it into a new
show_part_end function instead. Also, the show_part function now
accepts a new "first" argument that is set not only for the first MIME
part of a message, but also for each first MIME part within a series
of multipart parts. This "first" argument controls the omission of a
preceding comma when printing a part (for json).

Many thanks to David Edmondson <dme@dme.org> for originally
identifying the lack of nesting in the json output and submitting an
early implementation of this feature. Thanks as well to Jameson Graef
Rollins <jrollins@finestructure.net> for carefully shepherding David's
patches through a remarkably long review process, patiently explaining
them, and providing a cleaned up series that led to this final
implementation. Jameson also provided the new emacs code here.

13 years agonotmuch show: Include output for the enclosing multipart part of a MIME mail
Carl Worth [Tue, 17 May 2011 05:28:36 +0000 (22:28 -0700)]
notmuch show: Include output for the enclosing multipart part of a MIME mail

Previously, the outer multipart part of any multipart/mixed,
multipart/signed, etc. MIME message was silently omitted from the
"notmuch show" output. This prevented any client from correctly
determining to which parts a signature applies, for example.

Now, we actually emit these parts as their own parts. The output is
still flattened---the contained parts are not yet included "within"
the multipart part---so it's still not possible to determine to which
parts a signature applies, but this is one step along the path.

The test suite is updated to reflect this change, (though we'll
eventually want to fix the emacs interface to not display buttons for
the multipart enclosure parts as there's nothing useful for the user
to actually do with them).

13 years agotest: Add a test of "notmuch show" with a multipart message
Carl Worth [Mon, 16 May 2011 22:25:25 +0000 (15:25 -0700)]
test: Add a test of "notmuch show" with a multipart message

This tests "notmuch show" with both --format=text and --format=json on
a message with some non-trivial MIME multipart nesting, (multiple parts
within a multipart/mixed part which is within a multipart/signed part).

The test captures the current behavior (where only the leaf nodes of
the MIME structure are emitted as a flat list---the multipart parts
are effectively ignored). We plan to soon change the json output at
least to emit an actual hierarchy matching the MIME structure, (at
which point we will update this test).

13 years agoMark some structures in the library interface with visibility=default attribute.
Carl Worth [Wed, 11 May 2011 20:23:13 +0000 (13:23 -0700)]
Mark some structures in the library interface with visibility=default attribute.

As of gcc 4.6, there are new warnings from -Wattributes along the lines of:

warning: ‘_notmuch_messages’ declared with greater visibility
than the type of its field ‘_notmuch_messages::iterator’
[-Wattributes]

To squelch these, we decorate all such containing structs with
__attribute__((visibility("default"))). We take care to let only the
C++ compiler see this, (since the C compiler would otherwise warn
about ignored visibility attributes on types).

13 years agoRemove some variables which were set but not used.
Carl Worth [Wed, 11 May 2011 19:34:13 +0000 (12:34 -0700)]
Remove some variables which were set but not used.

gcc (at least as of version 4.6.0) is kind enough to point these out to us,
(when given -Wunused-but-set-variable explicitly or implicitly via -Wunused
or -Wall).

One of these cases was a legitimately unused variable. Two were simply
variables (named ignored) we were assigning only to squelch a warning about
unused function return values. I don't seem to be getting those warnings
even without setting the ignored variable. And the gcc docs. say that the
correct way to squelch that warning is with a cast to (void) anyway.

13 years agoemacs: Only compile replacement functions for emacs < emacs-23
Carl Worth [Wed, 11 May 2011 19:24:46 +0000 (12:24 -0700)]
emacs: Only compile replacement functions for emacs < emacs-23

This avoids the emacs lisp compiler from emitting warnings on this
replacement code, (which warnings would be hard for us to eliminate
since we didn't write the code but copied it verbatim from emacs 23).

13 years agovim: parse 'from' address
Felipe Contreras [Sun, 6 Feb 2011 12:53:58 +0000 (14:53 +0200)]
vim: parse 'from' address

In order to pass it to sendmail.

Signed-off-by: Felipe Contreras <felipe.contreras@gmail.com>
13 years agovim: use sendmail directly
Felipe Contreras [Wed, 16 Jun 2010 22:39:04 +0000 (01:39 +0300)]
vim: use sendmail directly

The problem with 'mailx' is that it's not standardized, and it doesn't
allow the -f option, which is pretty important on many sendmail
configurations.

Signed-off-by: Felipe Contreras <felipe.contreras@gmail.com>
13 years agovim: add myself (Felipe Contreras) as contributor
Felipe Contreras [Wed, 4 May 2011 20:14:37 +0000 (23:14 +0300)]
vim: add myself (Felipe Contreras) as contributor

Signed-off-by: Felipe Contreras <felipe.contreras@gmail.com>
13 years agovim: implement archive in show view
Felipe Contreras [Thu, 9 Dec 2010 19:16:38 +0000 (21:16 +0200)]
vim: implement archive in show view

Signed-off-by: Felipe Contreras <felipe.contreras@gmail.com>
13 years agovim: refactor tagging stuff
Felipe Contreras [Thu, 9 Dec 2010 18:58:03 +0000 (20:58 +0200)]
vim: refactor tagging stuff

Signed-off-by: Felipe Contreras <felipe.contreras@gmail.com>
13 years agoTODO: Add item for remembing files detected as not email.
Carl Worth [Mon, 25 Apr 2011 22:27:02 +0000 (15:27 -0700)]
TODO: Add item for remembing files detected as not email.

As recently discussed on the mailing list. See, for example:

id:"874o5m802d.fsf@servo.factory.finestructure.net"

13 years agotest: Mark the search-insufficient-from-quoting tests to expect success
Carl Worth [Mon, 25 Apr 2011 22:16:49 +0000 (15:16 -0700)]
test: Mark the search-insufficient-from-quoting tests to expect success

Theses were expected failures only due to a bug in GMime (with
versions of GMime before 2.4.18). As of GMime version 2.4.18 this bug
is fixed and these tests now pass.

13 years agoMerge remote-tracking branch 'amdragon/eager-metadata-v4'
Carl Worth [Mon, 25 Apr 2011 21:26:42 +0000 (14:26 -0700)]
Merge remote-tracking branch 'amdragon/eager-metadata-v4'

13 years agouse custom-face-edit value-type in notmuch-search-line-faces
Jameson Graef Rollins [Sun, 27 Mar 2011 05:16:32 +0000 (22:16 -0700)]
use custom-face-edit value-type in notmuch-search-line-faces

This enables the proper face customization UI for
notmuch-search-line-faces.

13 years agoFixup string list author
Austin Clements [Mon, 21 Mar 2011 06:33:55 +0000 (02:33 -0400)]
Fixup string list author

13 years agoAdd the tag list to the unified message metadata pass.
Austin Clements [Thu, 9 Dec 2010 06:54:34 +0000 (01:54 -0500)]
Add the tag list to the unified message metadata pass.

Now each caller of notmuch_message_get_tags only gets a new iterator,
instead of a whole new list.  In principle this could cause problems
with iterating while modifying tags, but through the magic of talloc
references, we keep the old tag list alive even after the cache in the
message object is invalidated.

This reduces my index search from the 3.102 seconds before the unified
metadata pass to 1.811 seconds (1.7X faster).  Combined with the
thread search optimization in b3caef1f0659dac8183441357c8fee500a940889,
that makes this query 2.5X faster than when I started.

13 years agoAdd the file name list to the unified message metadata pass.
Austin Clements [Thu, 9 Dec 2010 03:19:55 +0000 (22:19 -0500)]
Add the file name list to the unified message metadata pass.

Even if the caller never uses the file names, there is little cost to
simply fetching the file name terms.  However, retrieving the full
paths requires additional database work, so the expansion from terms
to full paths is performed lazily.

This also simplifies clearing the filename cache, since that's now
handled by the generic metadata cache code.

This further reduces my inbox search from 3.102 seconds before the
unified metadata pass to 2.206 seconds (1.4X faster).

13 years agoAdd a generic function to get a list of terms with some prefix.
Austin Clements [Thu, 9 Dec 2010 05:32:35 +0000 (00:32 -0500)]
Add a generic function to get a list of terms with some prefix.

Replace _notmuch_convert_tags with this and simplify
_create_filenames_for_terms_with_prefix.  This will also come in handy
shortly to get the message file name list.

13 years agoImplement an internal generic string list and use it.
Austin Clements [Thu, 9 Dec 2010 00:26:05 +0000 (19:26 -0500)]
Implement an internal generic string list and use it.

This replaces the guts of the filename list and tag list, making those
interfaces simple iterators over the generic string list.  The
directory, message filename, and tags-related code now build generic
string lists and then wraps them in specific iterators.  The real wins
come in later patches, when we use these for even more generic
functionality.

As a nice side-effect, this also eliminates the annoying dependency on
GList in the tag list.

13 years agoUse a single unified pass to fetch scalar message metadata.
Austin Clements [Sat, 27 Nov 2010 04:34:29 +0000 (23:34 -0500)]
Use a single unified pass to fetch scalar message metadata.

This performs a single pass over a message's term list to fetch the
thread ID, message ID, and reply-to, rather than requiring a pass for
each.  Xapian decompresses the term list anew for each iteration, so
this reduces the amount of time spent decompressing message metadata.

This reduces my inbox search from 3.102 seconds to 2.555 seconds (1.2X
faster).

13 years agopython: Remove completed TODO item
James Vasile [Wed, 16 Mar 2011 09:13:26 +0000 (10:13 +0100)]
python: Remove completed TODO item

Really just a left-over TODO item in the code, nothing spectacular to
see here.

Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
13 years agoemacs: Define notmuch-search-process-filter-data before first use.
Carl Worth [Fri, 11 Mar 2011 01:59:53 +0000 (17:59 -0800)]
emacs: Define notmuch-search-process-filter-data before first use.

To avoid a wraning about a reference to a free variable when compiling.

13 years agoemacs: Don't drop error messages from "notmuch search"
Carl Worth [Fri, 11 Mar 2011 00:53:46 +0000 (16:53 -0800)]
emacs: Don't drop error messages from "notmuch search"

With the previous commit, unexpected output before or between search results
would be displayed. However, trailing junk from the "notmuch search" output
would still be silently swallowed.

The most common case for an error message from "notmuch search" would be
an invalid command-line, and in that case, there would be no search results
and the trailing error message would get swallowed.

We fix the process sentinel to check for leftover data and add it to the
final buffer. We also add a test case to ensure this works.

13 years agoemacs: Fix notmuch-search-process-filter to handle incomplete lines
Carl Worth [Thu, 10 Mar 2011 23:29:24 +0000 (15:29 -0800)]
emacs: Fix notmuch-search-process-filter to handle incomplete lines

This fixes the recently-added emacs-large-search-buffer test. This is
as simple as saving any trailing input and then pre-prepending it on
the next call.

MAny thanks to Thomas Schwinge <thomas@schwinge.name> for tracking
down this problem and contributing a preliminary version of this fix.

13 years agoemacs: Display any unexpected output from notmuch search
Carl Worth [Thu, 10 Mar 2011 23:18:40 +0000 (15:18 -0800)]
emacs: Display any unexpected output from notmuch search

Rather than silently swallowing unexpected output, the emacs interface will now
display it. This will allow error messages to actually arrive at the emacs
interface (though not in an especially pretty way). This also allows for easier
investigation of the inadvertent swallowing of search results that span page
boundaries (as demonstrated by the recent added emacs-large-search-buffer test).

The page-boundary bug has been present since a commit from 2009-11-24:
93af7b574598637c2766dd1f8ef343962c9a8efb

Many thanks to Thomas Schwinge for tracking that bug down and
contributing the test for it.

13 years agoRename/rewrite the new emacs-forgetfulness test (to emacs-large-search-buffer)
Carl Worth [Thu, 10 Mar 2011 21:22:04 +0000 (13:22 -0800)]
Rename/rewrite the new emacs-forgetfulness test (to emacs-large-search-buffer)

The new name is more descriptive of the bug being tested. Also, the test
is rewritten slightly so that it's much more plain to see how the bug
manifests itself, (that messages are droped from the emacs result at
regular intervals). Primarily, this is by collapsing the large blobs
used to inflate the message subjects.

13 years agoNew test: Emacs' forgetfulness.
Thomas Schwinge [Wed, 2 Feb 2011 23:56:38 +0000 (00:56 +0100)]
New test: Emacs' forgetfulness.

Signed-off-by: Thomas Schwinge <thomas@schwinge.name>
13 years agonew: Update comments for add_files_recursive
Carl Worth [Thu, 10 Mar 2011 19:56:16 +0000 (11:56 -0800)]
new: Update comments for add_files_recursive

The most recent commit optimized the implementation of this
function. This commit simply updates the relevant comments to match
the new implementation.

13 years agonew: read db_files and db_subdirs only if mtime changed
Karel Zak [Fri, 4 Feb 2011 21:44:31 +0000 (22:44 +0100)]
new: read db_files and db_subdirs only if mtime changed

The db_files and db_subdirs are unnecessary for unchanged directories.

maildir with 10000 e-mails:

old version:
$ time ./notmuch new
No new mail.

real    0m0.053s
user    0m0.028s
sys     0m0.026s

new version:
$ time ./notmuch new
No new mail.

real    0m0.032s
user    0m0.009s
sys     0m0.023s

Signed-off-by: Karel Zak <kzak@redhat.com>
Reviewed-by: Austin Clements <amdragon@mit.edu>
Looks good (faster than, but provably equivalent to the original code!
notmuch_directory_get_child_* are side-effect free,
db_files/db_subdirs aren't used between where they were set in the old
code and where they are set in the new code, and db_files/db_subdirs
are initialized to NULL when declared).

Another timing data point:
Old code: ./notmuch new  0.77s user 0.28s system 99% cpu 1.051 total
New code: ./notmuch new  0.09s user 0.27s system 98% cpu 0.368 total

13 years agobuild: Save configure options and re-use them for automatic runs of configure
Carl Worth [Thu, 10 Mar 2011 19:30:06 +0000 (11:30 -0800)]
build: Save configure options and re-use them for automatic runs of configure

This supports the case of a user running "configure --prefix=/foo" then later
updating the soruce (including the configure script) and re-running make.

In this case, the make invocation will re-run configure. Before this change,
this run of configure would lose the user's carefully chosen prefix. This
is now fixed so that configrue is re-run with the user's options.

13 years agobuild: Fix a plain "make" to automatically run configure.
Carl Worth [Thu, 10 Mar 2011 19:29:13 +0000 (11:29 -0800)]
build: Fix a plain "make" to automatically run configure.

The recent change to support non-source-directory builds broke this case.

13 years agoMerge remote branch 'amdragon/search-perf-3'
Carl Worth [Thu, 10 Mar 2011 19:05:49 +0000 (11:05 -0800)]
Merge remote branch 'amdragon/search-perf-3'

13 years agobuild: Add support for non-source-directory builds.
Carl Worth [Wed, 9 Mar 2011 23:02:42 +0000 (15:02 -0800)]
build: Add support for non-source-directory builds.

Such as:

     mkdir build
     cd build
     ../configure
     make

This is implemented by having the configure script set a srcdir
variable in Makefile.config, and then sprinkling $(srcdir) into
various make rules. We also use vpath directives to convince GNU make
to find the source files from the original source directory.

13 years agoTODO: Add item for the build system to support a non-source-dir build
Carl Worth [Sun, 30 Jan 2011 19:30:07 +0000 (05:30 +1000)]
TODO: Add item for the build system to support a non-source-dir build

This would be a handy feature that many people would expect to just
work.

13 years agojson: Fix search result with no matches to be a valid json object.
Carl Worth [Sun, 30 Jan 2011 19:26:04 +0000 (05:26 +1000)]
json: Fix search result with no matches to be a valid json object.

In the original json code, search matching nothing would return a
valid, empty json array (that is, "[]"). I broke this in commit
6dcb7592e32ed5140ea0c0357ce78d6a37af6066 when adding support for
--output=threads|messages|tags. This time, while fixing the bug also
add a test to the test suite to help avoid future regressions.

13 years agotest: Rename and clarify the search-lwn test
Carl Worth [Fri, 28 Jan 2011 20:21:25 +0000 (06:21 +1000)]
test: Rename and clarify the search-lwn test

Now that we understand the bug here, we rename this test to
search-insufficient-from-quoting to clarify the bug being exercised,
(which occurs when the From: line contains an unquoted '.' character).

We also mark these tests as expected failures until the bug gets fixed.

13 years agosearch: Move lwn tests into their own file.
Carl Worth [Fri, 28 Jan 2011 19:39:21 +0000 (05:39 +1000)]
search: Move lwn tests into their own file.

Since it's much easier to debug and fix these if they can be run on
their own.

13 years agoAdd a few tests for searching LWN emails.
Thomas Schwinge [Thu, 27 Jan 2011 10:31:49 +0000 (03:31 -0700)]
Add a few tests for searching LWN emails.

These tests should pass -- but they currently don't.

Signed-off-by: Thomas Schwinge <thomas@schwinge.name>
13 years agoupdate for go-release-2011-02-01:
Sebastien Binet [Thu, 3 Feb 2011 08:36:58 +0000 (09:36 +0100)]
update for go-release-2011-02-01:
  * M bindings/go/cmds/notmuch-addrlookup.go
    log.Exitf -> log.Fatalf

13 years agoruby: Add generated files to .gitignore
Ali Polatel [Wed, 2 Feb 2011 23:40:38 +0000 (01:40 +0200)]
ruby: Add generated files to .gitignore

13 years agoSimplify _notmuch_doc_id_set_init interface.
Austin Clements [Mon, 31 Jan 2011 02:01:56 +0000 (21:01 -0500)]
Simplify _notmuch_doc_id_set_init interface.

Don't require the caller of _notmuch_doc_id_set_init to pass in a
correct bound; instead compute it from the array.  This simplifies the
caller and makes this interface easier to use correctly.

13 years agoRemove code repetition in the doc ID bitmap code.
Austin Clements [Mon, 31 Jan 2011 01:58:27 +0000 (20:58 -0500)]
Remove code repetition in the doc ID bitmap code.

Remove the repeated "sizeof (doc_ids->bitmap[0])" that bothered cworth
by instead defining macros to compute the word and bit offset of a
given bit in the doc ID set bitmap.

13 years agoClarify usage of `additional_headers' in test/test-lib.sh:generate_message.
Thomas Schwinge [Thu, 27 Jan 2011 09:17:21 +0000 (02:17 -0700)]
Clarify usage of `additional_headers' in test/test-lib.sh:generate_message.

Signed-off-by: Thomas Schwinge <thomas@schwinge.name>
13 years agoemacs: send notmuch-query stderr to /dev/null
Jameson Rollins [Wed, 26 Jan 2011 20:57:50 +0000 (12:57 -0800)]
emacs: send notmuch-query stderr to /dev/null

The call-process to notmuch in notmuch-query.el was previously sending
stderr into the output buffer.  This means that if there is any stderr
the JSON parsing breaks.  Unfortunately call-process does not support
sending stderr to a separate buffer or to the minibuffer [0], but it
does support sending it to /dev/null.  So we do that here instead.

[0] a bug was filed against emacs (#7842)

13 years agonew: Print progress estimates only when we have sufficient information
Michal Sojka [Wed, 26 Jan 2011 13:06:57 +0000 (14:06 +0100)]
new: Print progress estimates only when we have sufficient information

Without this patch, it might happen that the remaining time or processing
rate were calculated just after start where nothing was processed yet.
This resulted into division by a very small number (or zero) and the
printed information was of little value.

Instead of printing nonsenses we print only that the operation is in
progress. The estimates will be printed later, after there is enough data.

13 years agoMakefile: Quote variables used as filenames in shell commands
Carl Worth [Wed, 26 Jan 2011 13:29:15 +0000 (23:29 +1000)]
Makefile: Quote variables used as filenames in shell commands

This allows support for filenames with spaces in them.

13 years agoconfigure: Drop global setting of IFS (without space in it).
Michal Sojka [Wed, 26 Jan 2011 13:13:21 +0000 (23:13 +1000)]
configure: Drop global setting of IFS (without space in it).

This was originally intended to help support filenames with spaces in
them, but this actually breaks things when someone sets a command with
a space in it, (such as CC="ccache cc").

Instead, we now only set a custom IFS when acting on the
newline-separated list of files from /sbin/ldconfig.

13 years agoemacs: View the output of pipe command when it fails
Michal Sojka [Sun, 23 Jan 2011 10:45:50 +0000 (11:45 +0100)]
emacs: View the output of pipe command when it fails

Previously, the user didn't know whether the pipe command succeeded or
not. It was only possible to find it out by manually inspecting
the work done (or not done) by the command or by manually switching to
*notmuch-pipe* buffer and determine it from command output. For this
the user had to first find the text corresponding to the last run of
pipe command as the buffer accumulated the output from all pipe commands.

This patch changes the following. The *notmuch-pipe* buffer is erased
before every pipe command so it contains only the output from the last
command. Additionally, when the command failed, the *notmuch-pipe* buffer
is shown and an error message is displayed.
with the output of pipe command.

13 years agotest: Make it easier to resolve conflicts when adding new tests
Michal Sojka [Sun, 23 Jan 2011 11:12:24 +0000 (12:12 +0100)]
test: Make it easier to resolve conflicts when adding new tests

Currently, there are two places in the test framework that contain very
long list on a single line. Whenever a test is added (or changed) in
several branches and these branches are merged, it results in conflict
which is hard to resolve because one has to go through the whole long
line to find where the conflict is.

This patch splits these long lists to several lines so that the
conflicts are easier to resolve.

13 years agoFix installation of zsh completion
Michal Sojka [Sun, 23 Jan 2011 20:57:07 +0000 (21:57 +0100)]
Fix installation of zsh completion

13 years agoconfigure: add options to disable emacs/zsh/bash and choose install dir.
Cédric Cabessa [Sun, 23 Jan 2011 13:33:43 +0000 (14:33 +0100)]
configure: add options to disable emacs/zsh/bash and choose install dir.

add --bashcompletiondir and --zshcompletiondir (like --emacslispdir) to choose
installation dir for bash/zsh completion files

Make some features optional:
  --without-emacs / --with-emacs=no do not install lisp file
  --without-bash-completion / --with-bash-completion=no  do not install bash
files
  --without-zsh-completion / --with-zsh-completion=no do not install zsh files
By default, everything is enabled. You can reenable something with
  --with-feature=yes

13 years agonew: Enhance progress reporting
Michal Sojka [Fri, 21 Jan 2011 09:59:37 +0000 (10:59 +0100)]
new: Enhance progress reporting

notmuch new reports progress only during the "first" phase when the
files on disk are traversed and indexed. After this phase, other
operations like rename detection and maildir flags synchronization are
performed, but the user is not informed about them. Since these
operations can take significant time, we want to inform the user about
them.

This patch enhances the progress reporting facility that was already
present. The timer that triggers reporting is not stopped after the
first phase but continues to run until all operations are finished. The
rename detection and maildir flag synchronization are enhanced to report
their progress.

13 years agonew: Add all initial tags at once
Michal Sojka [Fri, 21 Jan 2011 09:59:36 +0000 (10:59 +0100)]
new: Add all initial tags at once

If there are several tags applied to the new messages, it is beneficial
to store them to the database at one, because it saves some time,
especially when the notmuch new is run for the first time.

This patch decreased the time for initial import from 1h 35m to 1h 14m.

13 years agoDo not defer maildir flag synchronization for new messages
Austin Clements [Wed, 26 Jan 2011 11:52:54 +0000 (21:52 +1000)]
Do not defer maildir flag synchronization for new messages

This is a simplified version of a patch originally by Michal Sojka
<sojkam1@fel.cvut.cz> which is designed to have the same performance
benefits. Michal said the following:

  When notmuch new is run for the first time, it is not necessary to
  defer maildir flags synchronization to later because we already know
  that no files will be removed.

  Performing the maildinr flag synchronization immediately after the
  message is added to the database has the advantage that the message
  is likely hot in the disk cache so the synchronization is faster.
  Additionally, we also save one database query for each message,
  which must be performed when the operation is deferred.

  Without this patch, the first notmuch new of 200k messages (3 GB)
  took 1h and 46m out of which 20m was maildir flags
  synchronization. With this patch, the whole operation took only 1h
  and 36m.

Unlike Michal's patch, this version does the deferral for any new
message, rather than doing it only on the first run of "notmuch new".

13 years agovim: Get user email address from notmuch config file.
Peter John Hartman [Wed, 26 Jan 2011 06:57:12 +0000 (16:57 +1000)]
vim: Get user email address from notmuch config file.

Here's a bitty patch to the vim plugin; it now calculates the primary email
of the user based on a call to notmuch config.  There's still a lot of work
that needs to get done on notmuch.vim, e.g., the ability to have multiple
emails/accounts.

13 years agolib: Save and restore term position in message while indexing.
Carl Worth [Wed, 26 Jan 2011 05:53:14 +0000 (15:53 +1000)]
lib: Save and restore term position in message while indexing.

This fixes the recently addead search-position-overlap bug as
demonstrated in the test of the same name.

13 years agoAdd test demonstrating a position overlap bug.
Carl Worth [Wed, 26 Jan 2011 05:19:31 +0000 (15:19 +1000)]
Add test demonstrating a position overlap bug.

Currently, whenever we call index_terms multiple times for a single
field, the term generator is being reset to position 0 each time. This
means that with text such as:

To: a@b.c, x@y.z

one can get a bogus match by searching for:

To: a@y.c

Thanks to Mark Anderson for reporting the bug, (and providing a nice,
minimal test case that inspired what is used here).

13 years agonotmuch search: Clean up some memory leaks during search loop.
Carl Worth [Tue, 25 Jan 2011 13:40:35 +0000 (23:40 +1000)]
notmuch search: Clean up some memory leaks during search loop.

With talloc, we were already freeing all memory by the time we exited
the loop, but that didn't help with excess use of memory inside the
loop, (which was mostly from tallocing some objects with the incorrect
parent).

Thanks to Andrew Tridgell for sitting next to me and teaching me to
use talloc_report_full to find these leaks.

13 years agoMigrate to goconfig pkg
Sebastien Binet [Thu, 16 Dec 2010 22:13:09 +0000 (23:13 +0100)]
Migrate to goconfig pkg

13 years agobindings/go: Add a todo file
Sebastien Binet [Wed, 10 Nov 2010 17:27:04 +0000 (18:27 +0100)]
bindings/go: Add a todo file

13 years agoA minor, cosmetic change
Sebastien Binet [Wed, 10 Nov 2010 17:24:14 +0000 (18:24 +0100)]
A minor, cosmetic change

Just trying to keep the line lengths in check.

13 years agoInitial import of Go bindings for notmuch
Sebastien Binet [Wed, 10 Nov 2010 17:18:44 +0000 (18:18 +0100)]
Initial import of Go bindings for notmuch

13 years agoTODO: Note that folder: has now been implemented.
Carl Worth [Thu, 20 Jan 2011 01:04:10 +0000 (17:04 -0800)]
TODO: Note that folder: has now been implemented.

This feature has been mentioned in the TODO file (more than once) for
quite some time now.

13 years agoruby: Add wrapper for message_get_filenames
Ali Polatel [Mon, 10 Jan 2011 14:03:31 +0000 (16:03 +0200)]
ruby: Add wrapper for message_get_filenames

13 years agoruby: Add wrappers for maildir sync. interface
Ali Polatel [Mon, 10 Jan 2011 14:02:43 +0000 (16:02 +0200)]
ruby: Add wrappers for maildir sync. interface

New wrappers:
notmuch_message_maildir_flags_to_tags(): MESSAGE.maildir_flags_to_tags
notmuch_message_tags_to_maildir_flags(): MESSAGE.tags_to_maildir_flags

13 years agoruby: Add wrappers for query_get_s{ort,tring}
Ali Polatel [Mon, 10 Jan 2011 13:59:18 +0000 (15:59 +0200)]
ruby: Add wrappers for query_get_s{ort,tring}

New wrappers:
notmuch_query_get_sort(): QUERY.sort
notmuch_query_get_query_string(): QUERY.to_s

13 years agoAdd documentation for the recently added folder-based searching.
Carl Worth [Sat, 15 Jan 2011 22:18:45 +0000 (14:18 -0800)]
Add documentation for the recently added folder-based searching.

Augmenting both the builtin help as well as the manual page.

13 years agoAdd support for folder-based searching.
Carl Worth [Sat, 15 Jan 2011 22:09:04 +0000 (14:09 -0800)]
Add support for folder-based searching.

A new "folder:" prefix in the query string can now be used to match
the directories in which mail files are stored.

The addition of this feature causes the recently added
search-by-folder tests to now pass.

13 years agotest: Add new tests for folder-based searching.
Carl Worth [Sat, 15 Jan 2011 22:05:10 +0000 (14:05 -0800)]
test: Add new tests for folder-based searching.

This is a new feature which is not implemente yet, so these tests mostly
fail currently. A subsequent commit will add the feature and cause these
tests to start passing.

These tests verify that we can search for containing folders of mail files
by word or by phrase and that the search terms are updated correctly when
directories are renamed.

13 years agoFix misspelling in a comment.
Carl Worth [Sat, 15 Jan 2011 22:04:38 +0000 (14:04 -0800)]
Fix misspelling in a comment.

Just a simple change I noticed in passing.

13 years agotest/basic: Ignore new files ending in ~
Carl Worth [Sat, 15 Jan 2011 22:03:33 +0000 (14:03 -0800)]
test/basic: Ignore new files ending in ~

We don't need to complain that temporary editor backups are not added
to the list of tests to be run.

13 years agoTighten up a memory allocation.
Carl Worth [Sat, 15 Jan 2011 22:01:43 +0000 (14:01 -0800)]
Tighten up a memory allocation.

Using the local talloc context ensures that the memory we are using
here will be freed shortly, (rather than hanging on for a long time
with the notmuch database object).

13 years agoCorrect some minor typos in a comment
Carl Worth [Thu, 13 Jan 2011 01:04:10 +0000 (17:04 -0800)]
Correct some minor typos in a comment

Nothing too important here. Just some misspellings I noticed while reading
nearby code.

13 years agopython: Update metainformation to point to new URL and version number
Sebastian Spaeth [Thu, 13 Jan 2011 11:35:59 +0000 (12:35 +0100)]
python: Update metainformation to point to new URL and version number

Convert the meta information to point to the notmuchmail.org repository, rather
than the old cnotmuch location. I will delete the "cnotmuch" package
from http://pypi.python.org/pypi/cnotmuch and create a new "notmuch"
package there that contains the current versions.

Also bump the version number to 0.4. I will need to upgrade the API
first before I can release the 0.5 of the bindings, there are still some
methods missing.

Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
14 years agoOptimize thread search using matched docid sets.
Austin Clements [Wed, 17 Nov 2010 19:28:26 +0000 (14:28 -0500)]
Optimize thread search using matched docid sets.

This reduces thread search's 1+2t Xapian queries (where t is the
number of matched threads) to 1+t queries and constructs exactly one
notmuch_message_t for each message instead of 2 to 3.
notmuch_query_search_threads eagerly fetches the docids of all
messages matching the user query instead of lazily constructing
message objects and fetching thread ID's from term lists.
_notmuch_thread_create takes a seed docid and the set of all matched
docids and uses a single Xapian query to expand this docid to its
containing thread, using the matched docid set to determine which
messages in the thread match the user query instead of using a second
Xapian query.

This reduces the amount of time required to load my inbox from 4.523
seconds to 3.025 seconds (1.5X faster).

14 years agoUpdate test suite for 7 tests that were recently fixed.
Carl Worth [Wed, 8 Dec 2010 00:35:47 +0000 (16:35 -0800)]
Update test suite for 7 tests that were recently fixed.

These tests had been broken, but were fixed by the preceding commit,
so update the test suite to print PASS rather than FIXED for these.