notmuch.git
14 years agoAdd functions notmuch-show-get-(bcc, cc, date, from, subject, to).
David Bremner [Fri, 12 Feb 2010 02:44:10 +0000 (22:44 -0400)]
Add functions notmuch-show-get-(bcc, cc, date, from, subject, to).

Return the corresponding header field for the current message as a
string.  These are thin wrappers around notmuch-show-get-header, which
means they each cause a full parse of the RFC822 header. The main idea
is to fix an api.

14 years agonotmuch-show-get-header: new function; return alist of parsed header fields.
David Bremner [Fri, 12 Feb 2010 02:34:31 +0000 (22:34 -0400)]
notmuch-show-get-header: new function; return alist of parsed header fields.

This function parses the displayed message to recover header
fields. It uses mailheader.el to do the actual header parsing, after
preprocessing to remove indentation.  It relies on the variables
notmuch-show-message-begin-regexp, notmuch-show-header-begin-regexp,
and notmuch-show-message-end-regexp.

14 years agonotmuch.el: Delete some trailing whitespace.
Carl Worth [Sat, 20 Feb 2010 20:11:31 +0000 (12:11 -0800)]
notmuch.el: Delete some trailing whitespace.

I'm not sure when this managed to creep in, but we don't want it.

14 years agonotmuch.el: Fix bug from message with ':' in the From address.
Carl Worth [Fri, 12 Feb 2010 00:19:37 +0000 (16:19 -0800)]
notmuch.el: Fix bug from message with ':' in the From address.

Eric reported that a particular thread was non-functional in the
notmuch-search mode in the emacs client. It was easy enough to trace
the bug down to a broken regular expression (using ':' instead of
';'). The bug would be triggered by a message with ':' in the
From address.

This is something I hope to add to the test suite as soon as we have
support for testing the emacs interface there.

14 years agonotmuch.el: Handle attached images via an external viewer.
Carl Worth [Wed, 10 Feb 2010 20:40:47 +0000 (12:40 -0800)]
notmuch.el: Handle attached images via an external viewer.

We temporarily override the mm-inline-media-tests variable so that the
only parts inserted into the temporary buffer (and lost) are those
parts that the user has already seen in the notmuch-show buffer.

Anything else, (such as images), will now be left to be handled via
mailcap, just like other attachment types.

14 years agonotmuch.el: Fix indentation.
Carl Worth [Wed, 10 Feb 2010 20:40:18 +0000 (12:40 -0800)]
notmuch.el: Fix indentation.

This line was indented incorrectly which can be confusing.

14 years agonotmuch.el: Avoid infinite loop marking up message with no parts.
Carl Worth [Wed, 10 Feb 2010 19:40:56 +0000 (11:40 -0800)]
notmuch.el: Avoid infinite loop marking up message with no parts.

The infinite loop was triggered by a message consisting of a single
attachment within the body, (and no "part") tags.

We need to do things in response to this bug (beyond this specific
fix):

1. Create a test suite that exercises our emacs frontend so that bugs
   like this do not come back to haunt us after we fix them once.

2. Switch from our ad-hoc regexp based search of message-part delimeters
   to known-good code for parsing a structured document, (for example,
   the outstanding JSON patches).

14 years agoReintroduce HTML inlining, with a much needed optimization
Alexander Botero-Lowry [Wed, 10 Feb 2010 01:55:21 +0000 (17:55 -0800)]
Reintroduce HTML inlining, with a much needed optimization

Now instead of requiring every single message be parsed, we now check
the Content-type in the parsed headers and only do HTML inlining if it's
text/html

14 years agoFix printf for when uint64_t != unsigned long long int
Carl Worth [Tue, 9 Feb 2010 19:09:30 +0000 (11:09 -0800)]
Fix printf for when uint64_t != unsigned long long int

Thanks to Michal Sojka <sojkam1@fel.cvut.cz> for pointing out the
correct fix, which I verified in the freely-available WG14/N1124 draft
(from the C99 working group) which is available here:

http://www.open-std.org/JTC1/SC22/wg14/www/docs/n1124.pdf

14 years agoSwitch from random to sequential thread identifiers.
Carl Worth [Mon, 8 Feb 2010 19:33:33 +0000 (11:33 -0800)]
Switch from random to sequential thread identifiers.

The sequential identifiers have the advantage of being guaranteed to
be unique (until we overflow a 64-bit unsigned integer), and also take
up half as much space in the "notmuch search" output (16 columns
rather than 32).

This change also has the side effect of fixing a bug where notmuch
could block on /dev/random at startup (waiting for some entropy to
appear). This bug was hit hard by the test suite, (which could easily
exhaust the available entropy on common systems---resulting in large
delays of the test suite).

14 years agoTODO: Add some recently discussed items.
Carl Worth [Tue, 9 Feb 2010 19:02:18 +0000 (11:02 -0800)]
TODO: Add some recently discussed items.

These were collected either from the mailing list of from IRC
conversations. The good ideas probably aren't mine, and the mistakes
probably are.

14 years agoAllow folders with no messages to be elided from list
Keith Packard [Sun, 27 Dec 2009 00:34:18 +0000 (16:34 -0800)]
Allow folders with no messages to be elided from list

This makes it easier to see folders with messages.
Eliding empty folders is togged with the 'e' binding.

Signed-off-by: Keith Packard <keithp@keithp.com>
14 years agoLook at whitespace to separate folder name from count
Keith Packard [Sun, 27 Dec 2009 00:34:17 +0000 (16:34 -0800)]
Look at whitespace to separate folder name from count

This allows folder names to contain any non-blank characters

Signed-off-by: Keith Packard <keithp@keithp.com>
14 years agoAdd 'm' and ' ' bindings to notmuch-folder view
Keith Packard [Sun, 27 Dec 2009 00:34:16 +0000 (16:34 -0800)]
Add 'm' and ' ' bindings to notmuch-folder view

This allows the user to compose new mail from the folder view, and
also to use <space> to show the current folder.

Signed-off-by: Keith Packard <keithp@keithp.com>
14 years agoemacs: Use font-lock-comment-face to highlight citation button
Kan-Ru Chen [Mon, 14 Dec 2009 05:41:35 +0000 (13:41 +0800)]
emacs: Use font-lock-comment-face to highlight citation button

Signed-off-by: Kan-Ru Chen <kanru@kanru.info>
14 years agonotmuch.el: show some of citation even when hiding.
David Bremner [Fri, 25 Dec 2009 20:09:08 +0000 (16:09 -0400)]
notmuch.el: show some of citation even when hiding.

- rename notmuch-show-citation-lines-min to n-s-c-l-prefix
- call forward-line with the appropriate parameter to adjust
  region to be hidden.
- change citation button text so that it makes (some) sense when citation is shown

Reviewed-by: Kan-Ru Chen <kanru@kanru.info>
14 years agonotmuch.el: Refactor citation markup. Variables for minimum size, button text.
David Bremner [Fri, 25 Dec 2009 20:09:07 +0000 (16:09 -0400)]
notmuch.el: Refactor citation markup. Variables for minimum size, button text.

This is a fairly intrusive rewrite.

- I pulled the common code for the signature and citation case out
  into a separate function. This is not so much shorter, but I think it
  will be easier to maintain.

- I replaced the sequence of (looking-at blah) (forward-line)  with a single
  re-search-forward per citation.

New variables

- notmuch-show-signature-button-format, notmuch-show-citation-button-format
  Allow customization of button text.

- notmuch-show-citation-lines-min
  Do not buttonize citations below the given threshold.

Reviewed-by: Kan-Ru Chen <kanru@kanru.info>
14 years agonotmuch.h: Fix a couple of typos in the documentation.
Carl Worth [Sat, 6 Feb 2010 01:31:40 +0000 (17:31 -0800)]
notmuch.h: Fix a couple of typos in the documentation.

Obviously, the spell-checker isn't able to catch every mistake
I make.

14 years agonotmuch restore: Don't waste time with messages with unchanged tags.
Carl Worth [Sat, 6 Feb 2010 01:29:56 +0000 (17:29 -0800)]
notmuch restore: Don't waste time with messages with unchanged tags.

It's a simple optimization to look at a message and check that the
existing tags are actually different than the tags we are setting
before we do anything.

For my mail store this takes a "notmuch restore" that does nothing
from about 10 minutes down to 1 minute, so there's a significant
speedup here.

14 years agonotmuch-test: Add basic tests for "notmuch dump" and "notmuch restore"
Carl Worth [Sat, 6 Feb 2010 01:28:25 +0000 (17:28 -0800)]
notmuch-test: Add basic tests for "notmuch dump" and "notmuch restore"

Our test-suite coverage keeps improving, (slowly).

14 years agonotmuch-test: Fix misalignment in output.
Carl Worth [Sat, 6 Feb 2010 01:27:22 +0000 (17:27 -0800)]
notmuch-test: Fix misalignment in output.

Clean output is just so much easier to read (and validate).

14 years agonotmuch.el: Use emacs built-in forward-button and backward-button
Kan-Ru Chen [Tue, 1 Dec 2009 06:13:53 +0000 (14:13 +0800)]
notmuch.el: Use emacs built-in forward-button and backward-button

There are built-ins, so why not use them?

Signed-off-by: Kan-Ru Chen <kanru@kanru.info>
14 years agoUse forward-line instead of next-line
Carl Worth [Fri, 5 Feb 2010 19:14:25 +0000 (11:14 -0800)]
Use forward-line instead of next-line

We do this all the time, but at least emacs is kind enough to remind us,
(when compiling), that next-line is only intended for interactive use,
and we should use forward-line inside of lisp code.

14 years agonotmuch.el: Add missing documentation for the new 'h' keybinding.
Carl Worth [Fri, 5 Feb 2010 19:30:35 +0000 (11:30 -0800)]
notmuch.el: Add missing documentation for the new 'h' keybinding.

Without this, our help screen displayed 'h' with no description of
what it does.

14 years agonotmuch.el: Add keybinding to toggle display of message body and headers.
Kan-Ru Chen [Tue, 1 Dec 2009 06:13:51 +0000 (14:13 +0800)]
notmuch.el: Add keybinding to toggle display of message body and headers.

I really missed this feature. Added notmuch-show-toggle-current-body and
notmuch-show-toggle-current-header and bind them to 'b' and 'h'.

Signed-off-by: Kan-Ru Chen <kanru@kanru.info>
14 years agonotmuch-test: Add test to verify that uuencoded data is not indexed.
Carl Worth [Fri, 5 Feb 2010 18:24:51 +0000 (10:24 -0800)]
notmuch-test: Add test to verify that uuencoded data is not indexed.

As recently promised, no new features are being accepted to notmuch
without corresponding new tests for the test suite.

14 years agonotmuch-test: Add several tests of "notmuch search"
Carl Worth [Fri, 5 Feb 2010 18:00:04 +0000 (10:00 -0800)]
notmuch-test: Add several tests of "notmuch search"

These tests were surprisingly simple to write---not much code at all
and most of them worked the first time even with hand-prepared
versions of the expected output.

14 years agonotmuch-test: Add a new add_message function.
Carl Worth [Fri, 5 Feb 2010 01:34:48 +0000 (17:34 -0800)]
notmuch-test: Add a new add_message function.

The previous generate_message function is what's needed when testing
"notmuch new". But after that, we never want to generate a message
without also adding it to the index. So create a new add_message
function with this convenience.

14 years agoEliminate some useless gobject boilerplate.
Carl Worth [Fri, 5 Feb 2010 01:26:00 +0000 (17:26 -0800)]
Eliminate some useless gobject boilerplate.

If we had external users of this filter then they might expect some of
these macros to exist. But since this is just internal, that's just
unneeded noise.

14 years agonotmuch new: Don't index uuencoded data.
Carl Worth [Fri, 5 Feb 2010 01:08:11 +0000 (17:08 -0800)]
notmuch new: Don't index uuencoded data.

With modern MIME attachments, we're already avoiding indexing the
attachments. But for old-school uuencoded data in the mail, we have
been directly indexing the encoded data as terms, (which is not useful
at all---nobody will ever ytry to search based on the seemingly random
uuencoded data).

Additionally, indexing a modestly large uuencoded file seems to make
Xapian go insane, (consuming *lots* of memory).

We fix both problems by detecting uuencoded content and not performing
any indexing of it.

14 years agonotmuch reply: Rename the mailing_list_munged_reply_to function
Carl Worth [Thu, 4 Feb 2010 20:39:11 +0000 (12:39 -0800)]
notmuch reply: Rename the mailing_list_munged_reply_to function

This function detects whether the address in the Reply-To header
already appears in either To or Cc. So give it a name that reflects
what it does (reply_to_header_is_redundant) rather than the old name
which described one possible use of the function, (as a simple
heuristic for detecting whether a mailing list had applied reply-to
munging).

14 years agonotmuch reply: Prevent GMIME assertion complaints for empty Reply-to header.
Carl Worth [Thu, 4 Feb 2010 20:32:28 +0000 (12:32 -0800)]
notmuch reply: Prevent GMIME assertion complaints for empty Reply-to header.

Apparently, GMime doesn't want to create a valid address list object
for an empty string. That's annoying, but it's easy enough to test for
the empty string and avoid the problem.

14 years agonotmuch reply: Use strstr instead of strcasestr for portability.
Carl Worth [Thu, 4 Feb 2010 20:29:04 +0000 (12:29 -0800)]
notmuch reply: Use strstr instead of strcasestr for portability.

This change was already recommended in a comment in the original
implementation of this patch. If someone really wants to support
un-munging in the case of To: and Reply-To: having the same address
but different case, then they can provide a portable approach for
that.

14 years agonotmuch reply: Fix the support for reply-to un-munging.
Carl Worth [Thu, 4 Feb 2010 20:27:42 +0000 (12:27 -0800)]
notmuch reply: Fix the support for reply-to un-munging.

The condition was using a reversed sense for the test of the return
value of strcasestr, (perhaps confusing it with the usage of strcmp?).

14 years agonotmuch-reply: Add missing whitespace.
Carl Worth [Thu, 4 Feb 2010 20:12:01 +0000 (12:12 -0800)]
notmuch-reply: Add missing whitespace.

Some recently-added functions were to hard for me to read without
the expected whitespace. Fix these.

14 years agonotmuch-test: Test "notmuch reply" with a reply-to-munged mail
Carl Worth [Thu, 4 Feb 2010 20:08:13 +0000 (12:08 -0800)]
notmuch-test: Test "notmuch reply" with a reply-to-munged mail

This is a test for the recently added feature where we detect that the
reply-to address already exists in the To: or Cc: header so will
already be replied to. In this case we want to include the From:
address in our reply, (where, otherwise we would use the Reply-To
address *instead* of the address in the From header).

14 years agonotmuch-reply.c: Handle munged `Reply-To' headers.
Jed Brown [Sat, 28 Nov 2009 23:19:36 +0000 (00:19 +0100)]
notmuch-reply.c: Handle munged `Reply-To' headers.

Some mailing lists engage in the evil practice of changing the Reply-To
header so that replies from all mailers go to the list by default, at
the expense of not responding to the person who actually sent the
message.  When this is detected, we reply to `From' and remove the
duplicate response to the mailing list.  Consider a reply to the
following message.

  From: Some User <some.user@example.com>
  To: Sample users list <sample-users@sample.org>
  Reply-To: Sample users list <sample-users@sample.org>

Prior to this patch, `notmuch reply' produces

  To: Sample users list <sample-users@sample.org>,
      Sample users list <sample-users@sample.org>

and after the patch,

  To: Some User <some.user@example.com>,
      Sample users list <sample-users@sample.org>

Signed-off-by: Jed Brown <jed@59A2.org>
14 years agonotmuch-reply.c: Factor adding recipients into common function
Jed Brown [Sat, 28 Nov 2009 23:19:35 +0000 (00:19 +0100)]
notmuch-reply.c: Factor adding recipients into common function

This code was already duplicated. We move it to a new, shared
add_recipients_from_message function, in preparation for more
sophisticated mailing list logic.

Signed-off-by: Jed Brown <jed@59A2.org>
14 years agonotmuch-test: Test "notmuch reply" with multiple recipients
Carl Worth [Thu, 4 Feb 2010 20:03:30 +0000 (12:03 -0800)]
notmuch-test: Test "notmuch reply" with multiple recipients

The feature tested here is that we reply to both the sender and to
others addresses on the To: line of the original message, but that we
don't reply to our own address.

14 years agonotmuch-test: Add test for "notmuch reply" with a CC header.
Carl Worth [Thu, 4 Feb 2010 19:55:57 +0000 (11:55 -0800)]
notmuch-test: Add test for "notmuch reply" with a CC header.

A simple test to help round out the set of tested features.

14 years agonotmuch-test: Add "notmuch reply" test for reply-to support.
Carl Worth [Thu, 4 Feb 2010 19:52:08 +0000 (11:52 -0800)]
notmuch-test: Add "notmuch reply" test for reply-to support.

This is the standard support of reply-to, (replying to that address
rather than the from address). It has nothing to do with the proposed
feature for extra-clever handling of a mail from a mailing-list that
has munged the reply-to header.

14 years agonotmuch-test: Test "notmuch reply" from alternate address.
Carl Worth [Thu, 4 Feb 2010 19:39:31 +0000 (11:39 -0800)]
notmuch-test: Test "notmuch reply" from alternate address.

When reply to a message addresses to an address configured in the
other_email setting in the configuration file, the reply should use
that address in the From header. Test this.

14 years agonotmuch-test: Add a basic test of "notmuch reply"
Carl Worth [Thu, 4 Feb 2010 19:33:56 +0000 (11:33 -0800)]
notmuch-test: Add a basic test of "notmuch reply"

Simply ensuring that the reply template is formatted as expected.

14 years agonotmuch-test: Eliminate sleeps to speed up test suite run
Carl Worth [Thu, 4 Feb 2010 19:31:01 +0000 (11:31 -0800)]
notmuch-test: Eliminate sleeps to speed up test suite run

We were sleeping merely to ensure that our updates to the mail store
would result in the mtime of the appropriate directories being
updated. We make the test suite run faster by not sleeping, but
instead explicitly updating the mtime of the directory to a future
time with touch.

We're careful to ensure that the time is not merely in the future
compared to the current time, but also later than any previous update
to the same directory mtime.

14 years agonotmuch-test: Allow custom headers when generating messages
Carl Worth [Thu, 4 Feb 2010 16:39:23 +0000 (08:39 -0800)]
notmuch-test: Allow custom headers when generating messages

This provides the control that future tests will need, (for example,
adding a CC field to ensure proper handling with "notmuch reply",
etc.)

14 years agonotmuch-test: Use named-parameters for generate_message function
Carl Worth [Thu, 4 Feb 2010 01:24:01 +0000 (17:24 -0800)]
notmuch-test: Use named-parameters for generate_message function

This makes the test suite bash-specific, but that's not much of
an issue for me, (if somebody else would prefer some other language
then they can rewrite the test suite and maintain it).

The advantage here is that we'll now be able to easily generate
custom messages for testing operations that depend on the message
content, (such as "notmuch reply", etc.).

14 years agoAdd actual testing to notmuch-test
Carl Worth [Wed, 3 Feb 2010 00:24:53 +0000 (16:24 -0800)]
Add actual testing to notmuch-test

We verify that each command creates output exactly as expected (after
ignoring variable output such as the report of how long an operation
took).

14 years agoAdd a very rough beginning of a test suite.
Carl Worth [Sat, 23 Jan 2010 18:36:39 +0000 (07:36 +1300)]
Add a very rough beginning of a test suite.

This notmuch-test script simply runs a few different notmuch operations,
(things that I found were useful while testing the rename-support code).

It's not useful as a test suite yet, since it doesn't actually check
the results of any operation, (the user of the suite has to know what
the results should be and must manually verify them. So there's no
integration with the build system yet, (no "make test" target).

But I didn't want to lose what I had so far, so here it is.

14 years agonotmuch.desktop: Clarify that this is the emacs interface to notmuch.
Carl Worth [Sat, 23 Jan 2010 18:28:19 +0000 (07:28 +1300)]
notmuch.desktop: Clarify that this is the emacs interface to notmuch.

And the name is notmuch not not much.

14 years agoAdd install target for notmuch.desktop file.
Jeffrey C. Ollie [Sun, 22 Nov 2009 21:17:11 +0000 (15:17 -0600)]
Add install target for notmuch.desktop file.

Add an install target that uses desktop-file-install to install the
desktop file in the appropriate location.  The location of the install
can be modified by changing the desktop_dir variable.

Signed-off-by: Jeffrey C. Ollie <jeff@ocjtech.us>
14 years agoFix misspelling of DT_UNKNOWN.
Carl Worth [Sat, 23 Jan 2010 09:45:23 +0000 (22:45 +1300)]
Fix misspelling of DT_UNKNOWN.

How foolish of me to advertise the fact that I pushed a commit without
compiling it first...

14 years agoREADME: Tighten up the text a bit.
Carl Worth [Sat, 23 Jan 2010 06:00:25 +0000 (19:00 +1300)]
README: Tighten up the text a bit.

As Keith pointed out, (with a humorous citation from Mark Twain),
the two uses of "very" added nothing to the description. Also,
"large collection of email" was repeated uselessly.

14 years agoAdd some comments to document the recently-fixed handling of d_type.
Carl Worth [Sat, 23 Jan 2010 05:58:30 +0000 (18:58 +1300)]
Add some comments to document the recently-fixed handling of d_type.

The fix was subtle, (requiring less code than originally expected), so
it behooves us to document it well.

14 years agonotmuch new: Fix to work on filesystems returning DT_UNKNOWN
Geo Carncross [Thu, 21 Jan 2010 20:10:18 +0000 (15:10 -0500)]
notmuch new: Fix to work on filesystems returning DT_UNKNOWN

Such as reiserfs or xfs. This has been broken since the merge of
support for rename and deletion of files from the mail store.

Here's the original justification for the patch:

A review of notmuch-new.c shows three uses of ->d_type:

Near line 153, in _entries_resemble_maildir() we can simply allow for
DT_UNKNOWN. This would fail if people have MH-style folders which have
three folders called "new" "cur" and "tmp", but that seems unlikely, in
which case the "tmp" folder would simply not be scanned.

Near line 273 in add_files_recursive() we have another check. If
DT_UNKNOWN, we fall through, then add_files_recursive() does a stat
almost immediately, returning with success if the path isn't a
directory.

Thus, the fallback is already written.

Finally, near line 343, in add_files_recursive() (a long function) we
have another check. Here we can simply treat DT_UNKNOWN as DT_LNK, since
the logic for the stat() results are the same.

14 years agoInstall zsh completion file
martin f. krafft [Fri, 8 Jan 2010 22:49:34 +0000 (11:49 +1300)]
Install zsh completion file

According to the Debian zsh maintainer Clint Adams, this is the first
time that a package installs its own completer into zsh. Part of the
reason this is not usually done is because zsh does not provide a stable
API.

We agreed to try it, given that notmuch is expected to change quite
a bit initially. If there are problems or the completer goes stable,
we'll move it into the upstream zsh repository.

Signed-off-by: martin f. krafft <madduck@debian.org>
14 years agonotmuch new: Print upgrade progress report as a percentage.
Carl Worth [Sun, 10 Jan 2010 01:38:23 +0000 (17:38 -0800)]
notmuch new: Print upgrade progress report as a percentage.

Previously we were printing a number of messages upgraded so far. The
original motivation for this was to accurately reflect the fact that
there are two passes, (so each message is processed twice and it's not
accurate to represent with a single count). But as it turns out, the
second pass takes zero time (relatively speaking) so we're still not
accounting for it.

If nothing else, the percentage-based reporting makes for a cleaner
API for the progress_notify function.

14 years agolib: Add non-content terms with a WDF value of 0.
Carl Worth [Sat, 9 Jan 2010 19:18:27 +0000 (11:18 -0800)]
lib: Add non-content terms with a WDF value of 0.

The WDF is the "within-document frequency" value for a particular
term. It's intended to provide an indication of how frequent a term is
within a document, (for use in computing relevance). Xapian's term
generator already computes WDF values when we use that, (which we do
for indexing all mail content).

We don't use the term generator when adding single terms for things
that don't actually appear in the mail document, (such as tags, the
filename, etc.). In this case, the WDF value for these terms doesn't
matter much.

But Xapian's flint backend can be more efficient with changes to terms
that don't affect the document "length". So there's a performance
advantage for manipulating tags (with the flint backend) if the WDF of
these terms is 0.

14 years agolib: Explicitly set BoolWeight when searching.
Carl Worth [Sat, 9 Jan 2010 19:16:40 +0000 (11:16 -0800)]
lib: Explicitly set BoolWeight when searching.

All notmuch searches currently sort by value (either date or message
ID) so it's just wasted effort for Xapian to compute relevance values
for each result. We now explicitly tell Xapian that we're uninterested
in the relevance values.

14 years agolib: Split the database upgrade into two phases for safer operation.
Carl Worth [Sat, 9 Jan 2010 19:13:12 +0000 (11:13 -0800)]
lib: Split the database upgrade into two phases for safer operation.

The first phase copies data from the old format to the new format
without deleting anything. This allows an old notmuch to still use the
database if the upgrade process gets interrupted. The second phase
performs the deletion (after updating the database version number). If
the second phase is interrupted, there will be some unused data in the
database, but it shouldn't cause any actual harm.

14 years agolib: Delete stale timestamp documents during database upgrade.
Carl Worth [Fri, 8 Jan 2010 17:52:25 +0000 (09:52 -0800)]
lib: Delete stale timestamp documents during database upgrade.

Once we move the timestamp to the new directory document, we don't
need the old one anymore.

14 years agonotmuch new: Don't prevent database upgrade from being interrupted.
Carl Worth [Fri, 8 Jan 2010 16:45:16 +0000 (08:45 -0800)]
notmuch new: Don't prevent database upgrade from being interrupted.

Our signal handler is designed to quickly flush out changes and then
exit. But if a database upgrade is in progress when the user
interrupts, then we just want to immediately abort. We could do
something fancy like add a return value to our progress_notify
function to allow it to tell the upgrade process to abort. But it's
actually much cleaner and robust to delay the installation of our
signal handler so that the default abort happens on SIGINT.

14 years agonotmuch new: Fix progress notification on database upgrade.
Carl Worth [Fri, 8 Jan 2010 05:24:44 +0000 (21:24 -0800)]
notmuch new: Fix progress notification on database upgrade.

This was firing continuously rather than just once per second as
intended.

14 years agonotmuch new: Automatically upgrade the database if necessary.
Carl Worth [Fri, 8 Jan 2010 02:30:32 +0000 (18:30 -0800)]
notmuch new: Automatically upgrade the database if necessary.

This takes advantage of the recently added library support to detect
if the database needs to be upgraded and then automatically performs
that upgrade, (with a nice progress report).

14 years agolib: Implement versioning in the database and provide upgrade function.
Carl Worth [Fri, 8 Jan 2010 02:26:31 +0000 (18:26 -0800)]
lib: Implement versioning in the database and provide upgrade function.

The recent support for renames in the database is our first time
(since notmuch has had more than a single user) that we have a
database format change. To support smooth upgrades we now encode a
database format version number in the Xapian metadata.

Going forward notmuch will emit a warning if used to read from a
database with a newer version than it natively supports, and will
refuse to write to a database with a newer version.

The library also provides functions to query the database format
version:

notmuch_database_get_version

to ask if notmuch wants a newer version than that:

notmuch_database_needs_upgrade

and a function to actually perform that upgrade:

notmuch_database_upgrade

14 years agonotmuch new: Fix deletion support to recurse on removed directories.
Carl Worth [Fri, 8 Jan 2010 02:20:28 +0000 (18:20 -0800)]
notmuch new: Fix deletion support to recurse on removed directories.

Previously, when notmuch detected that a directory had been deleted it
was only removing files immediately in that directory. We now
correctly recurse to also remove any directories (and files, etc.)
within sub-directories, etc.

14 years agoTODO: Add a couple of ideas that came up during recent coding.
Carl Worth [Fri, 8 Jan 2010 02:17:38 +0000 (18:17 -0800)]
TODO: Add a couple of ideas that came up during recent coding.

The notmuch_query_count_messages functions duplicates a lot of code
undesirably.

14 years agoPrefer READ_ONLY consistently over READONLY.
Carl Worth [Thu, 7 Jan 2010 18:29:05 +0000 (10:29 -0800)]
Prefer READ_ONLY consistently over READONLY.

Previously we had NOTMUCH_DATABASE_MODE_READ_ONLY but
NOTMUCH_STATUS_READONLY_DATABASE which was ugly and confusing. Rename
the latter to NOTMUCH_STATUS_READ_ONLY_DATABASE for consistency.

14 years agolib: Consolidate checks for read-only database.
Carl Worth [Thu, 7 Jan 2010 18:19:44 +0000 (10:19 -0800)]
lib: Consolidate checks for read-only database.

Previously, many checks were deep in the library just before a cast
operation. These have now been replaced with internal errors and new
checks have instead been added at the beginning of all top-levelentry
points requiring a read-write database.

The new checks now also use a single function for checking and
printing the error message. This will give us a convenient location to
extend the check, (such as based on database version as well).

14 years agolib: Clarify internal documentation of _notmuch_database_filename_to_direntry
Carl Worth [Thu, 7 Jan 2010 17:31:58 +0000 (09:31 -0800)]
lib: Clarify internal documentation of _notmuch_database_filename_to_direntry

The original wording made it sound like this function was just doing
some string manipulation. But this function actually creates new
directory documents as a side effect. So make that explicit in its
documentation.

14 years agonotmuch_message_get_filename: Support old-style filename storage.
Carl Worth [Thu, 7 Jan 2010 17:22:34 +0000 (09:22 -0800)]
notmuch_message_get_filename: Support old-style filename storage.

When a notmuch database is upgraded to the new database format, (to
support file rename and deletion), any message documents corresponding
to deleted files will not currently be upgraded. This means that a
search matching these documents will find no filenames in the expected
place.

Go ahead and return the filename as originally stored, (rather than
aborting with an internal error), in this case.

14 years agonotmuch new: Never ask the database for any names from a new directory.
Carl Worth [Wed, 6 Jan 2010 22:35:56 +0000 (14:35 -0800)]
notmuch new: Never ask the database for any names from a new directory.

When we know that we are adding a new directory to the database, (and
we therefore are using inode rather than strcmp-based sorting of the
filenames), then we *never* want to see any names from the
database. If we get any names that could only make us inadvertently
remove files that we just added.

Since it's not obvious from the Xapian documentation whether new terms
being added as part of new documents will appear in the in-progress
all-terms iteration we are using, (and this might differ based on
Xapian backend and also might differ based on how many new directories
are added and whether a flush threshold is reached).

For all of these reasons, we play it safe and use NULL rather than a
real notmuch_filenames_t iterator in this case to avoid any problem.

14 years agolib: Treat NULL as a valid (and empty) notmuch_filenames_t iterator.
Carl Worth [Wed, 6 Jan 2010 22:35:11 +0000 (14:35 -0800)]
lib: Treat NULL as a valid (and empty) notmuch_filenames_t iterator.

This will be convenient to avoid some special-casing in higher-level
code.

14 years agonotmuch new: Fix bug resulting in file removal on initial build of database.
Carl Worth [Wed, 6 Jan 2010 21:54:39 +0000 (13:54 -0800)]
notmuch new: Fix bug resulting in file removal on initial build of database.

The bug here was that we would see that the database did not know
anything about a directory so would get results from the filesystem in
inode rather than strcmp order.

However, we wouldn't actually ask for the list of files from the
database until after recursing into the sub-directories. So by the
time we traverse the filenames looking for deletions, the database
*does* have entries and we end up detecting erroneous deletions
because our filename list from the filesystem isn't in strcmp order.

So ask for the list of names from the database before doing any
additions to avoid this problem.

14 years agonotmuch new: Fix to detect deletions of names at the end of the list.
Carl Worth [Wed, 6 Jan 2010 21:26:47 +0000 (13:26 -0800)]
notmuch new: Fix to detect deletions of names at the end of the list.

Previously we only scanned the list of filenames in the filesystem and
detected a deletion whenever that scan skipped a name that existed in
the database. That much was fine, but we *also* need to continue
walking the list of names from the database when the filesystem list
is exhausted.

Without this, removing the last file or directory within any
particular directory would go undetected.

14 years agonotmuch new: Fix regression preventing addition of symlinked mail files.
Carl Worth [Wed, 6 Jan 2010 18:30:08 +0000 (10:30 -0800)]
notmuch new: Fix regression preventing addition of symlinked mail files.

As described in the previous commit message, we introduced multiple
symlink-based regressions in commit
3df737bc4addfce71c647792ee668725e5221a98

Here, we fix the case of symlinks to regular files by doing an extra
stat of any DT_LNK files to determine if they do, in fact, link to
regular files.

14 years agonotmuch new: Fix regression preventing recursion through symlinks.
Carl Worth [Wed, 6 Jan 2010 18:09:17 +0000 (10:09 -0800)]
notmuch new: Fix regression preventing recursion through symlinks.

In commit 3df737bc4addfce71c647792ee668725e5221a98 we switched from
using stat() to using the d_type field in the result of scandir() to
determine whether a filename is a regular file or a directory. This
change introduced a regression in that the recursion would no longer
traverse through a symlink to a directory. (Since stat() would resolve
the symlink but with scandir() we see a distinct DT_LNK value in
d_type).

We fix this for directories by allowing both DT_DIR and DT_LNK values
to recurse, and then downgrading the existing not-a-directory check
within the recursion to not be an error. We also add a new
not-a-directory check outside the recursion that is an error.

14 years agoFix typo in comment.
Carl Worth [Wed, 6 Jan 2010 18:08:51 +0000 (10:08 -0800)]
Fix typo in comment.

The difference between "now" and "not" ends up being fairly dramatic.

14 years agonotmuch new: Print counts of deleted and renamed messages.
Carl Worth [Wed, 6 Jan 2010 18:07:49 +0000 (10:07 -0800)]
notmuch new: Print counts of deleted and renamed messages.

It's nice to be able to see a report indicating that the recently
added support for detecting file rename and deletion is working.

14 years agolib: Indicate whether notmuch_database_remove_message removed anything.
Carl Worth [Wed, 6 Jan 2010 18:06:00 +0000 (10:06 -0800)]
lib: Indicate whether notmuch_database_remove_message removed anything.

Similar to the return value of notmuch_database_add_message, we now
enhance the return value of notmuch_database_remove_message to
indicate whether the message document was entirely removed (SUCCESS)
or whether only this filename was removed and the document exists
under other filenamed (DUPLICATE_MESSAGE_ID).

14 years agolib: Update documentation of notmuch_database_add_message.
Carl Worth [Wed, 6 Jan 2010 18:03:58 +0000 (10:03 -0800)]
lib: Update documentation of notmuch_database_add_message.

Previously, adding a filename with the same message ID as an existing
message would do nothing. But we recently fixed this to instead add
the new filename to the existing message document. So update the
documentation to match now.

14 years agoIndex content from citations and signatures.
Carl Worth [Wed, 6 Jan 2010 03:14:07 +0000 (19:14 -0800)]
Index content from citations and signatures.

In the presentation we often omit citations and signatures, but this
is not content that should be omitted from the index, (especially
when the citation detection is wrong---see cases where a line
beginning with "From" is corrupted to ">From" by mail processing
tools).

14 years agonotmuch new: Proper support for renamed and deleted files.
Carl Worth [Wed, 6 Jan 2010 02:59:18 +0000 (18:59 -0800)]
notmuch new: Proper support for renamed and deleted files.

The "notmuch new" command will now efficiently notice if any files or
directories have been removed from the mail store and will
appropriately update its database.

Any given mail message (as determined by the message ID) may have
multiple corresponding filenames, and notmuch will return one of
them. When a filen is deleted, the corresponding filename will be
removed from the message in the database. When the last filename is
removed from a message, that message will be entirely removed from the
database.

All file additions are handled before any file removals so that rename
is supported properly.

14 years agonotmuch new: Store detected removed filenames for later processing.
Carl Worth [Wed, 6 Jan 2010 01:56:11 +0000 (17:56 -0800)]
notmuch new: Store detected removed filenames for later processing.

It is essential to defer the actual removal of any filenames from the
database until we are entirely done adding any new files. This is to
avoid any information loss from the database in the case of a renamed
file or directory.

Note that we're *still* not actually doing any removal---still just
printing messages indicating the filenames that were detected as
removed. But we're at least now printing those messages at a time when
we actually *can* do the actual removal.

14 years agonotmuch new: Detect deleted (renamed) files and directories.
Carl Worth [Wed, 6 Jan 2010 01:43:03 +0000 (17:43 -0800)]
notmuch new: Detect deleted (renamed) files and directories.

This takes advantage of the notmuch_directory_t interfaces added
recently (with cooresponding storage of directory documents in the
database) to detect when files or entire directories are deleted or
renamed within the mail store.

This also fixes the recent regression where *all* files would be
processed by every run of "notmuch new", (now only new files are
processed once again).

The deleted files and directories are only detected so far. They
aren't properly removed from the database.

14 years agoadd_files_recursive: Make the maildir detection more efficient.
Carl Worth [Wed, 6 Jan 2010 00:42:14 +0000 (16:42 -0800)]
add_files_recursive: Make the maildir detection more efficient.

Previously, we were re-scanning the entire list of entries for every
directory entry. Instead, we can simply check if the entries look like
a maildir once, up-front.

14 years agoadd_files_recursive: Separate scanning for directories and files for legibility.
Carl Worth [Wed, 6 Jan 2010 00:35:02 +0000 (16:35 -0800)]
add_files_recursive: Separate scanning for directories and files for legibility.

We now do two scans over the entries returned from scandir. The first
scan is looking for directories (and making the recursive call). The
second scan is looking for new files to add to the database.

This is easier to read than the previous code which had a single loop
and some if statements with ridiculously long bodies. It also has the
advantage that once the directory scan is complete we can do a single
comparison of the filesystem and database mtimes and entirely skip the
second scan if it's not needed.

14 years agoadd_files_recursive: Use consistent naming for array and count variables.
Carl Worth [Wed, 6 Jan 2010 00:15:43 +0000 (16:15 -0800)]
add_files_recursive: Use consistent naming for array and count variables.

Previously we had an array named "namelist" and its count named
"num_entries". We now use an array name of "fs_entries" and a count
named "num_fs_entries" to try to preserve sanity.

14 years agonotmuch new: Remove an unnecessary stat of every regular file in the mail store.
Carl Worth [Wed, 6 Jan 2010 00:06:46 +0000 (16:06 -0800)]
notmuch new: Remove an unnecessary stat of every regular file in the mail store.

We were previousl using the stat for two reasons. One was to obtain
the mtime of the file. This usage was removed in the previous commit,
(since the mtime is unreliable in the case of a file being moved into
the mail store).

The second reason was to identify regular and directory file
types. But this information is already available in the result we get
from scandir.

What's left is simply a stat for each directory in the mailstore,
(which we are still using to compare filesystem mtime with the mtime
stored in the database).

14 years agonotmuch new: Eliminate the check on the mtime of regular files before adding.
Carl Worth [Tue, 5 Jan 2010 23:59:11 +0000 (15:59 -0800)]
notmuch new: Eliminate the check on the mtime of regular files before adding.

This check was buggy in that moving a pre-existing file into the mail
store, (where the file existed before the last run of "notmuch new"),
does not update the mtime of the file. So the message would never be
added to the database.

The fix here is not practical in the long run, (since it causes *all*
files in the mail store to be processed in every run of "notmuch new"
(!)). But this change will let us drop a stat() call that we don't
otherwise need and will help move us toward proper database-backed
detection of new files, (which will fix the bug without the
performance impact of the current fix).

14 years agonotmuch new: Fix internal documentation of add_files_recursive.
Carl Worth [Tue, 5 Jan 2010 23:52:59 +0000 (15:52 -0800)]
notmuch new: Fix internal documentation of add_files_recursive.

To make it more clear that the mtime of a directory does not affect
whether further sub-directories are examined, (they are examined
unconditionally).

14 years agonotmuch new: Rename the various timestamp variables to be more clear.
Carl Worth [Tue, 5 Jan 2010 23:31:56 +0000 (15:31 -0800)]
notmuch new: Rename the various timestamp variables to be more clear.

The previous name of "path_mtime" was very ambiguous. The new names
are much more obvious (fs_mtime is the mtime from the filesystem and
db_mtime is the mtime from the database).

14 years agonotmuch new: Avoid updating directory timestamp if interrupted.
Carl Worth [Tue, 5 Jan 2010 23:23:52 +0000 (15:23 -0800)]
notmuch new: Avoid updating directory timestamp if interrupted.

This was a very dangerous bug. An interrupted "notmuch new" session
would still update the timestamp for the directory in the
database. This would result in mail files that were not processed due
to the original interruption *never* being picked up by future runs of
"notmuch new". Yikes!

14 years agonotmuch-new: Remove dead add_files_callback code.
Carl Worth [Tue, 5 Jan 2010 23:13:16 +0000 (15:13 -0800)]
notmuch-new: Remove dead add_files_callback code.

Always satisfying to delete code (even if tiny).

14 years agoMake the add_files function static within notmuch-new.c.
Carl Worth [Tue, 5 Jan 2010 23:11:21 +0000 (15:11 -0800)]
Make the add_files function static within notmuch-new.c.

No other files need this function so we don't need it exported in
notmuch-client.h.

14 years agoMakefiles: Use .DEFAULT to support arbitrary targets from sub directories.
Carl Worth [Tue, 5 Jan 2010 23:05:57 +0000 (15:05 -0800)]
Makefiles: Use .DEFAULT to support arbitrary targets from sub directories.

Taking advantage of the .DEFAULT construct means that we won't need to
explicitly list targets such as "clean", etc. in each sub-Makefile.

14 years agoAdd missing comment for NOTMUCH_STATUS_READONLY_DATABASE.
Carl Worth [Tue, 5 Jan 2010 23:01:58 +0000 (15:01 -0800)]
Add missing comment for NOTMUCH_STATUS_READONLY_DATABASE.

And adjust the string representation of the same to match.

14 years agolib: Implement new notmuch_directory_t API.
Carl Worth [Tue, 5 Jan 2010 21:29:23 +0000 (13:29 -0800)]
lib: Implement new notmuch_directory_t API.

This new directory ojbect provides all the infrastructure needed to
detect when files or directories are deleted or renamed. There's still
code needed on top of this (within "notmuch new") to actually do that
detection.

14 years agoRevamp the proposed directory-tracking API slightly.
Carl Worth [Tue, 5 Jan 2010 21:06:24 +0000 (13:06 -0800)]
Revamp the proposed directory-tracking API slightly.

This commit contains my changes to the API proposed by Keith. Nothing
is dramatically different. There are minor things like changing
notmuch_files_t to notmuch_filenames_t and then various things needed
for completeness as noticed while implementing this, (such as
notmuch_directory_destroy and notmuch_directory_set_mtime).

14 years agoPrototypes for directory tracking
Keith Packard [Wed, 9 Dec 2009 08:51:52 +0000 (00:51 -0800)]
Prototypes for directory tracking

There's no functionality here yet---just a sketch of what the
interface could look like.

14 years agodatabase: Add new, public notmuch_database_remove_message
Carl Worth [Mon, 21 Dec 2009 23:14:32 +0000 (15:14 -0800)]
database: Add new, public notmuch_database_remove_message

This will allow applications to support the removal of messages, (such
as when a file is deleted from the mail store). No removal support is
provided yet in commands such as "notmuch new".