Introduce xkb_keysym_t type

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

Remove atom functions from public API

They're no longer needed since we don't expose any atoms in the
published API anymore.  As a result, we don't need to support external
atom implementations either.  Result!

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

Stringify public name types

Ensure that all names under xkb_desc are strings, rather than atoms.

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

Ensure we always have a complete keymap

Refuse to compile without keycodes + compat + types + symbols.

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

filecomp: Fix xkbcomp warning

Ironically, the test for named.xkb included a call that relied on the
default keymap, without the file naming an explicit default.  Go figure.

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

Convert CompileKeymap to early-return style

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

Use compile_keymap in xkb_compile_keymap_from_components

We already had this exact function sitting right next to us, so use it
rather than open-coding.

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

Add explicit braces

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

xkbcomp: Turn an array into an anonymous struct

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

Remove geometry support, again

It still parses geometry, but happily throws it away.

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

Remove KcCGST names from the map

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

Merge remote-tracking branch 'ran/fixes-cont'

Conflicts:
src/xkbcomp/expr.c

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

Merge remote-tracking branch 'ran/fixes'

Fix "Could not resolve keysym" errors

On many layouts, the following error appears:
Internal error:   Could not resolve keysym 10005b0
(Which is like the trademark of libxkbcommon now, and makes
unicode-heavy symbol files pretty useless).

This occurs when a keysym string (in this case, 10005b0) is passed to
xkb_string_to_keysym, but cannot be resolved.
This in turn happens because the parser passes on hexadecimal keysym
strings without the leading "0x", thus leaving the resolving function
without a way to disambiguate it as a number.

Therefore, make sure to pass on the "0x". The file symbols.c in xkbcomp
project does the same; it probably got lost in translation.

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

Refactor XkbFindFileInPath

Also fixes a bug, where the check (typeLen < 1) should have been
(pathLen < 1).

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

Free all atoms along with keymap

The code to do this is taken from xserver, dix/atom.c.

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

Free IncludePath when no longer needed

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

Free scanFile when no longer needed

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

Fix other misc leaks

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

Don't leak the scanner's buffer

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

Don't leak DoodadInfo's from GeometryInfo

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

Don't leak ActionInfo's

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

Don't leak KeyNamesInfo

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

Don't leak the "minimum"/"maximum" string

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

Don't leak the various *Info's names when overriding them

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

Free XkbFile's when no longer needed

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

Add function to free XkbFile's

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

Don't cache loaded rules files

This needlessly occupies memory for the lifetime of the library, and
does not make a noticeable difference otherwise.

This rules file won't be loaded more than once in most cases anyway, so
just load it again when it happens.

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

Don't cache parsed files

This needlessly occupies memory for the lifetime of the library, and
does not make a noticeable difference otherwise.

Instead, just parse the same file again when it happens.

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

Remove unneeded freeing mechanisms

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

Make the sections array local to the keymap compiling function

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

Remove unused global type tokens

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

Use global tables for action string handling

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

Constify some more text functions

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

Constify global tables

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

Use limits.h to find min/max short value

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

Remove useless casts

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

Use strchr instead of index

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

Use memset instead of bzero

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

Remove return's at the end of void functions

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

Remove all uses of the register keyword

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

Use strdup instead of strlen + malloc + strcpy

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

Don't check for NULL before free()

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

Use yacc-generated header instead of tokens.h

The yacc implementation can generate all the necessary token
definitions itself; there is no need to maintain a hand written
file for that.

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

makekeys: update to match the rest of libX11 makekeys

This integrates two commits from libX11:

ebd6ef0a4db0ddef0ae17ad14571518ccdeea5ba
   XStringToKeysym: Special case for XF86 keysyms

    Some XFree86 keysyms were in XKeysymDB as XF86_foo, despite really being
    XF86foo.  So, if we get to the bottom of XStringToKeysym and haven't
    found our XF86_foo, try it again as XF86foo.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
00175397480b76d32bf82b0c7c94c91a2a95954e
    makekeys: Scan vendor keysyms as well as core

    Since we can't really live without vendor keysyms, scan them all in to
    generate ks_tables.h, rather than only doing the core ones, and leaving
    the vendor syms to be manually synchronised with XKeysymDB.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Notice that the xkey.sh test is changed to match libX11 behavior, i.e.
XKeysymToString(0x1008FE20) -> "XF86Ungrab" as opposed to "XF86_Ungrab".

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

makekeys: Receive the keysym files as arguments

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

When makekeys fails to find a good hash, print error instead of divide-by-zero

This matches commit 78a894231ae8ec4959007b6d9b8d2a15d6333a1e from
libX11:
Signed-off-by: Alan Coopersmith <alan.coopersmith@sun.com>
Signed-off-by: Ran Benita <ran234@gmail.com>

makekeys: Fix build/target word size mismatch when cross-compiling

This matches commit 24283d40b1e4314c6647dda49d2a159833341a8b from
libX11:

    Since makekeys is built using build environment's compiler and
    runs natively, we have to make sure that the size of the
    Signature type is the same on both the native environment
    and the target, otherwise we get mismatches upon running X,
    and some LSB test failures (xts5).

    Use an unsigned 32-bit integer on all platforms.

Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
Reviewed-by: Daniel Stone <daniel@fooishbar.org>
Signed-off-by: Ran Benita <ran234@gmail.com>

Fix remaining warnings

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

Fix warnings in scanner and parser

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

Fix an incorrect sizeof

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

Fix all constness warnings

These are all trivial/obvious fixes which clear a bunch of warnings.

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

Fix possible null dereferences

Fix all reported null dereferences from clang-analyzer.
There seems to be one false negative (in file indicators.c), but it is
fixed anyway.

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

Remove unneeded assignments and variables

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

makekeys: clear const and shadow warnings

The name 'index' is used by the standard library, so use 'ndx' like some
other files.

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

makekeys: make buf local to main()

All the functions already accept it as a parameter. This clears -Wshadow
warnings.

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

Remove ancient compatibility check

Other headers include malloc just fine.

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

Remove unused includes of "tokens.h"

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

Remove unused debug #defines

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

Perform bounds checking in ExprResolveLevel

Both callers perform the same bounds check, so move it into
ExprResolveLevel itself.

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

Fix typo in ExprResolveKeyCode

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

Perform bounds checking in ExprResolveGroup

Every caller did the exact same check on the group bounds after calling
ExprResolveGroup, so might as well do it inside.

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

Switch expression resolution priv from char to void

Avoids a lot of really lame casts.

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

Remove chaining support from LookupModMask

Turns out we never use it, so.

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

Invert LookupModMask/LookupVModMask order

We never want to solely lookup a virtual modifier without also looking
up core modifiers.  So, rather than chaining the vmod lookup inside the
core modifier lookup, invert the ordering.

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

Remove unused LookupPriv::priv

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

Move some definitions from expr.h to expr.c

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

Remove field reference lookup support

None of the lookup functions anyone ever used supported field
references, so don't pretend we do in the API.

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

Move groupNames mask definition to its user

groupNames was declared in compat.c as a global to anything which
included compat.h (for which groupNames was its sole reason to exist),
but only ever used in indicators.c.

Which is kind of fortunate, given that e314931e removed identical
definitions of groupNames (as integers, not masks) from both action.c
and symbols.c.

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

Turn ExprResolveInteger into a simple wrapper

Move the bulk of ExprResolveInteger into an internal function called
ExprResolveIntegerLookup, and introduce ExprResolveInteger as a simple
wrapper which doesn't take priv/lookup arguments.

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

Open-code a radio group SimpleLookup

Because, joke's on you, it wasn't actually looking up radio groups.
Just checking to see if it was a string that was "none", or an integer.

Lord give me strength.

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

Add ExprResolveGroup helper

Just a dumb wrapper around ExprResolveInteger.

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

Add ExprResolveButton helper

Just a dumb wrapper around ExprResolveInteger.

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

Introduce ExprResolveLevel helper

Which returns an integer representing the level number represented by
the given expression.

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

Introduce ExprResolveRadioGroup helper

Just a tiny special-cased version of ExprResolveInteger.

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

Remove priv arguments from ExprResolveString

They've never been used.

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

Remove priv arguments from ExprResolveKeySym

They've never been used by its one caller.

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

Remove priv arguments from ExprResolveKeyName

They've never been used.

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

Remove priv arguments from ExprResolveBoolean

They've never been used.

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

Remove priv arguments from ExprResolveFloat

They've never been used.

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

Remove priv arguments from ExprResolveModMask

What with them now being unused and all.

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

Introduce ExprResolveVModMask

Which is just a slightly more typesafe wrapper around the chained
ExprResolveModMask everyone was using earlier.

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

Respect explicit minimum/maximum keycodes

Make sure we carry over an explicit minimum/maximum keycode setting,
rather than just using the computed minimum/maximum; this got broken
while changing the keycode range to be unsigned.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Reported-by: Pekka Paalanen <ppaalanen@gmail.com>

Still more memory leak fixes

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

Fix xkbparse.y compilation

Thanks to autotools happily building stale generated sources, I hadn't
actually ever built my xkbparse.y changes.  Fix that so it not only
compiles, but works.  This seems to parse long keycodes correctly,
although I very much would not recommend testing this by declaring
0x1fffffff as your highest keycode.

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

Don't leak SymbolsInfo

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

tests: Free returned XKB map

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

Enable stderr debugging by default

(If debugFlags is set to non-zero.)

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

Raise keycode range to 32 bits

\o/

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

vmod: Pass xkb_desc explicitly to vmod functions

Some error paths don't set info->xkb correctly, so just do like most
utility functions and pass the xkb_desc explicitly.

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

Introduce xkb_keycode_t for keycodes

And use it consistently everywhere, including with a special long-safe
internal keycode type, to ease the transition to large keycodes.

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

xkbscan: Copy, rather than assign, file name

For some reason, lex decided to reduce a strcpy into an assignment,
leading to entirely justified valgrind warnings about invalid reads,
when scanFile was set to a string which may have only ever lived on the
stack of a now-exited function.

Make it a strdup() instead.

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

Move include path from X11/extensions/ to xkbcommon/

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

xkbcomp: Silence warning of return of uninitialised value

Some routes through HandleGeometryVar do not set a return value. Set a default
value for the return variable to avoid returning an uninitialised value.

Add xkbc_free_keymap helper

Which just calls XkbcFreeKeyboard with the only arguments you'd ever
pass it.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Signed-off-by: Daniel Stone <daniel@fooishbar.org>

xkbcomp: Copy the string into the buffer rather than updating pointer

Avoids assigning the global pointer to a value that may only have a stack
lifetime:

Fixes valgrind warnings such as:

==24795== Invalid read of size 1
==24795==    at 0x4A06E9A: strcpy (mc_replace_strmem.c:311)
==24795==    by 0x4E54D68: ProcessIncludeFile (misc.c:73)
==24795==    by 0x4E59726: HandleIncludeSymbols.constprop.3 (symbols.c:829)
==24795==    by 0x4E59D8E: HandleSymbolsFile (symbols.c:1673)
==24795==    by 0x4E5A068: CompileSymbols (symbols.c:2211)
==24795==    by 0x4E51A61: CompileKeymap (keymap.c:155)
==24795==    by 0x4E5B410: xkb_compile_keymap_from_components (xkbcomp.c:236)
==24795==    by 0x4E5B587: xkb_compile_keymap_from_rules (xkbcomp.c:161)
==24795==    by 0x405ED2: display_create (window.c:2007)
==24795==    by 0x403732: main (desktop-shell.c:320)
==24795==  Address 0x7fefff0a0 is just below the stack ptr.  To suppress, use:
--workaround-gcc296-bugs=yes
==24795==
==24795== Source and destination overlap in strcpy(0x7fefff430, 0x7fefff430)
==24795==    at 0x4A06F3D: strcpy (mc_replace_strmem.c:311)
==24795==    by 0x4E54D68: ProcessIncludeFile (misc.c:73)
==24795==    by 0x4E59726: HandleIncludeSymbols.constprop.3 (symbols.c:829)
==24795==    by 0x4E59D8E: HandleSymbolsFile (symbols.c:1673)
==24795==    by 0x4E5A068: CompileSymbols (symbols.c:2211)
==24795==    by 0x4E51A61: CompileKeymap (keymap.c:155)
==24795==    by 0x4E5B410: xkb_compile_keymap_from_components (xkbcomp.c:236)
==24795==    by 0x4E5B587: xkb_compile_keymap_from_rules (xkbcomp.c:161)
==24795==    by 0x405ED2: display_create (window.c:2007)
==24795==    by 0x403732: main (desktop-shell.c:320)

Add prototype for CheckDefaultMap().

Those warnings disappear accordingly:
|   CC     parseutils.lo
| parseutils.c:742: warning: no previous prototype for ‘CheckDefaultMap’
|   CC     xkbscan.lo
| xkbscan.l: In function ‘XKBParseString’:
| xkbscan.l:220: warning: implicit declaration of function ‘CheckDefaultMap’
| xkbscan.l:220: warning: nested extern declaration of ‘CheckDefaultMap’

Reviewed-by: Dirk Wallenstein <halsmit@t-online.de>
Signed-off-by: Cyril Brulebois <kibi@debian.org>

Also update required kbproto version in .pc.in file

configure: Lower kbproto requirement to 1.0.4

There's nothing in 1.0.5 that we need.

Revert "config: remove incorrectly implemented visibility compiler option"

This reverts commit bf9fdceef63507da889a22eb1a583d2b6376f5ea.

We really only want to expose symbols that we explicitly mark as part of
the API.  This may not work with other platforms or compilers, but the
fact that private symbols are not available on Linux+GCC is enough of an
incentive to not use those.

Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>