xkbcomp/types: fix types being assigned the wrong number of levels in some circumstances

The buggy code assigned the number of levels based on the number of
level names in the definition, instead of the actual number of levels!
This would completely break type definitions which do not give names to
levels.

This was not noticed for so long because xkeyboard-config always gives
names to all levels.

This regressed in 61fed8dab9b8e27981f36ffc96666d7376546e30.

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

keysym: add xkb_keysym_to_{lower,upper} to public API

These can be useful in some odd cases.

There is already an implementation (+ tests) for internal use, so all
that's needed is to export them.

If xkbcommon were to provide a way to convert a Unicode codepoint to a
keysym, this could have been implemented externally as follows:

    uint32_t codepoint = xkb_keysym_to_utf32(keysym);
    uint32_t upper_codepoint = my_unicode_library_to_upper(codepoint);
    xkb_keysym_t upper_keysym = theoretical_xkb_keysym_from_utf32(upper_codepoint);

However keysym -> codepoint is not injective so such a function is not
possible strictly speaking.

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

test/data: sync with xkeyboard-config 2.22

Some tweaks to the de(neo) keyseq tests were required. It seems to have
improved.

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

Merge pull request #53 from whitslack/missing-ucs-codes

keysym-utf: Add missing codes for signifblank and permille keysyms

keysym-utf: Add missing codes for signifblank and permille keysyms

build/meson: require meson >= 0.41.0

With previous versions, the compilation fails with linker errors.

Fixes https://github.com/xkbcommon/libxkbcommon/issues/52
Reported-by: @rezso
Signed-off-by: Ran Benita <ran234@gmail.com>

expr: paper over a maybe-uninitialized warning

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

test/interactive-wayland: avoid unused function warning due to configuration

test/interactive-wayland.c:95:1: warning: ‘set_cloexec_or_close’ defined but not used [-Wunused-function]

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

build: add missing configure function checks for test/interactive-wayland

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

xkbcommon-compose.h: change recommended locale fallback code to treat empty string same as unset

The previous code would cause failures to find the Compose file if one
of the environment variables LC_ALL, LC_CTYPE or LANG are set to the
empty string.

The description of the fallback procedure in loclale(7) talks about
"non-null environment variable"; I interpreted this to mean the
environment variable is unset, but it actually means unset or empty (I
verified this by looking at what glibc and musl do).

A recent bug in systemd https://github.com/systemd/systemd/issues/6407
exposed this issue. It causes these these variables to be set to the
empty string in TTY sessions.

Reported by "doodoo" in https://bbs.archlinux.org/viewtopic.php?id=228658

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

Bump version to 0.7.2

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

Update NEWS

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

build/autotools: fix doc installation

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

test/symbols-leak-test.bash: make it easier to read

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

build/autotools: fix symbol versioning configure check

Regressed in 4309735. I thought `$top_srcdir` works in configure too,
apparently not!

See: https://github.com/xkbcommon/libxkbcommon/issues/50#issuecomment-319693694

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

build/autotools: make doxygen's input relative

Was meant to be a part of 41bea9a.

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

build/autotools: dist the meson build files

So that it's possible to build with meson from autotools-generated
tarballs.

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

build/autotools: install documentation under html/

To be consistent with the meson build; also makes more sense as doxygen
can generate more than html (though we currently are not doing that).

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

build: make doxygen run from the source tree

I couldn't find any other way to make this work!

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

Merge pull request #49 from bluetech/meson

build: add meson build system

build: add meson build system

Meson is easier to maintain, much faster, encourages better practices,
and is not built on a pile of shell scripts.

The autotools build system is kept intact for now, in order to ease the
migration. The intention is to remove it sooner rather than later, if
all goes well.

Run `meson build && mesonconf build` to see the configuration options
for the new system. Conversion should be straightforward. Environment
variables like CFLAGS work the same.

If meson is used, xorg-util-macros is not required.

In terms of functionality the two systems have about the same
capabilities. Here are some differences I noticed:

- Meson uses `-g` by default, autotools uses `-g -O2`.

- In autotools the default behavior is to install both static and shared
  versions of the libraries. In meson the user must choose exactly one
  (using -Ddefault_library=static/shared).

  It is possible to workaround if needed (install twice...), but
  hopefully meson will add the option in the future.

- Autotools has builtin ctags/cscope targets, meson doesn't.

  Easy to run the tools directly.

- Meson has builtin benchmarks target. Handy.

- Meson has builtin support for sanitizers/clang-analyzer/lto/pgo/
  coverage etc. Also handy.

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

build: remove Android.mk

I don't think it will work with meson, and I also don't think anyone is
using it.

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

build: remove -uninstalled pkg-config files

Meson doesn't support them directly, and I don't think anyone is using
them.

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

travis: do the build out-of-tree

To catch problems like the previous commit.

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

build: fix out-of-tree build

The change in d44ba48 removed -I$(top_builddir)/src/xkbcomp, but this is
needed in order to find the generated parser.h file which is put in the
build dir.

I also added -I$(top_builddir)/src in order to match the meson behavior.

Fixes https://github.com/xkbcommon/libxkbcommon/issues/50

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

build: use top_srcdir consistently

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

build: change doxygen target to be properly dependency-based

This hackery (thanks libinput) is clearer and more precise than the
previous hackery.

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

build: move custom targets to scripts/ and remove from makefile

These scripts generate source code that is committed to git and hence do
not really belong in the build system. A maintainer runs them as needed.

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

build: rewrite the update-keysyms sed script in python

A bit more manageable this way, and the other part of the target is
already using python.

The output is the same, except I removed the reference to Makefile.am.

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

build: remove unneeded preprocessor include flags

Better to avoid these unexpected include paths.

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

build: explain why we don't AC_DEFINE a couple of constants

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

x11/keymap,test/interactive-evdev: fix a couple of clang-analyzer warnings

From my analysis these values cannot be null, but the analyzer cannot
see this. So assert it.

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

test/x11comp: fix compiler warnings

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

Merge pull request #48 from namandixit/patch-1

Fixed a minor bug in error detection in Wayland test

Fixed a minor bug in error detection in Wayland test

Merge pull request #46 from jwrdegoede/master

Sync Keysyms with recent xproto additions

Sync Keysyms with recent xproto additions

xproto recently has been extended with 4 new keysyms:
XF86XK_Keyboard
XF86XK_WWAN
XF86XK_RFKill
XF86XK_AudioPreset

This commit is the result of running "make update-keysyms" on a system
with the updated xproto installed.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>

doc/compat: correct the XKB protocol version from 1.1 to 1.0

There is no XKB 1.1!

Thanks to Oded Arbel for catching this.

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

test/interactive-x11: handle NULL from xcb_wait_for_event

Can happen in cases like:
- There was an error between the error check and the call.
- The internal poll() fails.

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

state: cure boolean blindness in the filter functions' result

Makes it a little easier to understand the filters.

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

state: remove unneeded NULL check

xkb_filter_new() cannot return NULL.

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

state: reorder new() functions before the set() functions in the code

So that they may be read more naturally in chronological order.

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

test/x11comp: be a bit more careful with kill()

We did it correctly but better be safe and appease clang.

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

test/interactive-wayland: mark a local function static

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

test/interactive-wayland: handle unrecognized SHM format

The enum seems large, and we don't handle all of the values in it.
Previously if we got an unrecognized SHM format we would use an
uninitialized `stride`.

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

test/interactive-wayland: fix uninitialized `ret` in error path

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

Merge pull request #45 from fooishbar/xdg-shell-v6

Wayland: xdg-shell v6 support

interactive-wayland: Port to xdg-shell v6

Mutter only implements v6 now, and Weston also implements that. Port
interactive-wayland to this so people can keep on using it.

Signed-off-by: Daniel Stone <daniels@collabora.com>

Add explicit fallthrough case statements

When we fall through to another label in a case, add an explicit comment
noting so, to quiet GCC 7's warnings.

Signed-off-by: Daniel Stone <daniels@collabora.com>

Bump version to 0.7.1

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

Update NEWS

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

compose/doc: note that it is safe to pass the result of getenv() as locale

See the NOTES section of getenv(3). Somewhat obscure but it doesn't hurt
to reassure the readers who know about this.

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

compose: remove the keysym_from_name cache

The hit rate is high, but either the cache is slow or the function is
not fast enough -- the cache no longer holds its weight, leading only to
very modest improvements. If it's the former, it can definitely be
improved, the code is very dumb (though it worked just as well as any
other I tried back then). But instead, let's just kill it.

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

doc: note that XKB_KEYSYM_CASE_INSENSITIVE does C folding only

and not locale-dependent.

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

keysym: fix locale dependence in xkb_keysym_from_name()

We currently use strcasecmp, which is locale-dependent. In particular,
one well-known surprise even if restricted just ASCII input is found in
the tr_TR (Turkish) locale, see e.g.
https://msdn.microsoft.com/en-us/library/ms973919.aspx#stringsinnet20_topic5

We have known to avoid locale-dependent functions before, but in this
case, we forgot.

Fix it by implementing our own simple ASCII-only strcasecmp/strncasecmp.
Might have been possible to use strcasecmp_l() with the C locale, but
went the easy route.

Side advantage is that even this non-optimized version is faster than
the optimized libc one (__strcasecmp_l_sse42) since it doesn't need to
do the locale stuff. xkb_keysym_from_name(), which uses strcasecmp
heavily, becomes faster, and so for example Compose file parsing, which
uses xkb_keysym_from_name() heavily, becomes ~20% faster.

Resolves https://github.com/xkbcommon/libxkbcommon/issues/42
Signed-off-by: Ran Benita <ran234@gmail.com>

utils: rename popcount to avoid conflict in NetBSD

Resolves https://github.com/xkbcommon/libxkbcommon/issues/41
Signed-off-by: Ran Benita <ran234@gmail.com>

Bump version to 0.7.0

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

Update NEWS

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

Merge pull request #31 from bluetech/consumed-modes

Consumed modifiers modes

state: add GTK consumed modifiers mode

This is more or less what is implemented here:
https://git.gnome.org/browse/gtk+/tree/gdk/x11/gdkkeys-x11.c?h=3.19.10#n1131

The implementation here is more technically correct but should provide
the same results.

Try it out with ./test/interactive-evdev -g (modifiers prefixed with "-"
are consumed).

https://bugzilla.gnome.org/show_bug.cgi?id=754110
https://github.com/xkbcommon/libxkbcommon/issues/17

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

state: allow different modes for calculating consumed modifiers

The current functions dealing with consumed modifiers use the
traditional XKB definition of consumed modifiers (see description in the
added documentation). However, for several users of the library (e.g.
GTK) this definition is unsuitable or too eager. This is exacerbated by
some less-than-ideal xkeyboard-config type definitions (CTRL+ALT seems
to cause most grief...).

So, because we
- want to enable alternative interpretations, but
- don't want to expose too much internal details, and
- want to keep things simple for all library users,
we add a high-level "mode" parameter which selects the desired
interpretation. New ones can be added as long as they make some sense.

All of the old consumed-modifiers functions keep using the traditional
("XKB") mode. I mark xkb_state_mod_mask_remove_consumed() and as
deprecated without adding a *2 variant because I don't it is very useful
(or used) in practice.

Alternative modes are added in subsequent commits (this commit only adds
a mode for the existing behavior).

https://github.com/xkbcommon/libxkbcommon/issues/17

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

utils: add popcount function

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

test/state: move wrongly-placed assert

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

bench: fix compilation on hurd

Patch by Samuel Thibault.

https://github.com/xkbcommon/libxkbcommon/issues/39

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

bench/compose: tabs -> spaces

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

README: Add basic build directions

In particular, highlight the use of configure flags to control locating
X11 keyboard stuff when building for Wayland.

Of particular note, if the locale root is not specified, then xkbcommon
will look for them under $prefix (i.e. /usr/local/share/X11/locale).
But unless the user has specifically installed them there, it is better
to look in the standard system location, /usr/share/X11/locale.

Otherwise, xkbcommon will error when it can't find them, e.g.:

  xkbcommon: ERROR: ~/.XCompose:4:9: failed to expand %L to the locale Compose file
  xkbcommon: ERROR: ~/.XCompose:4:12: unterminated string literal

Signed-off-by: Bryce Harrington <bryce@osg.samsung.com>
Signed-off-by: Ran Benita <ran234@gmail.com>

README: Bug *reports* are welcome

Signed-off-by: Bryce Harrington <bryce@osg.samsung.com>
Reviewed-by: Yong Bakos <ybakos@humanoriented.com>
Signed-off-by: Ran Benita <ran234@gmail.com>

bench: Check for errors opening Compose file

Otherwise it can segfault e.g. running ./compose inside the bench
directory.

Signed-off-by: Bryce Harrington <bryce@bryceharrington.org>
Signed-off-by: Ran Benita <ran234@gmail.com>

doc/compat: (! MODIFIER) syntax is parsed but ignored

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

doc/compat.md: xkbcomp ignores multiple-keysyms these days

https://cgit.freedesktop.org/xorg/app/xkbcomp/commit/?id=e119cbec7e750ffc4d4bd08b577db2c697035a30

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

doc: Also mention the wayland test client in the quick guide

Signed-off-by: Bryce Harrington <bryce@osg.samsung.com>

doc: Declare keymap for wayland example

keymap was defined in the X11 example, but also define it in the wayland
example just to make it a bit more standalone

Signed-off-by: Bryce Harrington <bryce@osg.samsung.com>

doc: Fix ctx type in example

xkb_context_new() returns a xkb_context pointer, so change the variable
definition to be consistent.

Signed-off-by: Bryce Harrington <bryce@osg.samsung.com>

PACKAGING: Update for interactive-wayland

Signed-off-by: Daniel Stone <daniels@collabora.com>

test/interactive-wayland: fix control reaches end of non-void function

AFAICS there is nothing that can fail directly in this function, so
change it to void.

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

keymap-dump: use consistent order set/latch/lock (style)

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

src/state: match_mod_masks can return bool instead of int

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

test/symbols-leak-test: use more portable shebang

Some BSDs don't want to give bash the honor of /bin and put it
elsewhere. So look it up in PATH instead.

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

test/interactive-wayland: don't ignore asprintf return value

Fixes warn_unused_result warning.

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

test: use termios instead of system() for disabling terminal echo

Takes care of GCC's annoyingly persistent warn_unused_result warnings.
But it's better to avoid system() I suppose.

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

gitignore: ignore generated wayland protocol stubs

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

interactive-wayland: Valgrind-proofing

More meticulously free everything we create, including hooking up the
buffer-release callback so we actually free those when required. Make
sure seats are actually in the display's seat list.

The xkbcommon object-unref functions don't actually require
NULL-checking, so we can elide those.

Signed-off-by: Daniel Stone <daniels@collabora.com>

test: Add interactive-wayland

interactive-wayland is very similar to x11/xev, and dumps out as much
state as possible.

It provides no titlebar and a completely random cursor, but such is
life.

Signed-off-by: Daniel Stone <daniels@collabora.com>

Bump version to 0.6.1

Our most minor release yet.

Signed-off-by: Daniel Stone <daniels@collabora.com>

Add LICENSE to distributed files

Signed-off-by: Daniel Stone <daniels@collabora.com>

doc: update reference to compat symbol to its new name

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

Bump version to 0.6.0

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

test: assert/ignore some warn_unused_result's

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

Merge pull request #33 from bluetech/travis-enable-x11

travis: remove --disable-x11

travis: remove --disable-x11

Looks like the CI machines can use Ubuntu 14.04, which has libxcb 1.10,
which is new enough. Lets see if it works.

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

Merge pull request #32 from fooishbar/master

doc: Fix Doxygen include for srcdir != builddir

travis: Enable documentation build

Make sure Doxygen is installed before we build.

Signed-off-by: Daniel Stone <daniels@collabora.com>

doc: Fix Doxygen include for srcdir != builddir

Instead of giving Doxygen a series of absolute paths to the source
files and a relative path to the output directory, run it from the
source directory with purely relative paths to the source files, and
give it an absolute path to the build directory.

This fixes the parsing of README.md with a separate build directory,
since the relative includes for doc/quick-guide.md and doc/compat.md
don't resolve otherwise. Doxygen's INCLUDE_PATH turns out not to fix
this either, since that's just a set of paths to open and parse, rather
than an analogue to cpp's -I.

Signed-off-by: Daniel Stone <daniels@collabora.com>

Update NEWS

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

NEWS: add some spacing

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

build: don't run test/x11comp in make check for now

For some reason, the dumped keymap started repeating the two key types
FOUR_LEVEL_PLUS_LOCK and FOUR_LEVEL_KEYPAD. I need to investigate, but
let's disable it for now until I do (the problem would likely not be in
the xkbcommon - at most in the test itself).

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

test/x11comp: fix memory leak

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

test: add a test that all symbol version file is updated

It is easy to forget to update these files when adding new symbols.

Stolen with slight changes from libinput (commit by Marek Chalupa):
https://cgit.freedesktop.org/wayland/libinput/commit/?id=a9f216ab47ea2f643f20ed741b741a2b5766eba3

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

Add symbol versions forgotten in 0ce17ef3ea3722c1cfe7af3

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

build: make it easier to add non-built (e.g. script) tests

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