doc: clarify "server state" and "client state" distinction

Add a common page for the concept and link to there from the relevant
functions.

Signed-off-by: Ran Benita <ran@unusedvar.com>

build: require meson >= 0.51, fix meson deprecations

Signed-off-by: Ran Benita <ran@unusedvar.com>

ci: don't run linux on push to non-master branches

It's redundant with the pull request run.

Signed-off-by: Ran Benita <ran@unusedvar.com>

context: add XKB_CONTEXT_NO_SECURE_GETENV flag (#312)

This flag is useful for clients that may have relatively benign capabilities
set, like CAP_SYS_NICE, that also want to use the xkb configuration from the
environment and user configs in XDG_CONFIG_HOME.

Fixes: https://github.com/xkbcommon/libxkbcommon/issues/308
Fixes: https://github.com/xkbcommon/libxkbcommon/issues/129
Signed-off-by: Ran Benita <ran@unusedvar.com>

build: remove -fsanitize-undefined-trap-on-error

Meson complains; it's probably not that important anymore when using the
`-Db_sanitize` options.

meson.build:36: WARNING: Consider using the built-in option for sanitizers instead of using "-fsanitize-undefined-trap-on-error".

Signed-off-by: Ran Benita <ran@unusedvar.com>

CI: bump a few actions to newer versions

Node 12 is deprecated so let's bump the actions to newer versions that
use Node 16. See
https://github.blog/changelog/2022-09-22-github-actions-all-actions-will-begin-running-on-node16-instead-of-node12/

build: reenable test-context in MSVC

Signed-off-by: Ran Benita <ran@unusedvar.com>

test: move mkdir & mkdtemp calls to common place and fix them on MSVC

Signed-off-by: Ran Benita <ran@unusedvar.com>

utils: move some MSVC compat stuff to common place

Signed-off-by: Ran Benita <ran@unusedvar.com>

ci/linux: store the test logs as artifacts on failure

This makes debugging a lot easier than having to reproduce locally.

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

ci/linux: split normal testing from valgrind testing

A test case failure usually also triggers valgrind leaks, sifting through those
to find the actual test failure is painful. So let's separate the tests and run
them separately.

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

interactive-wayland: Fix interface versioning

We need to request the lower version of the interface versions we
support and the server supports, not the higher version.

Using the higher version caused crashes due to unbound callbacks on
GNOME, which supports a higher version of `xdg_wm_base`.

Bump version to 1.4.1

Signed-off-by: Ran Benita <ran@unusedvar.com>

keysym-utf: remove bogus currency sign entries

These ended up being mapping in the 0x100xxxx Unicode range, which are
handled automatically ; these special keysyms don't exist (except for
EuroSign).

    #define XKB_KEY_EcuSign                    0x10020a0  /* U+20A0 EURO-CURRENCY SIGN */
    #define XKB_KEY_ColonSign                  0x10020a1  /* U+20A1 COLON SIGN */
    #define XKB_KEY_CruzeiroSign               0x10020a2  /* U+20A2 CRUZEIRO SIGN */
    #define XKB_KEY_FFrancSign                 0x10020a3  /* U+20A3 FRENCH FRANC SIGN */
    #define XKB_KEY_LiraSign                   0x10020a4  /* U+20A4 LIRA SIGN */
    #define XKB_KEY_MillSign                   0x10020a5  /* U+20A5 MILL SIGN */
    #define XKB_KEY_NairaSign                  0x10020a6  /* U+20A6 NAIRA SIGN */
    #define XKB_KEY_PesetaSign                 0x10020a7  /* U+20A7 PESETA SIGN */
    #define XKB_KEY_RupeeSign                  0x10020a8  /* U+20A8 RUPEE SIGN */
    #define XKB_KEY_WonSign                    0x10020a9  /* U+20A9 WON SIGN */
    #define XKB_KEY_NewSheqelSign              0x10020aa  /* U+20AA NEW SHEQEL SIGN */
    #define XKB_KEY_DongSign                   0x10020ab  /* U+20AB DONG SIGN */
    #define XKB_KEY_EuroSign                      0x20ac  /* U+20AC EURO SIGN */

Signed-off-by: Ran Benita <ran@unusedvar.com>

keysym-utf: remove bogus Korean Won sign entry

This entry maps a non existing special keysym 0x20a9.
The correct mapping for XKB_KEY_Korean_Won (0x0eff) already exists.

Signed-off-by: Ran Benita <ran@unusedvar.com>

Remove bogus euro sign entry from keysymtab

Not sure what it's doing here, but converting "€" to a keysym
doesn't work with this entry. 0x13a4 doesn't appear in
xkbcommon-keysyms.h. 0x20ac is the keysym documented in the
header (and it's the last entry in the table).

It's been in the table since it was introduced in e0524296d2e0
("Add API for getting unicode representation of a keysym").

Co-authored-by: Simon Ser <contact@emersion.fr>

Do not clear sibling entries when override.

lokid and hikid actually stores the sibling to current node, which
should not be cleared when override. This would break the sequence with
a common prefix when override another.

Fix #286

Signed-off-by: Weng Xuetian <wengxt@gmail.com>

ci/linux: bring back `apt update` to fix `apt install` 404s

Signed-off-by: Ran Benita <ran@unusedvar.com>

scanner: prefix functions with `scanner_` to avoid symbol conflicts

Particularly `eof()` in mingw-w64.

Fixes: https://github.com/xkbcommon/libxkbcommon/pull/285
Reported-by: Marko Lindqvist
Signed-off-by: Ran Benita <ran@unusedvar.com>

Improve misleading comments #270

ci: another possible fix for release workflow

Signed-off-by: Ran Benita <ran@unusedvar.com>

Bump version to 1.4.0

Signed-off-by: Ran Benita <ran@unusedvar.com>

test: add test for variant inheritance of ISO 639/3166 codes

See efa99624320f54e6e1ff24e4c4b1fcecdfba43d5

Variants should inherit iso639, iso3166 and brief from parent layout if omitted (#266)

registry: skip over invalid ISO639 or ISO3166 entries

If the XML file is somehow off, don't load entries that are against the spec.

test: show the tool exit status on failure

We already do so for the non-successful cases, let's do this here too so we know
whether it was a signal or a normal exit.

registry: don't call xmlCleanupParser()

From the documentation:
> It does not clean up parser state, it cleans up memory allocated by the library
> itself. It is a cleanup function for the XML library. It tries to reclaim all
> related global memory allocated for the library processing. [...]
> One should call xmlCleanupParser() only when the process has finished using the library.

http://xmlsoft.org/html/libxml-parser.html#xmlCleanupParser

Since we're a library ourselves we cannot know if something else in the same
proces uses the parser, so we must not call this.

Reported-by: M Hickford

CI: always run the linux workflow

This makes it easier for contributors to check if their code runs correctly
without having to file a PR.

The Mac and Windows workflows are a bit more involved, so let's keep those on
pull requests only.

CI: split and rename the workflows to windows/macos/linux

"main" is a bit non-descriptive, let's name them after the platforms we run them
on. Splitting them up allows us to be less selective on how we run the various
workflows, e.g. always running the linux one.

Update README.md

Co-authored-by: Ran Benita <ran@unusedvar.com>

Add instructions to run tests. Add build/ to .gitignore.

build: add enable-tools option

this is mainly useful for multilib, but may be useful for other users as well

Signed-off-by: Alex Xu (Hello71) <alex_y_xu@yahoo.ca>

tools/list: print an empty string for null vendor strings

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

tools/list: enclose the the various field names in quotes

Because otherwise the 'no' layout is treated as disagreement with whatever is to
be disagreed with. Fixed in YAML 1.2 but that's not universally supported.

Fixes #268

man: mention `xkbcli list` output is meant to be in YAML 1.2 format

Signed-off-by: Ran Benita <ran@unusedvar.com>

ci: possible fix for release workflow

Signed-off-by: Ran Benita <ran@unusedvar.com>

Bump version to 1.3.1

Signed-off-by: Ran Benita <ran@unusedvar.com>

x11: try to fix crash in xkb_x11_keymap_new_from_device error handling

In 1b3a1c277a033083fee669e92c8cad862716ebd1 we changed the error
handling in this code to not bail out immediately but only after
everything has been processed, to simplify the code. But I suspect the
code isn't prepared for this and that's what causing the crash reported
in the issue.

Bring back the short-circuit error handling which would hopefully fix
such crashes.

Fixes: https://github.com/xkbcommon/libxkbcommon/issues/252
Signed-off-by: Ran Benita <ran@unusedvar.com>

x11: stylistic fixes

Signed-off-by: Ran Benita <ran@unusedvar.com>

tools/how-to-type: add --keysym for how to type a keysym

Previously, could only check how to type a Unicode codepoint, but
searching for a keysym directly is also occasionally useful.

Signed-off-by: Ran Benita <ran@unusedvar.com>

tools/interactive-x11: use keysym to look for Esc

Don't assume that keycode 9 means Escape. Instead, use the keymap
to check for Esc.

Logic copied from the Wayland version.

Signed-off-by: Simon Ser <contact@emersion.fr>

bench/compose: fix compose file path

Forgotten in e2465c2.

Signed-off-by: Ran Benita <ran@unusedvar.com>

ci: some tweaks

- Reduce permissions
- Update some stuff

Signed-off-by: Ran Benita <ran@unusedvar.com>

ci: also link to tarball in github release

Signed-off-by: Ran Benita <ran@unusedvar.com>

ci: automatically create github release when tag is pushed

Fixes: https://github.com/xkbcommon/libxkbcommon/issues/233
Signed-off-by: Ran Benita <ran@unusedvar.com>

test/tool-option-parsing: don't depend on build prefix

Fixes: https://github.com/xkbcommon/libxkbcommon/issues/244
Signed-off-by: Ran Benita <ran@unusedvar.com>

test/compose: add tests for xkb_compose_table_new_from_locale lookup

Signed-off-by: Ran Benita <ran@unusedvar.com>

test/compose: clean up after a setenv

Signed-off-by: Ran Benita <ran@unusedvar.com>

tests/data: add files needed to fully test compose

Signed-off-by: Ran Benita <ran@unusedvar.com>

ignore: remove no longer relevant gitignore files

These were relevant for the autoconf build but now we're meson only.

Signed-off-by: Ran Benita <ran@unusedvar.com>

xkbcommon: deprecate XK_approxeq and XK_notapproxeq

Sync xorg-proto commit
https://gitlab.freedesktop.org/xorg/proto/xorgproto/-/commit/25f3278b85ec7d1c78bb150eaea52f9c98294ea4

Fixes: #82
Signed-off-by: Ran Benita <ran@unusedvar.com>

scripts/update-keysyms: fix path to the include files after de1b6943d

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

tools: add compose tool for Compose debugging

Not very useful so not exposed in xkbcli.

Signed-off-by: Ran Benita <ran@unusedvar.com>

compose: add mapped locale to Compose-file-not-found error log

Signed-off-by: Ran Benita <ran@unusedvar.com>

Bump version to 1.3.0

Signed-off-by: Ran Benita <ran@unusedvar.com>

atom: rewrite as a hash table

While the previous 1987-style[0] scheme was fun (and I reasonably
optimized it for a fair comparison), this task is more suited to a hash
table. Even a simple implementation beats the old one.

[0] Seems to have first appeared in X11R1, released September 1987.
    See server/dix/atom.c here: https://www.x.org/releases/X11R1/X.V11R1.tar.gz

Signed-off-by: Ran Benita <ran@unusedvar.com>

bench: add atom benchmark

Signed-off-by: Ran Benita <ran@unusedvar.com>

bench/rulescomp: decrease benchmark iterations

No need for more, and easier to convert to milliseconds per iteration.

Signed-off-by: Ran Benita <ran@unusedvar.com>

build: fix missing includes

Signed-off-by: Ran Benita <ran@unusedvar.com>

Move include files to include/ subdirectory

This way we don't specify `include_directorories('.')` which brings in
more than needed.

Signed-off-by: Ran Benita <ran@unusedvar.com>

build: move the subproject variables to a common section at the end

As suggested in:
https://github.com/xkbcommon/libxkbcommon/pull/240#discussion_r620784021

Signed-off-by: Ran Benita <ran@unusedvar.com>

Meson: Allow building as subproject

Specify where to find the headers for libxkbcommon_dep,
libxkbcommon_x11_dep, and libxkbregistry_dep,  which allows other
projects to correctly locate the headers when libxkbcommon is being
built as a Meson subproject.

The dep_libxkbregistry variable is renamed to libxkbregistry_dep,
to follow the usual convention for variables which hold declared
dependencies to be used from subproject builds.

Windows: Pass list of symbols to export to MSVC

Arrange for passing .def files with the lists of symbols to export from
DLLs when building on Windows with MSVC. Without this no symbols were
being exported at all.

The .def files are generated from the .map files at build time using
scripts/map-to-def, which avoids needing to maintain two different sets
of files.

tools: change xkbcli list to output YAML

We have a lot of keyboard layouts and the current output format is virtually
useless at searching for a specific one to debug any issues with either the
layout list or the output from libxkbregistry.

Let's use YAML instead because that can easily be post-processed to extract the
specific layouts wanted, e.g. to get the list of all layouts:
  xkbcli-list | yq -r ".layouts[].layout"
to get the list of all variants of the "us" layout:
  xkbcli-list | yq -r '.layouts[] | select(.layout == "us") | .variant
and the number of option groups:
  xkbcli-list | yq -r '.option_groups[] | length'

Note that the top-level nodes have been de-capitalized, so where it was "Models"
before it is now "models" and the "Options" node is now "option_groups".

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

tools: change the list separator handling

Slightly easier to read than the "bool first" approach.

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

keysym: fix underflow in binary searches

This is hit when passing an empty string and XKB_KEYSYM_CASE_INSENSITIVE
to xkb_keysym_from_name currently if `(lo + hi) / 2` is 0 and `cmp < 0`,
causing mid to underflow and the the array access into name_to_keysym on
the next iteration of the loop to be out of bounds .

We *would* use ssize_t here as it is the appropriate type, but windows
unfortunately does not define it.

test: allow skipping errors in certain conditions

Add a skipError test function that can analyse rc, stdout, stderr to skip a test
even when we expect an error. We use that to skip if we couldn't find a keyboard
in the interactive-evdev test.

Fixes #235

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

test: add extra info to the layout-tester --help output

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

test: print the layout-tester progress bar to stdout by default

tqdm prints to stderr by default but we're using that for failed keymap
compiles (which are the ones that really matter). Plus, whether we're using tqdm
is dependent on isatty(sys.stdout) anyway.

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

test: add an LVO argument to the XKB layout tester

Slightly easier to debug if we can have it only parse one single layout.

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

test: print the compiled keymaps to a given directory

With --keymap-output-dir, the given directory will contain a list of files named
after the layout + variant ('us', 'us(euro)', ...) that contain the keymaps for
each variant + option combination compiled.

This is still a lot, but better to sift through hundreds of keymaps than tens of
thousands.

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

test: rework the output for the xkeyboard-config layout tester

The previous output is largely unusable. The result in the CI test runs is a 6GB
file with every compiled keymap in it and while we can grep for ERROR, it's not
particularly useful.

Let's change this and print out YAML instead - that can be machine-processed.
This patch adds a new parent class that prints itself in YAML format,
the tool invocations are child classes of that class. The result looks like this:

Example output:
- rmlvo: ["evdev", "pc105", "us", "haw", "grp:rwin_switch"]
  cmd: "xkbcli-compile-keymap --verbose --rules evdev --model pc105 --layout us --variant haw --options grp:rwin_switch"
  status: 0
- rmlvo: ["evdev", "pc105", "us", "foo", ""]
  cmd: "xkbcli-compile-keymap --verbose --rules evdev --model pc105 --layout us --variant foo"
  status: 1
  error: "failed to compile keymap"

Special status codes are: 99 for "unrecognized keysym" and 90 for "Cannot open
display" in the setxkbmap case.

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

test: add proper --verbose handling to the xkeyboard-config tester

Instead of defaulting to verbose on/off depending on isatty, make it an
argument instead.

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

interactive-wayland: fallback to ftruncate() if needed

Fallback to ftruncate() if the underlying filesystem does not
support posix_fallocate().

Idea by: Jan Beich <jbeich@FreeBSD.org>, Niclas Zeising <zeising@FreeBSD.org>
Inspired by: Wayland cursor/os-compatibility.c
[ran: small adjustments]
Signed-off-by: Ran Benita <ran@unusedvar.com>

doc: fix user-configuration sample file

Support copy-pasting from the docs to get something functional.

Signed-off-by: Jouke Witteveen <j.witteveen@gmail.com>

scripts: update license note in perfect_hash.py

Ref: https://github.com/ilanschnell/perfect-hash/issues/5

Signed-off-by: Ran Benita <ran@unusedvar.com>

Bump version to 1.2.1

Signed-off-by: Ran Benita <ran@unusedvar.com>

x11: fix xkb_x11_keymap_new_from_device failing when a level name is empty

The numpad:mac option doesn't specify a name for the first level:

    // On Mac keypads, level 1 and 2 are swapped.
    partial xkb_types "mac" {
        type "KEYPAD" {
            modifiers = None;
            map[None] = Level2;
            level_name[Level2] = "Number";
        };
        include "extra(keypad)"
    };

This means the atom for level name is XCB_ATOM_NONE. We tried to get its
name, which fails. This regressed in 40c00b472144d1684d2fb97cafef39.

Instead, translate it to XKB_ATOM_NONE, same as the previous behavior.

Fixes: https://github.com/xkbcommon/libxkbcommon/issues/229
Signed-off-by: Ran Benita <ran@unusedvar.com>

Bump version to 1.2.0

Signed-off-by: Ran Benita <ran@unusedvar.com>

compose: fix max compose nodes check

Signed-off-by: Ran Benita <ran@unusedvar.com>

keysym: speed up the perfect hash function

Make it use a bit operation instead of an expensive modulo.

perf diff:

    Baseline  Delta Abs  Shared Object      Symbol
    ........  .........  .................  ...................................

      28.15%     -6.57%  bench-compose      [.] xkb_keysym_from_name

Signed-off-by: Ran Benita <ran@unusedvar.com>

keysym: avoid strtoul in xkb_keysym_from_name

Signed-off-by: Ran Benita <ran@unusedvar.com>

keysym: use a perfect hash function for case sensitive xkb_keysym_from_name

In 7d84809fdccbb5898d0838849ec7c321410182d5 I added a fast path for the
case-sensitive case, but it is still slowing down Compose parsing.

Instead of the binary search, use a perfect hash function, computed with
a simple python module I found (vendored).

It is faster -- perf diff is:

   Baseline  Delta Abs  Shared Object      Symbol
   ........  .........  .................  ...................................

     22.35%    -14.04%  libc-2.33.so       [.] __strcmp_avx2
     16.75%    +10.28%  bench-compose      [.] xkb_keysym_from_name
     20.72%     +2.40%  bench-compose      [.] parse.constprop.0
      2.29%     -1.97%  bench-compose      [.] strcmp@plt
      2.56%     +1.81%  bench-compose      [.] resolve_name
      2.37%     +0.92%  libc-2.33.so       [.] __GI_____strtoull_l_internal
     26.19%     -0.63%  bench-compose      [.] lex
      1.45%     +0.56%  libc-2.33.so       [.] __memchr_avx2
      1.13%     -0.31%  libc-2.33.so       [.] __strcpy_avx2

Also reduces the binary size:

Before:

      text    data     bss     dec     hex filename
    341111    5064       8  346183   54847 build/libxkbcommon.so.0.0.0

After:

      text    data     bss     dec     hex filename
    330215    5064       8  335287   51db7 build/libxkbcommon.so.0.0.0

Note however that it's still larger than before 7d84809fdccbb5898d08388:

      text    data     bss     dec     hex filename
    320617    5168       8  325793   4f8a1 build/libxkbcommon.so.0.0.0

Signed-off-by: Ran Benita <ran@unusedvar.com>

compose: use a ternary tree instead of a regular trie

Previously we used a simple trie with a linked list for each chain.
Unfortunately most compose files have very long chains which means the
constructions performs an almost quadratic number of comparisons.

Switch to using a ternary search tree instead. This is very similar to a
trie, only the linked list is essentially replaced with a binary tree.

On the en_US/Compose file, the perf diff is the following (the modified
function is `parse`):

  Event 'cycles:u'

  Baseline  Delta Abs  Shared Object     Symbol
  ........  .........  ................  .................................

    39.91%    -17.62%  bench-compose     [.] parse.constprop.0
    20.54%     +6.47%  bench-compose     [.] lex
    17.28%     +5.55%  libc-2.33.so      [.] __strcmp_avx2
    12.78%     +4.01%  bench-compose     [.] xkb_keysym_from_name
     2.30%     +0.83%  libc-2.33.so      [.] __GI_____strtoull_l_internal
     3.36%     +0.78%  bench-compose     [.] strcmp@plt

Thanks to some careful packing, the memory usage is pretty much the
same.

Signed-off-by: Ran Benita <ran@unusedvar.com>

ast: remove comment re. anonymous struct

C11 is not sufficient for this, needs `--ms-extensions` which we don't
want to enable.

Signed-off-by: Ran Benita <ran@unusedvar.com>

compose: use anonymous union

Signed-off-by: Ran Benita <ran@unusedvar.com>

build: require C11

I'd really like to use anonymous unions/structs. Supposedly even MSVC
supports it now. Let's try and see.

Signed-off-by: Ran Benita <ran@unusedvar.com>

keysym: properly handle overflow in 0x keysym names

Relatedly, strtoul allows a lot of unwanted stuff (spaces, +/- sign,
thousand seperators), we really ought not use it. But that's for another
time.

Signed-off-by: Ran Benita <ran@unusedvar.com>

compose: add a limit of 65535 sequences

Fits in uint16_t, which enables some future optimizations. But also a
good idea to have some limit. Not aware of any compose files which come
close.

Signed-off-by: Ran Benita <ran@unusedvar.com>

utils: assert on streq for NULL pointers

We have streq_null for that purpose

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

test: move an assert up to before the strlen() use

../../../test/keysym.c:80:24: warning: Null pointer passed to 1st parameter
expecting 'nonnull' [core.NonNullParamChecker]
  (unsigned) strlen(expected));

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

xkbcomp: remove useless assignment

../../../src/xkbcomp/compat.c:693:16: warning: Although the value stored to
'merge' is used in the enclosing expression, the value is never actually read
from 'merge' [deadcode.DeadStores]
    si.merge = merge = (def->merge == MERGE_DEFAULT ? merge : def->merge);

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

xkbcomp: use memcpy over strncpy to avoid analyzer warnings

The target buffer is 7 bytes long, null-termination is optional (as the comment
already suggests). Coverity is unhappy about this though so let's use memset and
memcpy instead.

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

man: add missing pointer to xkbcli-compile-keymap(1)

Signed-off-by: Ran Benita <ran@unusedvar.com>

keysym: fast path for case sensitive xkb_keysym_from_name

xkb_keysym_from_name() is called a lot in Compose file parsing. The
lower case handling slows things down a lot (particularly given we can't
use the optimized strcasecmp() due to locale issues). So add separate
handling for the non-case-sensitive case which is used by Compose.

To do this we need to add another version of the ks_tables table. This
adds ~20kb to the shared library binary. We can probably do something
better here but I think it's fine.

Signed-off-by: Ran Benita <ran@unusedvar.com>

keysym: inline find_sym function

It's easier when everything is in one place.

Signed-off-by: Ran Benita <ran@unusedvar.com>

keysym: use a more descriptive argument name

Signed-off-by: Ran Benita <ran@unusedvar.com>

keysym: open-code bsearch

We want to optimize things here which requires messing with the binary
search some.

Signed-off-by: Ran Benita <ran@unusedvar.com>

interactive-evdev: fix missing initialization

Accidentally got lost in 6b65be4.

Signed-off-by: Ran Benita <ran@unusedvar.com>

interactive-evdev: switch from epoll(2) to poll(2)

Turns out FreeBSD supports evdev, so this toll can work on it; however
it does not support epoll, so switch to poll, which is portable.

Reported-by: Evgeniy Khramtsov <evgeniy@khramtsov.org>
Signed-off-by: Ran Benita <ran@unusedvar.com>