platform/upstream/libxkbcommon.git
12 years agokbproto untanglement: XkbKeyNameLength
Daniel Stone [Tue, 21 Aug 2012 10:21:19 +0000 (11:21 +0100)]
kbproto untanglement: XkbKeyNameLength

Define it ourselves as XKB_KEY_NAME_LENGTH and use that, instead of the
one from XKB.h.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
12 years agorules: use goto instead of state variable
Ran Benita [Tue, 28 Aug 2012 15:04:25 +0000 (18:04 +0300)]
rules: use goto instead of state variable

There's no noticeable speed difference, but I think it's nicer and more
explicit than the previous code. Some people just don't like goto,
though..

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agorules: rewrite
Ran Benita [Sun, 26 Aug 2012 05:54:31 +0000 (08:54 +0300)]
rules: rewrite

Rewrite the rules file parser for clarity, performance and memory usage
reduction. The previous implementation was quite hard to navigate and
did a lot of unnecessary work and copying.
This implementation keeps along just the state necessary, and doesn't
perform any copying of the file's content (although the entire file is
mmap'ed as before). Hopefully it's also easier to understand, has better
documentation, and better error checking and reporting. We try to
reproduce the previous behavior in every case.

Note: the diff is pretty confusing; it's likely better to look at the
file directly.

Benchmarks:
    On an old 32-bit Intel processor.
    gcc -O2 -pg
    ./test/rulescomp bench
    grof test/rulescomp

    Before:
        compiled 1000 keymaps in 14.863564304s
        %   cumulative   self              self     total
        time   seconds   seconds    calls  ms/call  ms/call  name
        49.33      4.43     4.43 30610000     0.00     0.00  yylex
        17.93      6.04     1.61    31000     0.05     0.22  yyparse
        6.57      6.63     0.59     1000     0.59     0.59  load_rules
        3.23      6.92     0.29  3637000     0.00     0.00  AppendStmt
        2.45      7.14     0.22   472000     0.00     0.00  AddKeySymbols
        2.12      7.33     0.19  3591000     0.00     0.00  atom_intern
        2.12      7.52     0.19   518000     0.00     0.00  FindNamedKey
        2.00      7.70     0.18   230000     0.00     0.00  FreeStmt
        1.78      7.86     0.16     1000     0.16     0.17  UpdateModifiersFromCompat
        1.34      7.98     0.12   732000     0.00     0.00  AddKeyName
        1.34      8.10     0.12                             __x86.get_pc_thunk.bx

    After:
        compiled 1000 keymaps in 13.874666269s
        %   cumulative   self              self     total
        time   seconds   seconds    calls  ms/call  ms/call  name
        49.82      4.26     4.26 30610000     0.00     0.00  yylex
        22.22      6.16     1.90    31000     0.06     0.22  yyparse
        2.92      6.41     0.25  3591000     0.00     0.00  atom_intern
        2.57      6.63     0.22     1000     0.22     0.25  xkb_components_from_rules
        2.11      6.81     0.18  3637000     0.00     0.00  AppendStmt
        2.11      6.99     0.18   230000     0.00     0.00  FreeStmt
        1.99      7.16     0.17   518000     0.00     0.00  FindNamedKey
        1.99      7.33     0.17     1000     0.17     0.17  UpdateModifiersFromCompat
        1.99      7.50     0.17                             __x86.get_pc_thunk.bx
        1.52      7.63     0.13   150000     0.00     0.00  AddInterp
        1.40      7.75     0.12   472000     0.00     0.00  AddKeySymbols

    On a newer 64-bit Intel processor.
    gcc -O2
    ./test/rules-file bench

    Before:
        processed 20000 times in 15.940546625s

    After:
        processed 20000 times in 5.295026345s

Allocations:
    gcc -O2
    valgrind test/rulescomp

    Before:
    total heap usage: 257,519 allocs, 257,519 frees, 14,766,529 bytes allocated

    After:
    total heap usage: 240,756 allocs, 240,756 frees, 14,007,886 bytes allocated

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: make LEDInfo a wrapper around xkb_indicator_map
Ran Benita [Tue, 4 Sep 2012 14:40:09 +0000 (17:40 +0300)]
compat: make LEDInfo a wrapper around xkb_indicator_map

instead of duplicating the fields. The same is done in SymInterpInfo
which wraps xkb_sym_interpret.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoFold keymap->indicator_names into keymap->indicators
Ran Benita [Tue, 4 Sep 2012 14:20:46 +0000 (17:20 +0300)]
Fold keymap->indicator_names into keymap->indicators

This makes sense, since giving a name to an indicator 'activates' the
indicator_map in that index.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoUnify some string tables from xkbcomp, text and keymap-dump
Ran Benita [Sun, 2 Sep 2012 18:45:42 +0000 (21:45 +0300)]
Unify some string tables from xkbcomp, text and keymap-dump

We move the LookupEntry struct from expr.h to text.h, along with most of
the lookup tables. This makes them available everywhere.
Looking up a value in the LookupEntry format is slower than direct index
mapping, but it allows multiple names per value (with the canonical one
being first) and "all"- and "none"-type masks. These functions are not
used anywhere efficiency matters.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoConvert rest of names in xkb_keymap back to atoms
Ran Benita [Fri, 31 Aug 2012 16:26:51 +0000 (19:26 +0300)]
Convert rest of names in xkb_keymap back to atoms

These were kept as atoms, but since the keymap was exposed in the API,
we converted them to strings; no the keymap is no longer exposed, so we
can go back to atoms. They make the keymap smaller (at least on 64-bit
machines) and the comparisons faster.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoexpr: make ResolveString return an atom
Ran Benita [Fri, 31 Aug 2012 16:05:49 +0000 (19:05 +0300)]
expr: make ResolveString return an atom

Almost all callers do xkb_atom_intern on the currently returned string,
while ResolveString converts the atom to the string to begin with...
uselss double work.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoatom: add xkb_atom_lookup
Ran Benita [Fri, 31 Aug 2012 15:52:26 +0000 (18:52 +0300)]
atom: add xkb_atom_lookup

This will only lookup the string and return the atom if found; it will
not intern it if not. This is useful when e.g. getting a string from the
user (which may be arbitrary) and comparing against atoms.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoatom: separate lookup logic from atom_intern
Ran Benita [Fri, 31 Aug 2012 15:47:28 +0000 (18:47 +0300)]
atom: separate lookup logic from atom_intern

This would allow us to add a non-interning xkb_atom_lookup function.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoatom: make type and name of the 'a' field clearer
Ran Benita [Fri, 31 Aug 2012 15:22:03 +0000 (18:22 +0300)]
atom: make type and name of the 'a' field clearer

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agotest/rules-file: add benchmark
Ran Benita [Tue, 28 Aug 2012 16:07:07 +0000 (19:07 +0300)]
test/rules-file: add benchmark

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agotest: add rmlvo-to-kccgst tool
Ran Benita [Wed, 29 Aug 2012 08:54:05 +0000 (11:54 +0300)]
test: add rmlvo-to-kccgst tool

For a quick look at what components result from the rules.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agorules: fix check for appending '|' character when applying
Ran Benita [Wed, 29 Aug 2012 09:10:28 +0000 (12:10 +0300)]
rules: fix check for appending '|' character when applying

There are two ways to separate multiple files in XKB include statements:
'+' will cause the later file to override the first in case of conflict,
while '|' will cause it augment it (this is done by xkbcomp). '!' is
unrelated here.
Since '|' is practically never used, this wasn't noticed.

In the modified test, the '|some_compat' previously was just ignored.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoUse XKB_MOD_INVALID instead of XkbNoModifier
Ran Benita [Wed, 29 Aug 2012 12:17:00 +0000 (15:17 +0300)]
Use XKB_MOD_INVALID instead of XkbNoModifier

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoDon't printf NULL strings
Ran Benita [Wed, 29 Aug 2012 07:12:56 +0000 (10:12 +0300)]
Don't printf NULL strings

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoxkbcomp: add debug messages of the RMLVO and KcCGST
Ran Benita [Wed, 29 Aug 2012 07:06:47 +0000 (10:06 +0300)]
xkbcomp: add debug messages of the RMLVO and KcCGST

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoinclude: document ParseIncludeMap better
Ran Benita [Tue, 28 Aug 2012 21:44:58 +0000 (00:44 +0300)]
include: document ParseIncludeMap better

The format of the include statment is not explained anywhere, the code
is confusing and the comments misleading. Try to explain it better.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agotest/data/symbols: keypad can only have one default section
Ran Benita [Tue, 28 Aug 2012 12:55:35 +0000 (15:55 +0300)]
test/data/symbols: keypad can only have one default section

Avoids a warning, from xkeyboard-config:

commit 6676053f2c93596c2aaa9905151a5c76355a1540
Author: Peter Hutterer <peter.hutterer@who-t.net>
Date:   Fri Jun 29 09:53:45 2012 +1000

    symbols: keypad can only have one default section

    Warning:          Multiple default components in keypad
                      Using x11, ignoring pointerkeys

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoinclude: properly use the default map if none is found
Ran Benita [Tue, 28 Aug 2012 12:05:01 +0000 (15:05 +0300)]
include: properly use the default map if none is found

For some reason this piece of code wasn't copied from xkbcomp, which
causes all of the warnings like these:
Warning:       No map in include statement, but "pc" contains several; Using first defined map, "pc105"

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoSet log level for tests through env, not directly
Ran Benita [Thu, 30 Aug 2012 17:23:35 +0000 (20:23 +0300)]
Set log level for tests through env, not directly

This way the test logs have all the information, but we don't get eye
bleed every time we run them manually. One can always use
TESTS_ENVIRONMENT (we correctly use AM_TESTS_ENVIRONMENT now), or set
the envvars from the shell.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agolog: replace "priority" by "level" everywhere
Ran Benita [Mon, 27 Aug 2012 21:42:59 +0000 (00:42 +0300)]
log: replace "priority" by "level" everywhere

Now that we don't use syslog, "level" does sound more commonplace. We
should change it while there is still nobody using it.
Also leave some space between the integers of the xkb_log_level enum
values, if we ever need to shove more in between.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoaction: fix LookupEntry for "lockdevbtn"
Ran Benita [Mon, 3 Sep 2012 07:28:58 +0000 (10:28 +0300)]
action: fix LookupEntry for "lockdevbtn"

xkbcomp has that bug as well, guess no one uses it.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoaction: remove redundant check
Ran Benita [Sun, 2 Sep 2012 13:33:02 +0000 (16:33 +0300)]
action: remove redundant check

The NoAction handler always errors out with the same message.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoaction: keep array of default actions, instead of list of changes
Ran Benita [Sun, 2 Sep 2012 08:29:31 +0000 (11:29 +0300)]
action: keep array of default actions, instead of list of changes

The implementation of changing the default properties of actions, e.g. a
statements such as (from test/data/compat/basic):
    setMods.clearLocks= True;
    latchMods.clearLocks= True;
    latchMods.latchToLock= True;
works by keeping a list of ActionInfo's, each containing the neccesary
info from each statement, and then when some action comes up (e.g. in an
interpret statment) it goes through the list, and applies the relevent
ActionInfo's to the newly-constructed xkb_action.

Instead of doing this, we add a struct ActionsInfo, which contains an
array of xkb_actions, one for each type. When a default changing
statement appears, we change the action in the array; when a new action
comes up, we just copy from the array. This is simpler to figure out,
and pretty straightforward.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoaction: disallow setting an action_field default for all types
Ran Benita [Sun, 2 Sep 2012 13:16:57 +0000 (16:16 +0300)]
action: disallow setting an action_field default for all types

Currently where it is possible to write:
    setMods.clearLocks = True;
It's also possible to write:
    action.clearLocks = True;
This will set the default value for the clearLocks action field for
*all* action types, as opposed to just setMods in this case. If
subsequently an action is used for which this field does not make sense,
it will error out.
This doesn't make any sense, because any given field is only possible by at
most 3 or 4 action types... which you might as well write explicitly and
avoid the side effect mentioned above.
Needless to say this is one of xkbcomp's "hidden features" and is not
used anywhere; remove it.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoaction: clean up formatting of extern functions
Ran Benita [Sun, 2 Sep 2012 07:35:08 +0000 (10:35 +0300)]
action: clean up formatting of extern functions

Make it a bit easier to understand what they do.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agotypes: don't strdup a default name
Ran Benita [Fri, 31 Aug 2012 09:12:13 +0000 (12:12 +0300)]
types: don't strdup a default name

None of the other files does that.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agotypes: use darray for KeyTypesInfo instead of list
Ran Benita [Fri, 31 Aug 2012 09:06:50 +0000 (12:06 +0300)]
types: use darray for KeyTypesInfo instead of list

Simpler, uses less memory and more efficient.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agotypes: remove outdated comments
Ran Benita [Mon, 27 Aug 2012 20:00:07 +0000 (23:00 +0300)]
types: remove outdated comments

The code is pretty straightforward now...

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agokeymap: optimize FindInterpsForKey a bit
Ran Benita [Mon, 3 Sep 2012 07:23:44 +0000 (10:23 +0300)]
keymap: optimize FindInterpsForKey a bit

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agokeymap: refactor ApplyInterpsToKey
Ran Benita [Wed, 29 Aug 2012 21:12:41 +0000 (00:12 +0300)]
keymap: refactor ApplyInterpsToKey

There's really not much to explain what the function used to do there...

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoAdd xkbcomp/keymap.c and move some code there
Ran Benita [Wed, 29 Aug 2012 12:02:40 +0000 (15:02 +0300)]
Add xkbcomp/keymap.c and move some code there

Add CompileKeymap to do most of what compile_keymap_file does now, and
move UpdateKeymapFromModifiers along with it from (mostly unrelated)
compat.c.
Also rename UpdateKeymapFromModifiers to UpdateDerivedKeymapFields,
because it does more than update the modifiers.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoMove enum xkb_file_type to xkbcomp/ast.h
Ran Benita [Mon, 27 Aug 2012 16:20:41 +0000 (19:20 +0300)]
Move enum xkb_file_type to xkbcomp/ast.h

This is a more suitable place for this enum, since it's internal to
xkbcomp.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoxkbcomp: clean up compile_keymap function
Ran Benita [Mon, 27 Aug 2012 15:58:36 +0000 (18:58 +0300)]
xkbcomp: clean up compile_keymap function

We make the xkb_file_type enum sequential instead of masks, and then
we don't have to repeat the file types several times in the function.
Makes the code cleaner.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agomap, state: check for KeycodeInRange only in API functions
Ran Benita [Thu, 30 Aug 2012 09:14:29 +0000 (12:14 +0300)]
map, state: check for KeycodeInRange only in API functions

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agostate: fix mod_names_are_active
Ran Benita [Thu, 30 Aug 2012 09:13:37 +0000 (12:13 +0300)]
state: fix mod_names_are_active

This function was always returning -1.
Adding a test, we see that test/state.c treat the is_active functions as
returning booleans, which would treat -1 as success, so we test for > 0
instead (most users would probably get this wrong as well...).
Also update the documentation for the are_active functions, and add a
ATTR_NULL_SENTINEL for gcc __attribute__((sentinel)).

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agostate: fix type confusion within xkb_state_update_mask
Ran Benita [Thu, 30 Aug 2012 08:21:03 +0000 (11:21 +0300)]
state: fix type confusion within xkb_state_update_mask

idx should be xkb_mod_index_t, while mod is the mask.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agostate: remove unneeded optimization
Ran Benita [Thu, 30 Aug 2012 14:15:39 +0000 (17:15 +0300)]
state: remove unneeded optimization

The code that follows does exactly that.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agostate: light indicator when either condition is satisfied
Ran Benita [Mon, 27 Aug 2012 08:51:37 +0000 (11:51 +0300)]
state: light indicator when either condition is satisfied

For the indicator to be set, it is sufficient for at least one of the
group, modifier, or control state to match; this is in line with the
xkblib spec, section 8.2 and ComputeAutoState() in xserver/xkb/xkbLEDs.c
(though the xserver implementation differs from the spec on some
points...).

This also adds a tiny optimization to skip the entire check if the mask
is empty.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agostate: fix led_update_all group mask calculation
Ran Benita [Mon, 27 Aug 2012 08:25:23 +0000 (11:25 +0300)]
state: fix led_update_all group mask calculation

The one above uses which_mods, this one should use which_groups.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agokeycodes: remove outdated comments
Ran Benita [Mon, 27 Aug 2012 20:05:04 +0000 (23:05 +0300)]
keycodes: remove outdated comments

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agokeycodes: use darray for aliases instead of list
Ran Benita [Fri, 31 Aug 2012 15:15:01 +0000 (18:15 +0300)]
keycodes: use darray for aliases instead of list

Uses slightly more memory, but worth it.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agokeycodes: use array for indicator names instead of list
Ran Benita [Fri, 31 Aug 2012 08:40:35 +0000 (11:40 +0300)]
keycodes: use array for indicator names instead of list

Using a simple array here to mirror keymap->indicator_names makes much
more sense, and is simpler.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agokeycodes: ignore "virtual" in indicators
Ran Benita [Fri, 31 Aug 2012 08:13:24 +0000 (11:13 +0300)]
keycodes: ignore "virtual" in indicators

The distinction between real/virtual indicators is useless for us, we
can just ignore it.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agosymbols: call deinit functions Clear instead of Free
Ran Benita [Tue, 28 Aug 2012 11:45:03 +0000 (14:45 +0300)]
symbols: call deinit functions Clear instead of Free

Everywhere else Free is reserved for when the argument is free'd as
well.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agosymbols: remove comparison of unsigned >= 0
Ran Benita [Mon, 27 Aug 2012 06:37:01 +0000 (09:37 +0300)]
symbols: remove comparison of unsigned >= 0

clang warning.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agovmod: remove outdated comments
Ran Benita [Thu, 30 Aug 2012 10:49:23 +0000 (13:49 +0300)]
vmod: remove outdated comments

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agovmod: ClearVModInfo doesn't need the keymap
Ran Benita [Tue, 28 Aug 2012 08:14:54 +0000 (11:14 +0300)]
vmod: ClearVModInfo doesn't need the keymap

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agovmod: remove useless keymap initialization
Ran Benita [Tue, 28 Aug 2012 08:11:40 +0000 (11:11 +0300)]
vmod: remove useless keymap initialization

keymap->vmods is not touched until UpdateModifiersFromCompat,
where it initialized and used.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agovmod: remove support for resolving integer to virtual modifier
Ran Benita [Wed, 15 Aug 2012 19:07:37 +0000 (22:07 +0300)]
vmod: remove support for resolving integer to virtual modifier

This is only relevant to the virtualModifier= statement in interpret
statements in xkb_compat. The current code allows to write e.g.
    virtualModifier = 4
to get the virtual modifier whose index happens to be 4 (possibly
declared in other files or sections, i.e. xkb_types). Doing this is
undeterministic, will break without notice, and not used anywhere.
Don't allow it.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoexpr: move op_type/value_type_to_string functions to ast
Ran Benita [Wed, 15 Aug 2012 18:45:02 +0000 (21:45 +0300)]
expr: move op_type/value_type_to_string functions to ast

Generally the enum-to-string function should appear where the enum is
defined.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agovmod: remove support for direct vmod -> real mod mapping
Ran Benita [Tue, 14 Aug 2012 14:11:16 +0000 (17:11 +0300)]
vmod: remove support for direct vmod -> real mod mapping

The current code supports statements such as:
    virtual_modifiers NumLock = Mod2;
This would set the mapping from the NumLock vmod to the Mod2 real mod
directly, without going through the virtualModifier field in an
interpret statement (in xkb_compat) or vmods field in a key statement
(in xkb_symbols).

This is undocumented, unused and complicates things, so remove it.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoxkbcomp: seperate keymap-copying code from Compile functions
Ran Benita [Wed, 15 Aug 2012 07:10:44 +0000 (10:10 +0300)]
xkbcomp: seperate keymap-copying code from Compile functions

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: only compute 'bool report' once
Ran Benita [Sun, 2 Sep 2012 16:16:34 +0000 (19:16 +0300)]
compat: only compute 'bool report' once

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: disallow changing global defaults from within an interpret
Ran Benita [Sun, 2 Sep 2012 08:49:43 +0000 (11:49 +0300)]
compat: disallow changing global defaults from within an interpret

It's currently possible to write something like this:
    interpret Num_Lock+Any {
        virtualModifier = NumLock;
        action = LockMods(modifiers=NumLock);
        !indicator.allowExplicit;
    };
The final statement has the same effect as writing it in the global file
scope, which changes the default indicator (which all subsequent
indicators start off as). This very strange and also unused; if someone
does it he probably expects it to affect only the local scope, and he
might then get unexpected behavior. So don't allow it.

Also, HandleInterpVar is clearly a misnomer (as it can also change
indicator defaults) so rename it.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: use darray instead of list for interps
Ran Benita [Mon, 27 Aug 2012 19:06:50 +0000 (22:06 +0300)]
compat: use darray instead of list for interps

No need for a list here.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: remove "flags" field from xkb_indicator_map
Ran Benita [Sun, 2 Sep 2012 15:51:26 +0000 (18:51 +0300)]
compat: remove "flags" field from xkb_indicator_map

We don't set this field any more.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: ignore "allowExplicit" in indicator statements
Ran Benita [Thu, 30 Aug 2012 15:56:24 +0000 (18:56 +0300)]
compat: ignore "allowExplicit" in indicator statements

Using !allowExplicit sets the XkbIM_NoExplicit flag of the indicator,
which means that an XKB client cannot change the state of the indicator
using e.g. XkbSetNamedIndicator().
We do not support changing the state of an indicator; furthermore doing
it is probably only useful in conjunction with led-drives-keyboard
behavior, which we also do not support. This is because setting an
indicator without led-drives-keyboard would make the indicator and the
modifier/group it's bound to to get out of sync.

We can re-add this if we need this info.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: ignore "ledDrivesKbd" in indicator statements
Ran Benita [Thu, 30 Aug 2012 13:39:33 +0000 (16:39 +0300)]
compat: ignore "ledDrivesKbd" in indicator statements

We don't support it, as mentioned in the README, so we should stop
processing it and print a message about it.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: ignore "group" (compatibility) statements
Ran Benita [Mon, 27 Aug 2012 18:31:18 +0000 (21:31 +0300)]
compat: ignore "group" (compatibility) statements

Group compatibility statements are like the following:
    group 3 = AltGr;
This currently results in:
    keymap->groups[2].mask = <real mod mapped from AltGr vmod>
And we don't do any thing with this value later. The reason it exists in
XKB is to support non-XKB clients (i.e. XKB support disabled entirely in
the server), which do not know the concept of "group", and use some
modifier to distinguish between the first and second keyboard layouts
(usually with the AltGr key). We don't care about all of that, so we can
forget about it.

One artifact of this removal is that xkb_map_num_groups no longer
works, because it counted through keymap->groups (this wasn't entirely
correct BTW). Instead we add a new num_groups member to the keymap,
which just hold the maximum among the xkb_key's num_groups. This also
means we don't have to compute anything just to get the number of
groups.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: get rid of BindIndicators
Ran Benita [Mon, 27 Aug 2012 10:21:03 +0000 (13:21 +0300)]
compat: get rid of BindIndicators

Now that 1fba6189e67 removed support for binding indicator maps by index
instead of name, we can remove some more magic which happens now: if an
indicator map specifies an indicator name which was not previously
declared in a 'indicator 5 = "Caps Lock"'-like statement in
xkb_keycodes, we can just look at the next free index and assign it.

This also allows us to use a darray for the LEDInfo's instead of a list.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: ignore "index" field in indicator statements
Ran Benita [Mon, 27 Aug 2012 09:29:57 +0000 (12:29 +0300)]
compat: ignore "index" field in indicator statements

The current code allows to set the "index" field in an indicator
statment's body. This would bind the indicator to the specified index,
instead of by name (which was declared previously in xkb_keycodes).
Doing this is a bad idea, for the same reasons as in 3cd9704, and is
also happily not used anywhere.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: remove dead NoAutomatic code
Ran Benita [Mon, 27 Aug 2012 08:38:44 +0000 (11:38 +0300)]
compat: remove dead NoAutomatic code

The xkblib spec, table 7.1 (indicators), says:
XkbIM_NoAutomatic: Xkb does not automatically change the value of the
                   indicator based upon a change in the keyboard state,
                   regardless of the values for the other fields of the
                   indicator map.

xkbcomp (the real one) never actually implemented a way for an indicator
statement to set this flag, so it's just dead unused code. We definitely
don't want to implement it ourselves, so remove any mention of it.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: ignore "locking" field in sym interprets
Ran Benita [Tue, 14 Aug 2012 13:26:30 +0000 (16:26 +0300)]
compat: ignore "locking" field in sym interprets

This field is used in conjunction with key behaviors, which we don't
support since c1ea23da5. This is also unused in xkeyboard-config.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: small changes
Ran Benita [Tue, 14 Aug 2012 12:06:11 +0000 (15:06 +0300)]
compat: small changes

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agocompat: add general overview
Ran Benita [Tue, 14 Aug 2012 12:05:56 +0000 (15:05 +0300)]
compat: add general overview

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoaction: convert action field type to enum
Ran Benita [Sat, 11 Aug 2012 08:54:05 +0000 (11:54 +0300)]
action: convert action field type to enum

We can also hide the ActionInfo definition inside action.c.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agotypes: add "Effects on keymap" to overview
Ran Benita [Tue, 14 Aug 2012 11:42:57 +0000 (14:42 +0300)]
types: add "Effects on keymap" to overview

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agorulescomp: remove bad failtests
Ran Benita [Sun, 12 Aug 2012 15:16:52 +0000 (18:16 +0300)]
rulescomp: remove bad failtests

Since we now handle empty model/layout, the last couple of tests should
not fail. The reason they do is bacause they try to use a non-existent
"base" rules file. When the file is brought in these tests do not fail.
Since we already test for non-existent rules file, we can remove them,
and refine the other tests a bit.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoOrganize xkbcomp/ header files
Ran Benita [Mon, 13 Aug 2012 08:00:43 +0000 (11:00 +0300)]
Organize xkbcomp/ header files

Various non-functional changes:
- Re-add keycodes.h and move some stuff there.
- Add parser-priv.h for internal bison/flex stuff.
- Don't include headers from other headers, such that file dependencies
  are immediate in each file.
- Rename xkbcomp.h -> ast.h, parseutils.{c,h} -> ast-build.{c,h}
- Rename path.{c,h} -> include.{c,h}
- Rename keytypes.c -> types.c
- Make the naming of XkbFile-related functions more consistent.
- Move xkb_map_{new,ref,unref} to map.c.
- Remove most extern keyword from function declarations, it's just
  noise (XKB_EXPORT is what's important here).
- Append XKBCOMP_ to include guards.
- Shuffle some code around to make all of this work.

Splitting this would be a headache..

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoAllocate xkb_component_names on stack
Ran Benita [Tue, 14 Aug 2012 08:49:19 +0000 (11:49 +0300)]
Allocate xkb_component_names on stack

Instead of malloc'ing it as well. Also improve the error handling.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoMove ISEMPTY to utils.h
Ran Benita [Mon, 13 Aug 2012 10:49:17 +0000 (13:49 +0300)]
Move ISEMPTY to utils.h

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoMove 'no symbols defined for ...' message to a warning
Daniel Stone [Thu, 9 Aug 2012 14:47:53 +0000 (16:47 +0200)]
Move 'no symbols defined for ...' message to a warning

Shut up shut up shut up shut up shut up shut up.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
12 years agoCombine a couple of macros
Ran Benita [Fri, 10 Aug 2012 19:48:18 +0000 (22:48 +0300)]
Combine a couple of macros

Easier to see what it does without the trivial macros.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoaction: get rid of xkb_any_action
Ran Benita [Fri, 10 Aug 2012 19:38:07 +0000 (22:38 +0300)]
action: get rid of xkb_any_action

And use union xkb_action instead. We add xkb_private_action, which is
the same as xkb_any_action, but only used where the intention is clear.
This should take care of whatever sizing changes the action struct might
have.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoRemove XkbKeyTypeIndex and widen index type
Ran Benita [Fri, 10 Aug 2012 19:06:12 +0000 (22:06 +0300)]
Remove XkbKeyTypeIndex and widen index type

We don't need the macro, and using char for the kt_index is imaginably
too small.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoStore actions inside struct xkb_key
Ran Benita [Fri, 10 Aug 2012 15:14:35 +0000 (18:14 +0300)]
Store actions inside struct xkb_key

Cuts out a lot of useless redirection and space.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agokeycodes: save context in Info, not keymap
Ran Benita [Fri, 10 Aug 2012 10:32:58 +0000 (13:32 +0300)]
keycodes: save context in Info, not keymap

We don't need the keymap in this case, just makes things more verbose.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoRemove xkbcomp/misc.c
Ran Benita [Fri, 10 Aug 2012 10:30:44 +0000 (13:30 +0300)]
Remove xkbcomp/misc.c

The KeyName functions are more appropriate in keycodes.c.
The ProcessIncludeFile can go to path.c along with the other functions
dealing with includes.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoRemove left over keycodes.h
Ran Benita [Fri, 10 Aug 2012 10:26:36 +0000 (13:26 +0300)]
Remove left over keycodes.h

For some reason we still track this file in git even though we don't use
it any more.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoRemove AutoKeyNames feature
Ran Benita [Fri, 10 Aug 2012 10:08:03 +0000 (13:08 +0300)]
Remove AutoKeyNames feature

If this keymap flag is set, whenever a key name appears in one of the
sections which does not exist (i.e. has not been declared in keycodes),
it finds the first unused keycode and attaches it that name.

This might have been useful when you could compile the symbols section
or geometry section without a keycodes section, but we don't support
this anymore. It's also pretty useless for any real work, because the
user has no way of knowing the keycode and so it will never be used.
Finally the only obscure way left to set this flag is by including a
keycodes file called "computed".

Just remove it.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agomap: share some code
Ran Benita [Fri, 10 Aug 2012 07:17:32 +0000 (10:17 +0300)]
map: share some code

Make more extensive use of get_entry_for_key_state, and add
key_get_consumed to use in the other consume functions.

There's also a slight change in the consumed mods calculations, where
we use entry->mods.mask instead of type->mods.mask. The original was
copied from what libX11 does but what we do now is more logically
correct. The result is exactly the same though because:
type->mods.mask ⊇ entry->mods.mask ⊇ entry->preserve.mask

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoUse XKB_{GROUP,LEVEL}_INVALID instead of -1 for errors
Ran Benita [Fri, 10 Aug 2012 07:11:49 +0000 (10:11 +0300)]
Use XKB_{GROUP,LEVEL}_INVALID instead of -1 for errors

The group/level types are unsigned, so it's odd to return -1 for them.
Instead use their invalid values (which happen to be == -1).

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agostate: use global static const for fake action
Ran Benita [Fri, 10 Aug 2012 07:08:20 +0000 (10:08 +0300)]
state: use global static const for fake action

Requires constifying some arguments.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoModernize struct xkb_mods
Ran Benita [Wed, 8 Aug 2012 23:33:51 +0000 (02:33 +0300)]
Modernize struct xkb_mods

Currently xkb_mods has the following members:
- uint8_t real_mods - 8 X11 core mods
- xkb_mod_mask_t vmods - 16 virtual mods, zero-based index
- xkb_mod_mask_t mask - the computed effective *real* modifier mask,
  basically a cache for the first two which is:
  real_mods | real mods computed from vmods

Our API acts on masks which combine the real_mods and vmods into a
single value, which is:
8 first bits real mods | 16 next bits virtual mods
(XkbNumModifiers = 8, XkbNumVirtualMods = 16). This is also the format
which ResolveVModMask uses (which is where all the modifier masks really
"come from", e.g. "Shift+Lock+Level5" -> xkb_mod_mask_t).

What the code does now after getting the mask from ResolveVModMask, is
to break it into real part and virtual part and store them seperately,
and then join them back together when the effective mask is calculated.

This is all pretty useless work. We change xkb_mods to the following:
- xkb_mod_mask_t mods - usually what ResolveVModMask returns
- xkb_mod_mask_t mask - the computed mask cache
And try to consistently use the word "mods" for the original,
non-effective mods and mask for the effective mods (which can only
contain real mods for now, because things break otherwise).

The separation is also made clearer. The effective masks are computed by
UpdateModifiersFromCompat after all the sections have been compiled;
before this the mask field is never touched; after this (i.e. map.c and
state.c) the original mods field is never touched. This single execption
to this rule is keymap-dump.c: it needs to print out only the original
modifiers, not computed ones. This is also the reason why we actually
keep two fields instead keeping one and modifying it in place.

The next logical step is probably to turn the real mods into vmods
themselves, and get rid of the distinction entirely (in a compatible
way).

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoFix xkb_keymap::vmods type
Ran Benita [Wed, 8 Aug 2012 23:33:51 +0000 (02:33 +0300)]
Fix xkb_keymap::vmods type

It maps a vmod to a mask, of course.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agotypes: don't compute effective masks here as well
Ran Benita [Wed, 8 Aug 2012 22:30:22 +0000 (01:30 +0300)]
types: don't compute effective masks here as well

After compiling all of the sections, UpdateModifiersFromCompat does all
of the vmod -> real mods translations, including types/kt_entries.

keytypes.c also has code that does that, but it's unneeded:
- Later sections don't look at their effective masks, so doing it later
  is fine.
- When this code is executed, the vmods -> real mods mapping is empty
  (that is set up later), so VModsToReal has no effect here.
So we can just remove it.

However UpdateModifiersFromCompat didn't update the preserve mask, so do
that.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agotypes: get rid of PreserveInfo
Ran Benita [Wed, 8 Aug 2012 16:53:55 +0000 (19:53 +0300)]
types: get rid of PreserveInfo

We don't need the indirection. We store the preserve mask directly in
the entry, and create a new one if it doesn't exists (which is exactly
what the current code does in a roundabout way).

Incidentally this fixes a bug where the effective modifier mask of the
entries' preserve[] wasn't calculated, so the virtual modifiers had no
effect there.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agokeymap-dump: use VModMaskText
Ran Benita [Wed, 8 Aug 2012 17:47:51 +0000 (20:47 +0300)]
keymap-dump: use VModMaskText

The difference between the two are irrelevant here.

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoFix warning
Ran Benita [Wed, 8 Aug 2012 22:55:30 +0000 (01:55 +0300)]
Fix warning

Signed-off-by: Ran Benita <ran234@gmail.com>
12 years agoAdd xkb_map_mod_mask_remove_consumed
Daniel Stone [Wed, 8 Aug 2012 12:26:23 +0000 (14:26 +0200)]
Add xkb_map_mod_mask_remove_consumed

A fairly simple helper which, given an xkb_mod_mask_t, removes all
modifiers which are consumed during processing of a particular key.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
12 years agoAdd xkb_log_level enum rather than using syslog
Daniel Stone [Wed, 8 Aug 2012 12:01:46 +0000 (14:01 +0200)]
Add xkb_log_level enum rather than using syslog

Instead of relying on people including syslog.h, add our own
XKB_LOG_LEVEL_* defines.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
12 years agoIncrease log verbosity in tests
Daniel Stone [Wed, 8 Aug 2012 11:56:28 +0000 (13:56 +0200)]
Increase log verbosity in tests

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
12 years agoRules: mmap() rules file instead of using getc()
Daniel Stone [Fri, 3 Aug 2012 03:13:46 +0000 (05:13 +0200)]
Rules: mmap() rules file instead of using getc()

Good for a small performance win on my system.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
12 years agoandroid: add build files
Pekka Paalanen [Mon, 14 May 2012 11:33:29 +0000 (14:33 +0300)]
android: add build files

squashed: android: set xkb config path

Conflicts:
Makefile.am

12 years agostringcomp: Make test more punishing
Daniel Stone [Fri, 3 Aug 2012 02:07:33 +0000 (04:07 +0200)]
stringcomp: Make test more punishing

Recreate the old test/dump scenario, where we test the following map:
  - rules: evdev
  - model: pc104
  - layout #1: us
  - layout #2: ru
  - layout #3: ca(multix)
  - layout #4: de(neo)

This is ever so slightly altered from the xkbcomp output; running the
following:
setxkbmap -rules evdev -model pc105 -layout us,ru,ca,de -variant
,,multix,neo -print | xkbcomp -xkb - -

will give you a map with RCTL added to the modifier_map for both Control
and Mod3.  Running the output through xkbcomp -xkb - - again, will give
you RCTL only added to Mod3.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
12 years agotest/dump: Remove superfluous test
Daniel Stone [Fri, 3 Aug 2012 02:02:31 +0000 (04:02 +0200)]
test/dump: Remove superfluous test

No longer necessary now we have stringcomp doing a full round-trip test
for us.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
12 years agotest/stringcomp: Perform full round-trip test
Daniel Stone [Fri, 3 Aug 2012 02:01:21 +0000 (04:01 +0200)]
test/stringcomp: Perform full round-trip test

We now pass! \o/

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
12 years agostringcomp: Remove unnecessary Level1 mappings
Daniel Stone [Fri, 3 Aug 2012 01:54:44 +0000 (03:54 +0200)]
stringcomp: Remove unnecessary Level1 mappings

As a map will implicitly go to level one unless explicitly mentioned
otherwise, remove all explicit =Level1 mappings, except for those with
preserve entries.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>