README: move compatibility notes to their own page

Signed-off-by: Ran Benita <ran234@gmail.com>

README: link to online API docs

Signed-off-by: Ran Benita <ran234@gmail.com>

README: move note about xkeyboard-config to its own section

Signed-off-by: Ran Benita <ran234@gmail.com>

README: remove section on API stability

This was useful when we *weren't* stable. But now that we are this
should be obvious.

Signed-off-by: Ran Benita <ran234@gmail.com>

Replace some strncmp's with memcmp

Signed-off-by: Ran Benita <ran234@gmail.com>

compose/parser: save len in keysym_from_name cache

This reduces a lot of strcmp's, and allows to use a faster memcmp.

Signed-off-by: Ran Benita <ran234@gmail.com>

test/keyseq: test 'map[None] = Level2;' scenario

See previous commit for an explanation.

Signed-off-by: Ran Benita <ran234@gmail.com>

test/data: add files for model=applealu_ansi layout=us

Signed-off-by: Ran Benita <ran234@gmail.com>

test/data: sync to xkeyboard-config 2.13

(Run ./test/data/sync.sh).

Signed-off-by: Ran Benita <ran234@gmail.com>

state: correctly infer inactive type entries

The current test is incorrect, since 'map[None]' is entirely valid. In
most cases this doesn't cause any problems, since the default fallback
is Level1, and it's almost always 'map[None] = Level1' anyway. But in
one case in xkeyboard-config it isn't, in types/numpad(mac):

    type "KEYPAD" {
        modifiers = None;
        map[None] = Level2;
        level_name[Level2] = "Number";
    };

So before checking if no modifiers were mapped, make sure there *were*
any modifiers at all.

https://bugs.freedesktop.org/show_bug.cgi?id=85092

Reported-by: Gatis Paeglis <gatis.paeglis@digia.com>
Signed-off-by: Ran Benita <ran234@gmail.com>

symbols: don't warn about conflicting syms if they are the same

Signed-off-by: Ran Benita <ran234@gmail.com>

parser: bring back warning about includes of files with no default

Using the same format as xkbcomp.

Signed-off-by: Ran Benita <ran234@gmail.com>

Merge pull request #14 from jasperla/undefined

disable -Wl,--no-undefined on OpenBSD for it doesn't have DT_NEEDED for libc

disable -Wl,--no-undefined on OpenBSD for it doesn't have DT_NEEDED for libc

Merge pull request #13 from jasperla/librt

Make librt optional, required for at least OpenBSD

Make librt optional, required for at least OpenBSD

test/compose: test include statement

Signed-off-by: Ran Benita <ran234@gmail.com>

test/compose: test modifier syntax

Signed-off-by: Ran Benita <ran234@gmail.com>

compose/parser: fix parsing of multiple modifiers

Signed-off-by: Ran Benita <ran234@gmail.com>

compose/parser: parse (! mods) properly

We don't actually do anything with them. But if someone uses them we can
at least not choke.

Signed-off-by: Ran Benita <ran234@gmail.com>

compose/parser: resolve keysyms in parser instead of scanner

It will become context-sensitive.

Signed-off-by: Ran Benita <ran234@gmail.com>

compose/parser: use parameter as intended

Signed-off-by: Ran Benita <ran234@gmail.com>

compose/parser: one more skip_to_eol()

Signed-off-by: Ran Benita <ran234@gmail.com>

compose/parser: fix segfault when including

The keysym cache for the new scanner was not initialized.
To avoid such errors also in the future, require passing the priv
argument in scanner_init(), instead of initializing it separately.

Signed-off-by: Ran Benita <ran234@gmail.com>

COPYING: add copyright notice from libX11:modules/im/ximcp/imLcPrs.c

We have used some portions of it, so add the notice.

Signed-off-by: Ran Benita <ran234@gmail.com>

scanner-utils: optimize one-line comments

Compose files have a lot of those.

Signed-off-by: Ran Benita <ran234@gmail.com>

bench/compose: add new benchmark

Signed-off-by: Ran Benita <ran234@gmail.com>

test/interactive-evdev: add compose support

To try, do e.g.:
sudo ./test/interactive-evdev -l us -v intl -o compose:ralt -d

Signed-off-by: Ran Benita <ran234@gmail.com>

test/compose: add new test

Some results from the benchmark (compilation of en_US.UTF-8/Compose):

$ grep 'model name' /proc/cpuinfo
model name  : Intel(R) Core(TM)2 Duo CPU     E8400  @ 3.00GHz
model name  : Intel(R) Core(TM)2 Duo CPU     E8400  @ 3.00GHz

$ uname -a
Linux ran 3.16.1-1-ARCH #1 SMP PREEMPT Thu Aug 14 07:40:19 CEST 2014 x86_64 GNU/Linux

$ ./test/compose bench
compiled 1000 compose tables in 7.776488331s

So according to the above benchmark and valgrind --tool=massif, an
xkb_compose_table adds an overhead of about ~8ms time and ~130KB
resident memory.
For contrast, a plain US keymap adds an overhead of ~3ms time and 90KB
resident memory.

Signed-off-by: Ran Benita <ran234@gmail.com>

compose: add xkbcommon-compose - implementation

Signed-off-by: Ran Benita <ran234@gmail.com>

compose: add xkbcommon-compose - API

xkbcommon-compose is a Compose implementation for xkbcommon. It mostly
behaves like libX11's Compose, but the support is somewhat low-level and
is not transparent like in libX11. The user must add some supporting code
in order to utilize it.

The intended audience are users who use xkbcommon but not a full-blown
input method. With this they can add Compose support in a straightforward
manner, so they have a fairly complete keyboard input for Latin-like
languages at least.

See the header documentation for details.

Signed-off-by: Ran Benita <ran234@gmail.com>

scanner-utils: add priv member

For when a user of the scanner wants to pass something along with it.

Signed-off-by: Ran Benita <ran234@gmail.com>

darray: add darray_shrink()

If we have a big array which can be finalized, on average we can give
back 1/4 of its size, which the allocator might be able to use.

Signed-off-by: Ran Benita <ran234@gmail.com>

keysym: add function to test if a keysym is for a modifier

Needed for compose.

Signed-off-by: Ran Benita <ran234@gmail.com>

scanner-utils: add helper for appending an entire string

Signed-off-by: Ran Benita <ran234@gmail.com>

scanner-utils: add helper for hex string escape

Like the already existing oct.

Signed-off-by: Ran Benita <ran234@gmail.com>

Move benchmarks from tests to their own files in bench/

The tests only contain tests, and the benchmarks are more visible.

Signed-off-by: Ran Benita <ran234@gmail.com>

build: don't rebuild everything for libtest.la

We added a -fvisibilty=default to the libtest.la CFLAGS. When automake
sees that the CFLAGS are different from those of libxkbcommon.la, it
figures it should recompile all of the $(libxkbcommon_la_SOURCES),
instead of reusing the already-compiled object files.

But actually, the convenience library is not a shared object, it's just
used as an archive, thus the -fvisibility is entirely pointless. So
avoid the recompilations by just removing it.

Signed-off-by: Ran Benita <ran234@gmail.com>

scanner-utils: optimize str()/lit()

Replace the dog-slow unneeded strncasecmp() with an inlineable memcmp().

Before:
compiled 2500 keymaps in 8.348715629s

After:
compiled 2500 keymaps in 7.872640338s

Signed-off-by: Ran Benita <ran234@gmail.com>

keymap: rename XkbKeyGroupWidth to XkbKeyNumLevels

The "width" terminology comes from the group*width+level layout of the
keysyms in a key, as used in the old implementations. We don't keep all
the keysyms of a key in one array so change it to a more accurate name.

Signed-off-by: Ran Benita <ran234@gmail.com>

symbols: improve FindKeyForSymbol()

A bit more involved, but can short circuit.

Signed-off-by: Ran Benita <ran234@gmail.com>

symbols: use correct max value

xkb_level_index_t was initially uint16_t, now it's 32.

Signed-off-by: Ran Benita <ran234@gmail.com>

doc: disable alphabetical data structure index

It is useless - we have very little data structures.

Signed-off-by: Ran Benita <ran234@gmail.com>

doc: disable BRIEF_MEMBER_DESC

It's easier to quickly go through the functions in a module this way.
The names provide enough "brief description" to get someone to click for
the full description.

Signed-off-by: Ran Benita <ran234@gmail.com>

doc: disable tree view

It is redundant with the header navigation.

Signed-off-by: Ran Benita <ran234@gmail.com>

doc: linkify some references

Signed-off-by: Ran Benita <ran234@gmail.com>

x11, doc: add note about Key{Press,Release}->state field

Signed-off-by: Ran Benita <ran234@gmail.com>

x11, doc: add note about per-client flags like detectable autorepeat

Signed-off-by: Ran Benita <ran234@gmail.com>

keymap-dump: combine modifier_map's with the same modifier

A bit less efficient, but makes for shorter, nicer output.

Signed-off-by: Ran Benita <ran234@gmail.com>

rules: fix mlvo-not-used warning

An mlvo can also be used in an expansion, but we didn't mark them in
this case in commit d8a4f52cb95d989b4. This caused wrongful warnings on
something like -l ch -v fr -- the `fr` is only added via expansion.

Signed-off-by: Ran Benita <ran234@gmail.com>

quick-guide: fix xkb_state_key_get_syms() example

Signed-off-by: Ran Benita <ran234@gmail.com>

quick-guide: fix Java-esque array syntax

Signed-off-by: Ran Benita <ran234@gmail.com>

rules: warn when an RMLVO component isn't used

Due to wildcard matches in the rules file, this is only really useful
for misspelled or missing options, e.g.

$ ./test/rmlvo-to-kccgst -o comprose:ralt > /dev/null
xkbcommon: ERROR: Unrecognized RMLVO option "comprose:ralt" was ignored

Although it is more of a warning, it indicates a misconfiguration which
the user probably wants to see. Therefore the log level is ERROR.

Signed-off-by: Ran Benita <ran234@gmail.com>

parser: use "atom" instead of "sval" in yylval

"sval" is already used for "struct sval".

Signed-off-by: Ran Benita <ran234@gmail.com>

doc: reorder "Keymap Components" functions

Put the general keymap stuff before key-specific functions.

Signed-off-by: Ran Benita <ran234@gmail.com>

doc: move consumed modifier description to its own section

With small edits.

Signed-off-by: Ran Benita <ran234@gmail.com>

doc: order the modules better

Explicitly specify the files in INPUT so that "X11 Support" appears
after the main modules.

Signed-off-by: Ran Benita <ran234@gmail.com>

x11, doc: fix typo

Signed-off-by: Ran Benita <ran234@gmail.com>

x11, doc: add overview

Signed-off-by: Ran Benita <ran234@gmail.com>

test/keyseq: add test

Signed-off-by: Ran Benita <ran234@gmail.com>

configure.ac: add -Wdocumentation

It's supported by clang and warns about bad doxygen comments.

Signed-off-by: Ran Benita <ran234@gmail.com>

doc: fix @param display for xkb_x11_setup_xkb_extension()

Signed-off-by: Ran Benita <ran234@gmail.com>

symbols: use darray_foreach_from for nicer loop

Signed-off-by: Ran Benita <ran234@gmail.com>

Remove compatibility symbols

Remove the deprecated symbols that were used for ABI compatibility
during the transition period to the first stable version, 0.2.0.

The old *names* can still be used, programs which use the old names will
continue to work, as long as they were compiled against a stable
version (as they have been #defined to the new names from the start; see
xkbcommon/xkbcommon-compat.h). Namely, this will break binaries which:

1. Were compiled against a pre-stable version of libxkbcommon, and
2. Are linked against the next version of libxkbcommon, and
3. Expect to work.

This scenario is very unlikely, and will break in many other ways
anyway. Also, retaining support for these means including them in the
new symbol version file, which I would like to avoid.

Signed-off-by: Ran Benita <ran234@gmail.com>

build: use symbol versioning

Symbol versions provide a means by which ELF utilities can determine
whether a program is incompatible with a too-old library version so
that package management tools can autodetect version-based
dependencies and suggest upgrade paths.

[ran: swap xkbcommon.map and xkbcommon-x11.map]

Signed-off-by: Jan Engelhardt <jengelh@inai.de>
Signed-off-by: Ran Benita <ran234@gmail.com>

keycodes: fix keymap compilation with no aliases and malloc(0)==NULL

If the keymap doesn't have any key-aliases (which is certainly
possible), the calloc(num_key_aliases, ...) is allowed to return NULL
according to the C standard, but this is not an error.

Signed-off-by: Ran Benita <ran234@gmail.com>

ast-build: fix leak in error path

Signed-off-by: Ran Benita <ran234@gmail.com>

test: make most tests portable by copying linux/input.h locally

There is really no reason to deny these tests from different platforms
only for a few #defines.

The only linux-only test (or test program, it is not run by make check)
is interactive-evdev, which actually uses evdev.

Signed-off-by: Ran Benita <ran234@gmail.com>

Remove <linux/input.h> include from test/x11comp.c

libxkbcommon 0.4.3 introduces a new test, x11comp, which does not build
on non-Linux OSes because of the unconditional <linux/input.h> include.

This seems not needed even on Linux, so attached there is a simple patch
to remove it.

https://bugs.freedesktop.org/show_bug.cgi?id=83551

Signed-off-by: Ran Benita <ran234@gmail.com>

Merge pull request #11 from michaelforney/master

Fix typo in default-variant configure option

Fix typo in default-variant configure option

test/x11comp: don't hang if Xvfb is not available

If Xvfb is not present, posix_spawn still forks, but the child fails.
In that case, since we left the write fd of the pipe open in the parent,
we just kept waiting on the read() without noticing that the other side
is dead.

Signed-off-by: Ran Benita <ran234@gmail.com>

keymap: fix typo

Signed-off-by: Ran Benita <ran234@gmail.com>

keymap: fix description of xkb_keymap_key_get_syms_by_level()

The @level argument is restricted by xkb_keymap_num_levels_for_key(). Fix
the description to no longer mention xkb_keymap_num_layouts_for_key().

Signed-off-by: David Herrmann <dh.herrmann@gmail.com>

Bump version to 0.4.3

Signed-off-by: Ran Benita <ran234@gmail.com>

Update NEWS

Signed-off-by: Ran Benita <ran234@gmail.com>

test/state: add test_update_mask() test

Signed-off-by: Ran Benita <ran234@gmail.com>

state: make sure the mods are fully resolved after xkb_state_update_mask()

Virtual modifiers can have "mappings" to real modifiers, e.g. NumLock
may also set Mod2. In a normal turn of events, the various components
(depressed, latched, locked, and consequently effective) include the
mapped mods, because the masks are pre-resolved everywhere. However,
xkb_state_update_mask() accepts arbitrary mod masks, which may not be
resolved (if it comes from somewhere other than
xkb_state_serialize_mods()). So let's always resolve them ourselves.

Signed-off-by: Ran Benita <ran234@gmail.com>

Make the effective mod mask calculation available to other files

We will want to use that function in state.c as well.

Signed-off-by: Ran Benita <ran234@gmail.com>

state: no need for loop in xkb_state_update_mask()

Signed-off-by: Ran Benita <ran234@gmail.com>

test/x11comp: server writes \n to displayfd

Signed-off-by: Ran Benita <ran234@gmail.com>

test/x11-keyseq: new test

It is like test/stringcomp, only instead of using
xkb_keymap_new_from_string(), it uses xkbcomp to upload the keymap to a
dummy Xvfb X server and then xkb_x11_keymap_new_from_device().

If any of these components are not present or fails, the test is shown
as skipped.

The test is messy, fragile, limited and depends on external tools, but I
will improve on that later -- it's better to have a test.

Signed-off-by: Ran Benita <ran234@gmail.com>

x11/keymap: handle private actions

Previously we treated them as NoAction().

Signed-off-by: Ran Benita <ran234@gmail.com>

utils: add a STATIC_ASSERT macro

It'd be nicer to use C11's static_assert(), but it's easier to roll our
own C99 version using a trick I saw in xv6.

Signed-off-by: Ran Benita <ran234@gmail.com>

keymap: remove "flags" field of xkb_private_action

Private actions have no flags - only serialized data.

Signed-off-by: Ran Benita <ran234@gmail.com>

x11/keymap: be more defensive about the number of modifiers

There can be at most 16 vmods, and we rely on the facts that #vmods +
NUM_REAL_MODS (8) <= XKB_MAX_MODS (32) when accessing keymap->mods.mods.
But msb_pos() can potentially return up to #vmods = 32 if the server is
malicious, so we need to truncate it.

Signed-off-by: Ran Benita <ran234@gmail.com>

x11/keymap: don't forget to add the vmod offset in get_vmods

The first 8 modifiers in keymap->mods are the real modifiers; the virtual
 modifiers are then at slots 8-24. But XkbGetMap's virtualMods mask
starts the virtual modifiers at zero, so we need to add an offset (like
we do correctly in get_vmod_names()).

https://github.com/xkbcommon/libxkbcommon/issues/9

Reported-by: @rtcm
Signed-off-by: Ran Benita <ran234@gmail.com>

docs: move keysym-transformations page to a better position

Signed-off-by: Ran Benita <ran234@gmail.com>

docs: fix self-reference

Signed-off-by: Ran Benita <ran234@gmail.com>

interactive-evdev: don't use sysexits.h

Signed-off-by: Ran Benita <ran234@gmail.com>

ast-build: make sure InterpDef is freeable

With the following two rules:

    InterpretDecl   :       INTERPRET InterpretMatch OBRACE
                                VarDeclList
                            CBRACE SEMI
                            { $2->def = $4; $$ = $2; }
                    ;

    InterpretMatch  :       KeySym PLUS Expr
                            { $$ = InterpCreate($1, $3); }
                    |       KeySym
                            { $$ = InterpCreate($1, NULL); }
                    ;

And the fact that InterpCreate doesn't initialize ->def, if the
VarDeclList fails, the %destructor tries to recursively free the
uninitialized ->def VarDef. So always initialize it.

That was the only problematic code in the parser for %destructor (I'm
pretty sure).

Signed-off-by: Ran Benita <ran234@gmail.com>

parser: silence bison "unused value" warnings

Previous commit triggered these for some reason:

/home/ran/src/libxkbcommon/src/xkbcomp/parser.y:555.25-33: warning: unused value: $1 [-Wother]
 CoordList       :       CoordList COMMA Coord
                         ^^^^^^^^^

Signed-off-by: Ran Benita <ran234@gmail.com>

parser: don't leak AST nodes for discarded symbols

If the parser has symbols on the stack, and then enters an error, it
discards the symbols and fails. But their actions which allocate AST
nodes had already ran. So we must free these to avoid leaks.

We use %destructor declarations, see
http://www.gnu.org/software/bison/manual/html_node/Destructor-Decl.html

Note: byacc only supports %destructor when compiled with
--enable-btyacc. Also, it doesn't support using the parse-param in the
destructor. So we might revert this commit before the next release, or
forget about byacc.

https://github.com/xkbcommon/libxkbcommon/issues/8

Signed-off-by: Ran Benita <ran234@gmail.com>

test: add file with a syntax error

We didn't really have any. It also a exposes a memory leak, since the
parser doesn't clean up the AST nodes of the discarded symbols.

Signed-off-by: Ran Benita <ran234@gmail.com>

Replace darray_mem with a new darray_steal

That's a more declarative interface.

Signed-off-by: Ran Benita <ran234@gmail.com>

ast-build: use cast instead of ->common

Missed in 1b2bb204e0baa2246a6232aea762c1edb00cd44a.

Signed-off-by: Ran Benita <ran234@gmail.com>

types: refactor CopyKeyTypesToKeymap

So it's OOM-safe and doesn't clobber keymap on failure.

Signed-off-by: Ran Benita <ran234@gmail.com>

keycodes: split CopyKeyInfoToKeymap to several functions

It's a bit easier to read and self-documenting. Also handles OOM better.

Signed-off-by: Ran Benita <ran234@gmail.com>

parser: don't shadow "str"

It's a name of a function in scanner-utils.h and also of some
parameters.

https://bugs.freedesktop.org/show_bug.cgi?id=79898

Reported-by: Bryce Harrington <b.harrington@samsung.com>
Signed-off-by: Ran Benita <ran234@gmail.com>

x11: don't iterate on empty batches

If count % SIZE == 0 we did a useless iteration where start==stop. It's
harmless but strange, so don't do that.

Signed-off-by: Ran Benita <ran234@gmail.com>