There is absolutely no reason to make events start on noon, 12:00.
Switching to 00:00 seems totally reasonable here, and makes event
handling a bit easier, also.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Nasty off-by-one error here. An event should be associated with a day if
it starts at 12:00 a.m. and shouldn't be associated if it enters the
next day.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Now that we support more powerful duration strings, we should also
resize the input field for duration strings. Twelve characters is enough
space for "+999d23h59m".
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Add support for "1d23h42m"-like duration strings to parse_duration().
Also, switch to using a deterministic finite automaton to parse duration
strings, as things tend to get unclear.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Now that parse_time() and parse_duration() do all the validation work,
this isn't used (nor needed) any longer.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Make use of these new helpers at various places. Note that this patch
implies a few behavioural changes:
* Short forms such as "23:" and ":45" are allowed when entering times.
* Durations always need to be prefixed with a plus sign ("+"), with the
nice side effect that you can now use "+3:30" to declare an
appointment that lasts three hours and thirty minutes (that's much
more convenient than "+210").
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
These helpers can be used in a fashion similar to parse_date(). In
addition to check_time(), parse_time() and parse_duration() support
short forms such as "23:" (instead of "23:00") and ":45" (instead of
"00:45").
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Add a global count buffer to our main loop and pass it to keys_getch()
as well as to all motion commands.
This enables some fancy shortcuts:
* Type "2l" to jump to the day after tomorrow.
* Type "4k" (or "28l") to move forward four weeks.
* Type "2$" to go to the end of next week.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Key commands can be prefixed with a natural number - keys_getch() will
store this number in the buffer pointed to by the second parameter. Set
this parameter to NULL to disable count prefixes.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
This allows for moving more than one item up/down.
This currently isn't used anywhere but will be bound to a key with one
of the following patches.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Allows for moving more than one step forward/backward.
This is not used anywhere yet but a key binding will likely be added in
one of the following patches.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Add a window bitmask to wins_update() and only update windows that might
actually require an update in our main loop. This improves response
times of the user interface a bit.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Instead of using a hacky flag, simply move the storage update code to a
separate function and call it when needed.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Add an additional check to apoint_update_panel() and todo_update_panel()
and only highlight currently selected items if the corresponding panel
is active. This allows us to remove all the highlight pointer juggling
that we used whenever the panel selection changed.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
In contrast to LLIST_{,TS}_FIND_FOREACH, these convenience macros search
for the first match and return successors until there is an item that
isn't matched by the filter callback. Any items beyond the first cut-off
are discarded.
Should be used when results are known to be continuous, such as
appointments and events belonging to a specific day etc.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
This convenience function can be used to return the successor of a list
item if it is matched by a filter callback and return NULL otherwise.
We will use this for an improved version of the LLIST_FIND_FOREACH macro
that can be used whenever results are known to be continuous.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Adding a tail pointer to each list increases memory footprint by four
bytes, while reducing the runtime of llist_add() from O(n) to O(1). In
testing, the time required to append 100000 elements to a linked list
was reduced from 29.245s to 0.009s.
Our second main concern is to reduce the runtime of llist_add_sorted()
when inserting elements from a presorted list (this is reduced from O(n)
to O(1) as well), since the data files contain appointments in sorted
order and are always processed front to back.
Some local numbers show how this speeds up calcurse startup (test set
with 50000 appointments):
0.22user 0.12system 0:00.35elapsed 99%CPU (0avgtext+0avgdata 5396maxresident)k
0inputs+8outputs (0major+1398minor)pagefaults 0swaps
As opposed to the unpatched binary:
21.97user 0.25system 0:22.23elapsed 99%CPU (0avgtext+0avgdata 5388maxresident)k
0inputs+48outputs (0major+1396minor)pagefaults 0swaps
This is a ~10000% increase in speed. Timings for reading random input
files generated by a script stay the same (32.391s vs. 31.776s).
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Adds a "-g" option that allows for running the garbage collector for
note files manually. This is useful for users that do not use note files
at all or rarely edit/remove them.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
If "auto_gc" is enabled, the garbage collector for note files will be
run on every exit. As this is an experimental feature and may cause data
loss, this is disabled by default.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Now that we use hashes to identify notes, a garbage collector comes in
handy. We search for note files that aren't referenced anywhere. Such
files might come up if a note, that isn't connected with any other item,
is edited.
Huge parts of this code are very hackish due to our data structure
layout and the way our hash table implementation works.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Now that we use hash-based note file names, note files should never be
unlinked as a note file might be shared.
Also, remove the ERASE_FORCE_KEEP_NOTE flag that no longer makes any
sense.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Note file names are now generated based on their content. Items using
the same note will share a single note file.
Please note that this implies a few changes:
* Both random-style and hash-style note files need to be handled to
ensure we do not break backwards compatibility.
* Note files may not be moved or deleted if a note is changed or removed
since the original note file might be used by another item as well.
* A garbage collector to remove unreferenced note files needs to be
implemented.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Read up to the first blank in note_read() instead of assuming a
fixed-width note file name. Accept everything up to 40 characters (which
is the length of a SHA1 hash in hexadecimal representation).
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Add io_file_cp() which can be used to copy an existing source file to
another location. We will need this for our new hash-based note file
names.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Add an implementation of the SHA1 hash algorithm based on a public
domain project by Steve Reid <sreid@sea-to-sky.net>. You can get the
original sources from:
ftp://ftp.funet.fi/pub/crypt/hash/sha/sha1.c
We will need this for several things. File names of note files will be
generated based on a hash their content instead of using a random name.
Items can be uniquely identified using a hash.
In addition to the regular sha1_init(), sha1_update() and sha1_final()
operations, our implementation also contains wrappers for hashing a
string or a stream.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
This short paper contains information on how to create and submit
patches to calcurse. This used to be on our website - "doc/" seems to be
a better place for this, though.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Always invoke pthread_join() when we blow up a thread via
pthread_cancel() (avoid zombie threads).
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Update the number of todo items when importing an iCal file to prevent
some items from being inaccessible.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Use "vi" instead of "/usr/bin/vi" and "less" instead of "/usr/bin/less".
Hardcoding absolute paths is a bad idea:
$ uname -rsv
Linux 3.0-ARCH #1 SMP PREEMPT Tue Aug 30 07:32:23 UTC 2011
$ which less
/bin/less
The "$PATH" environment variable will almost always have a better idea
of where these binaries are located.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
This should fix all remaining build issues with documentation
generation. Finally.
The new "--disable-docs" option should be used to skip documentation
completely, whereas "--without-asciidoc" can be used if both manual and
man page should not be rebuilt. As a consequence, "--without-asciidoc"
can only be combined with "--enable-docs" if ready-made documentation
already exists in "doc/". This is true for release tarballs (where we
include prebuilt documentation for the sake of portability and
simplicity), as well as for Git checkouts where the documentation has
already been built earlier.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
Do not use snprintf() here as printf() behaviour is undefined if the
destination pointer is used as a parameter.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
This is a rather invasive change that introduces correct line folding to
our iCal parser.
From now on, ical_readline() should be used instead of fgets() to read
lines from an iCal file as it unfolds lines automatically. We also need
to use shared buffers as each ical_readline() invocation eats up the
first part of the next line and stores it in the "lstore" buffer.
Subsequent ical_readline() invocations copy the contents of this buffer
and append continuation lines. We currently use a single buffer pair
that is allocated in io_import_data() and pass it to all subroutines.
ical_readline_init() needs to be called once for every buffer pair. It
reads the first part of the current line and writes to "lstore",
clearing the target buffer at the same time.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
If "notify-all" is enabled, all non-flagged appointments will be
notified (instead of flagged ones). This is useful for users that want
to be notified of everything.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
We used to iterate over the list of regular appointments and the list of
recurrent appointments in separate loops, thus leading to recurrent
appointments being printed first and regular appointments being printed
afterwards, regardless of their start and end times. Merge these loops
to coerce precedence of the start time.
There's still a fair bit of hackery in here - we will fix that later.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
We passed the function arguments the wrong way round. This regression
was introduced in commit 77ef3fe76e4ce4c9a990e8a5904ad2d83420ca02.
Signed-off-by: Lukas Fleischer <calcurse@cryptocrack.de>
