Although I think it's a pretty bad idea to continue using the old API,
this allows both a more gentle transition for clients of the library,
and allows us to break one monolithic change into a series
Before this change with --auto-daemon but without --create-frame
emacs server was started but no clients stay connected to it
(in both graphical and terminal displays).
Note that this changes how --client --auto-daemon works on
graphical display; New emacs frame is now created for the
message (and message-exit-actions hook appended).
Make the default behaviour for --client the same as emacsclient
default: do not create a new frame. Add a new option --create-frame,
passing the same option to emacsclient to create a frame.
Automatically starting Emacs in daemon mode if the Emacs server is not
running is a matter of preference better not hard coded in
notmuch-emacs-mua. Add an option to control the behaviour.
The side effect is that all of add_files_state will be initialized to
zero, removing any lingering doubt that some of it might not be
initialized. It's not a small struct, and the initialization is
scattered around a bit, so this makes the code more readable.
Let each view have a "sort" key, typically used with values
"oldest-first" or "newest-first" (although all values in Query.SORT
are accepted), and sort the results accordingly. Oldest first remains
the default.
The dynamic approach of mapping sort values is as suggested by
W. Trevor King <wking@tremily.us>.
When upgrading emacs, the install file can (apparently) be called with
the lisp files already linked into the flavour specific
directories. In this case we should not fail.
The order of the results with --output=count and --deduplicate=address
are unspecified as they're based on a hash table traversal. This being
the case, optimize the query by explicitly requesting unsorted
results. Clarify the documentation accordingly.
First a simple smoke test first, next generate messages with multiple
email address variants and check the behaviour of deduplication
schemes with these.
Currently we key the address hash table with the case sensitive "name
<address>". Switch to case insensitive keying with just address, and
store the case sensitive name and address in linked lists. This will
be helpful in adding support for different deduplication schemes in
the future.
There will be a slight performance penalty for the current full case
sensitive name + address deduplication, but this is simpler as a whole
when other deduplication schemes are added, and I expect the schemes
to be added to become more popular than the current default.
Aparet from the possible performance penalty, the only user visible
change should be the change in the output ordering for
--output=count. The order is not guaranteed (and is based on hash
table traversal) currently anyway, so this should be of no
consequence.
It doesn't seem likely we can support simple date:<expr> expanding to
date:<expr>..<expr> any time soon. (This can be done with a future
version of Xapian, or with a custom query query parser.) In the mean
time, provide shorthand date:<expr>..! to mean the same. This is
useful, as the expansion takes place before interpetation, and we can
use, for example, date:yesterday..! to match from beginning of
yesterday to end of yesterday.
Idea from Mark Walters <markwalters1009@gmail.com>.
It isn't completely clear what we want to do here, but
1) We currently don't fail if we skip a whole test file (mainly because
we neglect to count those skipped tests properly). This change at least
makes the two kinds of skipping consistent.
2) Automated build environments may have good reasons for building with
a minimal set of prereqs, and we don't want to discourage running our
test suite by breaking builds.
Make it possible to use notmuch address as part of a | sort | uniq -c
pipe instead of forcing --output=count. This is useful for combining
results from multiple notmuch address queries.
Check argc mainly to fix unused parameter warning:
test/symbol-test.cc:7:14: warning: unused parameter ‘argc’ [-Wunused-parameter]
int main(int argc, char** argv) {
^
This makes more sense than telling the compiler it's unused on
purpose.
These functions are all just accessors, and it's pretty clear they don't
modify the query struct. This also fixes one warning I created when I
introduced status.c.
I think it would be no real problem to cut and paste the gdb based
error message test from count to the other clients modified here, but
I'm not currently convinced it's worth the trouble since the code path
being tested is almost the the same, and the tests are relatively
heavyweight.
This brings back status information that may have been hidden by the
great library logging conversion.
Note the change of the internal API / return-value for count_files. The
other count calls to the lib will also get error handling when that API
is updated in the lib.
The library does not have a function to remove a directory document
for a path. Usually this doesn't matter except for a slight waste of
space. However, if the same directory gets added to the filesystem
again, the old directory document is found with the old mtime. Reset
the directory mtime on removal to avoid problems.
The corner case that can hit this problem is renaming directories back
and forth. Renaming does not change the mtime of the directory in the
filesystem, and thus the old db directory document mtime may match the
fs mtime of the directory.
The long term fix might be to add a library function to remove a
directory document, however this is a much simpler and faster fix for
the time being.
When filtering the current search further with notmuch-search-filter,
wrap the current search in parens (if necessary).
This fixes unexpected behavior when the current search is
complex (like "(tag:this and date:one_week_ago..) or tag:that").
It's becoming a maintenance burden to do anything things with the
crypto glue code twice, once for 2.4 and once for 2.6. I don't have
any 2.4 version available to test on my development machine anymore,
so the 2.4 specific code paths are likely not very well tested.
The function notmuch_exit_if_unmatched_db_uuid is split from
notmuch_process_shared_options because it needs an open notmuch
database.
There are two exceptional cases in uuid handling.
1) notmuch config and notmuch setup don't currently open the database,
so it doesn't make sense to check the UUID.
2) notmuch compact opens the database inside the library, so we either
need to open the database just to check uuid, or change the API.
In the short term we need a way to get lastmod information e.g. for
the test suite. In the long term we probably want to add lastmod
information to at least the structured output for several other
clients (e.g. show, search).
This exposes the committed database revision to library users along
with a UUID that can be used to detect when revision numbers are no
longer comparable (e.g., because the database has been replaced).
This adds a new document value that stores the revision of the last
modification to message metadata, where the revision number increases
monotonically with each database commit.
An alternative would be to store the wall-clock time of the last
modification of each message. In principle this is simpler and has
the advantage that any process can determine the current timestamp
without support from libnotmuch. However, even assuming a computer's
clock never goes backward and ignoring clock skew in networked
environments, this has a fatal flaw. Xapian uses (optimistic)
snapshot isolation, which means reads can be concurrent with writes.
Given this, consider the following time line with a write and two read
transactions:
write |-X-A--------------|
read 1 |---B---|
read 2 |---|
The write transaction modifies message X and records the wall-clock
time of the modification at A. The writer hangs around for a while
and later commits its change. Read 1 is concurrent with the write, so
it doesn't see the change to X. It does some query and records the
wall-clock time of its results at B. Transaction read 2 later starts
after the write commits and queries for changes since wall-clock time
B (say the reads are performing an incremental backup). Even though
read 1 could not see the change to X, read 2 is told (correctly) that
X has not changed since B, the time of the last read. In fact, X
changed before wall-clock time A, but the change was not visible until
*after* wall-clock time B, so read 2 misses the change to X.
This is tricky to solve in full-blown snapshot isolation, but because
Xapian serializes writes, we can use a simple, monotonically
increasing database revision number. Furthermore, maintaining this
revision number requires no more IO than a wall-clock time solution
because Xapian already maintains statistics on the upper (and lower)
bound of each value stream.
