platform/upstream/libxkbcommon.git
10 years agoast: pack the ParseCommon struct
Ran Benita [Sat, 14 Dec 2013 15:39:11 +0000 (17:39 +0200)]
ast: pack the ParseCommon struct

This shows a measurable improvement in memory and performance for free,
on 64bit at least. Packing is (or should be) safe in this case.

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoatom: don't malloc every node separately
Ran Benita [Mon, 2 Dec 2013 15:13:50 +0000 (17:13 +0200)]
atom: don't malloc every node separately

Instead of having a darray of pointers to malloc'ed atom_node's, make it
a darray of atom_node's directly.

This makes the code a bit simpler, saves on some malloc's, and the
memory gain/loss even out.

Unfortunately, we are no longer Three Star Programmers ;(
http://c2.com/cgi/wiki?ThreeStarProgrammer

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agotest/atom: add test for atom table
Ran Benita [Mon, 2 Dec 2013 12:16:45 +0000 (14:16 +0200)]
test/atom: add test for atom table

Mostly a random test.

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoatom: tiny style fixes
Ran Benita [Mon, 2 Dec 2013 12:25:51 +0000 (14:25 +0200)]
atom: tiny style fixes

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoatom: allow passing NULLs to find_node_pointer()
Ran Benita [Mon, 2 Dec 2013 12:23:59 +0000 (14:23 +0200)]
atom: allow passing NULLs to find_node_pointer()

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoatom: allow interning empty string
Ran Benita [Mon, 2 Dec 2013 12:15:58 +0000 (14:15 +0200)]
atom: allow interning empty string

Which is different than XKB_ATOM_NONE, as in "" != NULL.

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoatom: drop {xkb_,}atom_strdup
Ran Benita [Mon, 2 Dec 2013 12:14:41 +0000 (14:14 +0200)]
atom: drop {xkb_,}atom_strdup

Even though in 112cccb18ad1bc877b3c4a87fa536ea085c761b5 I said it might
be useful, it's not. So remove it.

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoexpr: make Expr creation naming and file location consistent
Ran Benita [Sun, 1 Dec 2013 11:32:51 +0000 (13:32 +0200)]
expr: make Expr creation naming and file location consistent

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoexpr: add wrapper macro for ExprCreate
Ran Benita [Sun, 1 Dec 2013 11:29:30 +0000 (13:29 +0200)]
expr: add wrapper macro for ExprCreate

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoexpr: split expression types and allocate them separately
Ran Benita [Sun, 1 Dec 2013 10:08:47 +0000 (12:08 +0200)]
expr: split expression types and allocate them separately

Currently, we have one ExprDef type, which contains a tagged union with
the value of all expression types. Turns out, this union is quite
wasteful memory-wise. Instead, create separate types for all expressions
(e.g ExprBinary, ExprInteger) which embed the common fields
(ExprCommon), and malloc them per their size; ExprDef then becomes a
union of all these types, but is just used as a generic pointer.

[Instead of making ExprDef a union, another option is to use
ExprCommon as the generic pointer type and then do up-castings, like we
do with ParseCommon. But this makes the code much uglier.]

The diff is mostly straightforward mechanical adaptations. It could have
been much smaller with the help of C11 anonymous structs (which were
previously a gnu extension). This will have saved all of the 'op' ->
'expr->op', etc changes. But if we can be a bit more portable for a
little effort, we should.

Before (./test/rulescomp, x86 32 bit, -O2):
==12974==   total heap usage: 145,217 allocs, 145,217 frees, 10,476,238 bytes allocated

After:
==11145==   total heap usage: 145,217 allocs, 145,217 frees, 8,270,358 bytes allocated

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoparser, symbols: drop unnecessary casts
Ran Benita [Sun, 1 Dec 2013 08:45:52 +0000 (10:45 +0200)]
parser, symbols: drop unnecessary casts

It's casted into ExprDef and then uncasted for no reason.

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoexpr: use ExprCreate in more places
Ran Benita [Sun, 1 Dec 2013 08:31:27 +0000 (10:31 +0200)]
expr: use ExprCreate in more places

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoparser: remove 'uval' yylval type
Ran Benita [Sat, 30 Nov 2013 21:29:58 +0000 (23:29 +0200)]
parser: remove 'uval' yylval type

We don't care about DoodadType.

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoexpr: add constructor for boolean expressions
Ran Benita [Sat, 30 Nov 2013 21:24:18 +0000 (23:24 +0200)]
expr: add constructor for boolean expressions

Also add a 'bool set' to the ExprDef union, instead of using 'ival' as a
bool.

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoexpr: add constructors for more expression types
Ran Benita [Sat, 30 Nov 2013 21:12:45 +0000 (23:12 +0200)]
expr: add constructors for more expression types

This makes the parser a bit more declarative. But really it might make
error handling easier.

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoexpr: add 'ident' value to ExprDef union
Ran Benita [Sat, 30 Nov 2013 20:25:39 +0000 (22:25 +0200)]
expr: add 'ident' value to ExprDef union

This distinguishes between an identifier expression and a string
expression in the union.

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoResolve keysyms early in parser
Ran Benita [Wed, 27 Nov 2013 11:49:13 +0000 (13:49 +0200)]
Resolve keysyms early in parser

Instead of having the parser passing strings to the AST, and
symbols/compat etc. resolving them themselves. This simplifies the code
a bit, and makes it possible to print where exactly in the file the bad
keysym originates from.

The previous lazy approach had an advantage of not needlessly resolving
keysyms from unrelated maps. However, I think reporting these errors in
*any* map is better, and the parser is also a bit smarter then old
xkbcomp and doesn't parse many useless maps. So there's no discernible
speed/memory difference with this change.

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoscanner: restore lost DIVIDE token
Ran Benita [Wed, 27 Nov 2013 11:43:57 +0000 (13:43 +0200)]
scanner: restore lost DIVIDE token

I don't know how this could have happened. Luckily this token is
completely useless.

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agosymbols: modernize LookupKeysym
Ran Benita [Wed, 27 Nov 2013 11:12:19 +0000 (13:12 +0200)]
symbols: modernize LookupKeysym

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agosymbols: rename info.modMaps -> modmaps
Ran Benita [Wed, 27 Nov 2013 10:52:20 +0000 (12:52 +0200)]
symbols: rename info.modMaps -> modmaps

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoNEWS: add missing 'New API' entries for NO_FLAGS
Ran Benita [Tue, 26 Nov 2013 10:10:05 +0000 (12:10 +0200)]
NEWS: add missing 'New API' entries for NO_FLAGS

Forgot to add this.

Signed-off-by: Ran Benita <ran234@gmail.com>
10 years agoBump version to 0.3.2 xkbcommon-0.3.2
Ran Benita [Fri, 22 Nov 2013 19:02:03 +0000 (21:02 +0200)]
Bump version to 0.3.2

Add a NEWS file, with some retroactive entries. Also add 'check-news' to
configure.ac, though this might be a bit annoying.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoMake XkbFileCreate argument types match between header & implementation
Alan Coopersmith [Sat, 9 Nov 2013 01:08:35 +0000 (17:08 -0800)]
Make XkbFileCreate argument types match between header & implementation

Fixes build failure with Solaris Studio compilers:
"src/xkbcomp/ast-build.c", line 492: identifier redeclared: XkbFileCreate
  current : function(..., enum xkb_map_flags)
  previous: function(..., unsigned int) : "src/xkbcomp/ast-build.h", line 98

Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
11 years agoscanner-utils: remove outdated comment
Ran Benita [Sun, 27 Oct 2013 18:12:05 +0000 (20:12 +0200)]
scanner-utils: remove outdated comment

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agorules: fix scanning of line-continuation without leading space
Ran Benita [Mon, 14 Oct 2013 16:05:24 +0000 (19:05 +0300)]
rules: fix scanning of line-continuation without leading space

We were failing to scan something like\
this correctly.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoReplace ctype.h functions with ascii ones
Ran Benita [Mon, 14 Oct 2013 15:59:53 +0000 (18:59 +0300)]
Replace ctype.h functions with ascii ones

ctype.h is locale-dependent, so using it in our scanners is not optimal.
Let's be deterministic with our own simple functions.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoconfigure.ac: add AM_PROG_CC_C_O
Ran Benita [Sun, 13 Oct 2013 19:19:38 +0000 (22:19 +0300)]
configure.ac: add AM_PROG_CC_C_O

Older Automakes give this error without this directive:

Makefile.am: C objects in subdir but `AM_PROG_CC_C_O' not in `configure.ac'

In newer autotools this is included under AC_PROG_CC, but it's harmless
to add.

https://github.com/xkbcommon/libxkbcommon/issues/3

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agorulescomp: increase BENCHMARK_ITERATIONS from 1000 to 2500
Ran Benita [Wed, 9 Oct 2013 09:23:46 +0000 (12:23 +0300)]
rulescomp: increase BENCHMARK_ITERATIONS from 1000 to 2500

1000 is a bit too low for statistical significance on this 6 years old
CPU. Since the benchmark is run manually this shouldn't be a problem.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agokeymap: call strlen on keymap string instead of SIZE_MAX
Ran Benita [Wed, 9 Oct 2013 09:12:59 +0000 (12:12 +0300)]
keymap: call strlen on keymap string instead of SIZE_MAX

I wanted to avoid the strlen, but we'd better keep the scanner a bit
less surprising and encourage people to use xkb_keymap_new_from_buffer()
instead of they do in fact have access to the size.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agorules: simplify a bit of code
Ran Benita [Tue, 8 Oct 2013 20:09:01 +0000 (23:09 +0300)]
rules: simplify a bit of code

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agorules: use strlen_safe
Ran Benita [Tue, 8 Oct 2013 19:58:28 +0000 (22:58 +0300)]
rules: use strlen_safe

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agorules: improve error logging macros
Ran Benita [Tue, 8 Oct 2013 19:37:53 +0000 (22:37 +0300)]
rules: improve error logging macros

Improve safety with parenthesis, make the matcher macros use the scanner
ones, and make the 1 variant use %s instead of embedding the msg; this
way the compiler can reuse the string in the binary.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agorules: reduce variable scopes
Ran Benita [Tue, 8 Oct 2013 18:46:01 +0000 (21:46 +0300)]
rules: reduce variable scopes

There are some big functions there, and this might help reduce the
cognitive load a bit.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoscanner: remove unnecessary cast
Ran Benita [Tue, 8 Oct 2013 15:35:05 +0000 (18:35 +0300)]
scanner: remove unnecessary cast

'tok' is already an int now.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agokeysym-utf: Fix a warning about shadowing
Siddharth Heroor [Mon, 7 Oct 2013 08:41:36 +0000 (14:11 +0530)]
keysym-utf: Fix a warning about shadowing

Change variable names to avoid the name clash. The warning seen is

src/keysym-utf.c: In function 'bin_search':
src/keysym-utf.c:841: warning: declaration of 'min' shadows a global declaration
src/utils.h:109: warning: shadowed declaration is here
src/keysym-utf.c:842: warning: declaration of 'max' shadows a global declaration
src/utils.h:115: warning: shadowed declaration is here

Signed-off-by: Siddharth Heroor <heroor@ti.com>
11 years agosymbols: fix use of uninitialized variable
Ran Benita [Thu, 3 Oct 2013 07:02:49 +0000 (10:02 +0300)]
symbols: fix use of uninitialized variable

'tmp' is stack allocated so tmp->merge is used uninitialized by
AddModMapEntry(). The value doesn't matter much, but it used to
make some modmap merging decision (which doesn't have many
conflicts usually).

Bug inherited from xkbcomp.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoparser: add some notes about byacc working
Ran Benita [Sat, 28 Sep 2013 22:41:52 +0000 (01:41 +0300)]
parser: add some notes about byacc working

We now also work with byacc (version tested: 20130925) which some people
prefer, perhaps due to its license (public domain) or performance
(haven't compared).

When using byacc, currently the following warning comes up:
    src/xkbcomp/parser.c:954:14: warning: declaration shadows a variable in the global scope [-Wshadow]
        YYSTYPE  yylval;
                ^
    src/xkbcomp/parser.c:37:20: note: expanded from macro 'yylval'
    #define yylval     _xkbcommon_lval
                    ^
    ./src/xkbcomp/parser.h:96:16: note: previous declaration is here
    extern YYSTYPE _xkbcommon_lval;

This is due to a bug in byacc - it shouldn't output that extern line in
%pure-parser mode. So the warning stays.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoparser: add a workaround for byacc
Ran Benita [Sat, 28 Sep 2013 22:29:47 +0000 (01:29 +0300)]
parser: add a workaround for byacc

Unlike bison, byacc outputs its own parser code *after* our own parser.y
code, which includes the #undef. So this fix is needed for the 'scanner'
-> 'param->scanner' translation to work in the parser.c code generated
by byacc.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoparser: don't use %locations
Ran Benita [Sat, 28 Sep 2013 21:41:17 +0000 (00:41 +0300)]
parser: don't use %locations

byacc doesn't support this feature.

We print the line/col of the last scanned token instead. This is slightly
less in case of *parser* errors (not syntax errors), but I couldn't make
it point to another line, and this are pretty cryptic anyways. So it's
good enough. Also might be a bit faster, but haven't checked.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoparser: drop %name-prefix, use -p yacc argument instead
Ran Benita [Sat, 28 Sep 2013 21:24:50 +0000 (00:24 +0300)]
parser: drop %name-prefix, use -p yacc argument instead

Even though the %name-prefix is more sensible, byacc doesn't support it,
but both bison and byacc support the -p argument.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoparser: use %pure-parser instead of %define api.pure
Ran Benita [Sat, 28 Sep 2013 21:22:20 +0000 (00:22 +0300)]
parser: use %pure-parser instead of %define api.pure

Both bison and byacc support this syntax. Bison manpage says something
about this giving more or less options, but we don't care.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoparser: don't use enum yytokentype
Ran Benita [Sat, 28 Sep 2013 21:19:32 +0000 (00:19 +0300)]
parser: don't use enum yytokentype

byacc doesn't support this, it just puts out #define's for the tokens.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoMakefile.am: drop AM_LFLAGS
Ran Benita [Sat, 28 Sep 2013 21:16:39 +0000 (00:16 +0300)]
Makefile.am: drop AM_LFLAGS

We don't use a lex/flex anymore so this is not used.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoMake C++ happy.
Wander Lairson Costa [Thu, 26 Sep 2013 12:35:33 +0000 (09:35 -0300)]
Make C++ happy.

For most functions taking an enum flags parameter, we use 0 value to
indicate that no flags should be applied.

C++ has a stronger type system than C and will not implicitly convert
int's to enum's. Thus, we create valid 0 enum values for enum types
where it makes sense.

Signed-off-by: Wander Lairson Costa <wander.lairson@gmail.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agotest/state: fix missing xkb_state_unref
Ran Benita [Wed, 25 Sep 2013 07:05:26 +0000 (10:05 +0300)]
test/state: fix missing xkb_state_unref

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agodoc: use 'indices' instead of 'indexes' consistently
Ran Benita [Thu, 29 Aug 2013 16:49:40 +0000 (19:49 +0300)]
doc: use 'indices' instead of 'indexes' consistently

We use 'indices' in some function names so use that.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agodoc: try to explain leds and shift levels
Ran Benita [Thu, 29 Aug 2013 16:48:35 +0000 (19:48 +0300)]
doc: try to explain leds and shift levels

Remove the @todo's.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agobuild: distribute makekeys.py and keywords.gperf
Ran Benita [Thu, 29 Aug 2013 16:07:25 +0000 (19:07 +0300)]
build: distribute makekeys.py and keywords.gperf

This was an oversight: even though we ship the pre-built files, it is
still good behavior to include the original generators / source files in
the tarball.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agostate: apply capitalization transformation on keysyms
Ran Benita [Tue, 13 Aug 2013 15:57:43 +0000 (18:57 +0300)]
state: apply capitalization transformation on keysyms

The xkbproto spec says:
http://www.x.org/releases/current/doc/kbproto/xkbproto.html#Interpreting_the_Lock_Modifier

    If the Lock modifier is not consumed by the symbol lookup process,
    routines that determine the symbol and string that correspond to
    an event should capitalize the result.

This was not an issue until now, because most xkeyboard-config keymaps
do not utilize this "feature", and specify the keysyms for the Lock
modifier explicitly instead. However, some keymaps do depend on it, e.g.
ch(fr) for eacute and others.

The spec goes on to describe two options for doing this transformation:
locale-sensitive and locale-insensitive. We opt for the latter; it is
less desirable but we don't want *that* headache.

Also, only xkb_state_key_get_one_sym() is changed;
xkb_state_key_get_syms() is left as-is, and always reports the
untransformed keysyms. This is for the following reasons:

- The API doesn't allow it, since we return a const pointer directly to
  the keymap keysyms table and we can't transform that.

- The transformation doesn't make sense for multiple-keysyms.

- It can be useful for an application to get the "raw" keysyms if it
  wants to (e.g. maybe it wants to do the transformation itself).

Finally, note that xkb_state_mod_index_is_consumed() does *not*
report Lock as consumed even if it was used in the transformation. This
is what Xlib does.

This definitely doesn't fall under the "hard to misuse" API rule but
it's the best we can do.

https://bugs.freedesktop.org/show_bug.cgi?id=67167

Reported-By: Gatis Paeglis <gatis.paeglis@digia.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agotest/interactive: use xkb_state_key_get_one_sym() when there's only 1
Ran Benita [Tue, 13 Aug 2013 15:55:09 +0000 (18:55 +0300)]
test/interactive: use xkb_state_key_get_one_sym() when there's only 1

Kind of odd, but get_one_sym() will be getting a different behavior.
Real life users *should* pick one or the other.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agokeysym: add xkb_keysym_to_{lower,upper}
Ran Benita [Tue, 13 Aug 2013 15:52:46 +0000 (18:52 +0300)]
keysym: add xkb_keysym_to_{lower,upper}

These functions are needed later; they are not API functions. The
capitalization is not locale sensitive.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agobuild: use build-aux as autom4te cache directory
Ran Benita [Thu, 15 Aug 2013 06:53:32 +0000 (09:53 +0300)]
build: use build-aux as autom4te cache directory

Remove the generated directory ./autom4te.cache by reusing ./build-aux
as cache directory.

This was stolen from a libxcb commit by Daniel Martin:
http://lists.freedesktop.org/archives/xcb/2013-July/008431.html

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agobuild: use AC_CONFIG_AUX_DIR([build-aux])
Ran Benita [Thu, 15 Aug 2013 06:45:49 +0000 (09:45 +0300)]
build: use AC_CONFIG_AUX_DIR([build-aux])

Less clutter, why not?

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agobuild: build libtest.la only on 'make check'
Ran Benita [Thu, 15 Aug 2013 06:32:42 +0000 (09:32 +0300)]
build: build libtest.la only on 'make check'

For those just running 'make', compiling libtest is wasted heat.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agodoc: tweak the wording on xkb_state/xkb_keymap relationship
Ran Benita [Wed, 14 Aug 2013 08:36:04 +0000 (11:36 +0300)]
doc: tweak the wording on xkb_state/xkb_keymap relationship

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agotest: make sure keycode 0 works fine
Ran Benita [Wed, 14 Aug 2013 08:35:01 +0000 (11:35 +0300)]
test: make sure keycode 0 works fine

It is a legal keycode but we never tried it actually.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agodoc: un-clarify one subtle point
Ran Benita [Tue, 13 Aug 2013 12:17:59 +0000 (15:17 +0300)]
doc: un-clarify one subtle point

Including the X server is a bit of a borderline case; we should mostly
encourage people to use update_mask() only when xkbcommon itself
serializes the state on the other side. But it's not entirely wrong
either.. So rephrase a bit.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoscanner: don't fail over unknown escape sequence
Ran Benita [Tue, 13 Aug 2013 11:45:33 +0000 (14:45 +0300)]
scanner: don't fail over unknown escape sequence

This is too strict, and causes symbols/cz to fail parsing. Instead, just
emit a warning (not shown by default):
    xkbcommon: WARNING: cz:75:19: unknown escape sequence in string literal

https://bugs.freedesktop.org/show_bug.cgi?id=68056

Reported-By: Gatis Paeglis <gatis.paeglis@digia.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoxkbcomp: improve a few log messages
Ran Benita [Tue, 13 Aug 2013 06:57:07 +0000 (09:57 +0300)]
xkbcomp: improve a few log messages

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoscanner: fix compiler warning
Ran Benita [Fri, 2 Aug 2013 11:41:19 +0000 (14:41 +0300)]
scanner: fix compiler warning

src/xkbcomp/scanner.c:158:17: warning: comparison of constant -1 with expression of type 'enum yytokentype' is always true
      [-Wtautological-constant-out-of-range-compare]
        if (tok != -1) return tok;
            ~~~ ^  ~~

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agobuild: make all symbols in libtest visible
Ran Benita [Fri, 2 Aug 2013 08:07:27 +0000 (11:07 +0300)]
build: make all symbols in libtest visible

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agobuild: use AM_LDFLAGS for general flags
Ran Benita [Fri, 2 Aug 2013 07:21:52 +0000 (10:21 +0300)]
build: use AM_LDFLAGS for general flags

We want -no-undefined for every library so do it implictly.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoscanner: allow empty key name literals
Ran Benita [Thu, 1 Aug 2013 20:09:46 +0000 (23:09 +0300)]
scanner: allow empty key name literals

Some keymaps actually have this, like the quartz.xkb which is tested. We
need to support these.

https://bugs.freedesktop.org/show_bug.cgi?id=67654

Reported-By: Gatis Paeglis <gatis.paeglis@digia.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoxkbcomp: handle empty keymaps
Ran Benita [Thu, 1 Aug 2013 17:24:27 +0000 (20:24 +0300)]
xkbcomp: handle empty keymaps

We should handle empty xkb_keycode and xkb_symbol sections, since
xkbcomp handles them, and apparently XQuartz uses it. There are also
files for it in xkeyboard-config (rules=base model=empty layout=empty,
which translate to keycodes/empty and symbols/empty).

https://bugs.freedesktop.org/show_bug.cgi?id=67654

Reported-By: Gatis Paeglis <gatis.paeglis@digia.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agokeymap-dump: use correct format specifiers
Ran Benita [Thu, 1 Aug 2013 17:44:46 +0000 (20:44 +0300)]
keymap-dump: use correct format specifiers

For keycodes, groups, levels, etc, which are unsigned. The really proper
inttypes.h ones seem a bit much though.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agolog: change the log prefixes to be more library-like
Ran Benita [Sun, 28 Jul 2013 20:42:35 +0000 (23:42 +0300)]
log: change the log prefixes to be more library-like

"Error: " is not very informative when intermingled with other logs. The
format that seems most suitable is:
    "xkbcommon: ERROR: %s"

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agokeymap: update builtin fields directly in src/keymap.c
Ran Benita [Sun, 28 Jul 2013 13:21:40 +0000 (16:21 +0300)]
keymap: update builtin fields directly in src/keymap.c

This fields are part of our API and every keymap should have them, not
just xkbcomp/ ones.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agokeymap: be more careful in xkb_keymap_unref
Ran Benita [Sat, 27 Jul 2013 18:19:22 +0000 (21:19 +0300)]
keymap: be more careful in xkb_keymap_unref

To allow calling _unref on the keymap in whatever limbo state it happens
to be in.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agokeymap: remove struct xkb_key_redirect_action
Ran Benita [Thu, 25 Jul 2013 21:50:26 +0000 (00:50 +0300)]
keymap: remove struct xkb_key_redirect_action

The file src/xkbcomp/action.c already doesn't handle this action type
and fails if it encounters it. So lets not pretend to do something with
it, and ignore it rather than failing.

If we/someone wants this we can consider implementing it.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agodoc: try to clarify some subtle points
Ran Benita [Wed, 31 Jul 2013 07:38:30 +0000 (10:38 +0300)]
doc: try to clarify some subtle points

Which to choose: xkb_state_update_key() / xkb_state_update_mask(). If you
use update_mask(), you don't (and shouldn't) use update_key().
If you use update_key(), be careful not to have unmatched UP/DOWN
calls; this can bring about stuck modifiers etc.

Probably want to use XKB_STATE_{MODS,LAYOUT}_EFFECTIVE. The others are
only useful in very special cases.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agotest/keyseq: add de(neo) Level{6,7,8} tests
Ran Benita [Thu, 25 Jul 2013 10:21:33 +0000 (13:21 +0300)]
test/keyseq: add de(neo) Level{6,7,8} tests

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoGet rid of the usage of PATH_MAX
Pino Toscano [Wed, 24 Jul 2013 08:05:02 +0000 (10:05 +0200)]
Get rid of the usage of PATH_MAX

PATH_MAX is optional in POSIX, so avoid its unconditional usage
allocating and freeing buffers as needed.
To avoid too many malloc/free in the for loop in FindFileInXkbPath,
a buffer is grown according to the size needed at each iteration.

11 years agokeymap: move RANGE_WRAP to be the first in the enum
Ran Benita [Tue, 23 Jul 2013 08:36:01 +0000 (11:36 +0300)]
keymap: move RANGE_WRAP to be the first in the enum

This is the reasonable "zero-default" for this enum.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agotest: sync test/data from xkeyboard-config 2.9
Ran Benita [Mon, 22 Jul 2013 15:43:53 +0000 (18:43 +0300)]
test: sync test/data from xkeyboard-config 2.9

Needed for some tests. The tests need some adjustment, mostly because of
the resolution of xkeyboard-config bug
https://bugs.freedesktop.org/show_bug.cgi?id=50935

Also add the 'ch' symbols file for future tests.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoatom: really work with non-NUL-terminated strings
Ran Benita [Mon, 22 Jul 2013 07:51:22 +0000 (10:51 +0300)]
atom: really work with non-NUL-terminated strings

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoctx: adapt to the len-aware atom functions
Ran Benita [Sun, 21 Jul 2013 14:01:20 +0000 (17:01 +0300)]
ctx: adapt to the len-aware atom functions

xkb_atom_intern now takes a len parameter. Turns out though that almost
all of our xkb_atom_intern calls are called on string literals, the
length of which we know statically. So we add a macro to micro-optimize
this case.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoatom: allow interning non-NUL-terminated strings
Ran Benita [Sun, 21 Jul 2013 13:41:27 +0000 (16:41 +0300)]
atom: allow interning non-NUL-terminated strings

We need this later. The strlen was calculated anyway, so no loss here.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoatom: expand variable names
Ran Benita [Sun, 21 Jul 2013 13:32:21 +0000 (16:32 +0300)]
atom: expand variable names

A bit easier to understand at a glance.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agokeymap: don't use darray for sym_interprets
Ran Benita [Sun, 21 Jul 2013 06:48:12 +0000 (09:48 +0300)]
keymap: don't use darray for sym_interprets

We want xkb_keymap to be easy to handle everywhere.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoxkbcomp: escape the section names before storing them in the keymap
Ran Benita [Thu, 18 Jul 2013 11:50:21 +0000 (14:50 +0300)]
xkbcomp: escape the section names before storing them in the keymap

This ensures the names are escaped before having any interaction with
the user.

This was caught by noticing dump(compile(dump())) != dump. Since that's
a nice test we add it to stringcomp.

https://bugs.freedesktop.org/show_bug.cgi?id=67032

Reported-By: Auke Booij
Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoResync keysym database
Daniel Stone [Wed, 17 Jul 2013 21:46:48 +0000 (22:46 +0100)]
Resync keysym database

xproto 7.0.24 adds XF86AudioMicMute.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
11 years agodoc: Update Doxyfile
Ran Benita [Wed, 19 Jun 2013 18:13:52 +0000 (21:13 +0300)]
doc: Update Doxyfile

It gives:

Warning: Tag `SYMBOL_CACHE_SIZE' at line 346 of file doc/Doxyfile has become obsolete.
To avoid this warning please remove this line from your configuration file or upgrade it using "doxygen -u"

So run 'doxygen -u' as suggested.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoBump version to 0.3.1 xkbcommon-0.3.1
Daniel Stone [Mon, 3 Jun 2013 15:43:43 +0000 (16:43 +0100)]
Bump version to 0.3.1

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
11 years agoAdd keycode min/max and iteration API
Matthias Clasen [Thu, 9 May 2013 14:31:21 +0000 (15:31 +0100)]
Add keycode min/max and iteration API

Add three new pieces of API:
  - xkb_keymap_min_keycode does what it says on the tin
  - xkb_keymap_max_keycode likewise
  - xkb_keymap_key_for_each calls the provided function once for every
    valid key in the keymap

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
11 years agoWiden keycode range to 8/255 if possible (bug #63390)
Daniel Stone [Thu, 9 May 2013 13:47:09 +0000 (14:47 +0100)]
Widen keycode range to 8/255 if possible (bug #63390)

If the keycode range is smaller than 8 → 255, artifically widen it when
dumping the keymap as not to displease X.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
11 years agoReplace flex scanner with a hand-written one
Ran Benita [Sun, 12 Aug 2012 08:40:02 +0000 (11:40 +0300)]
Replace flex scanner with a hand-written one

The scanner is very similar in structure to the one in xkbcomp/rules.c.
It avoids copying and has nicer error reporting.

It uses gperf to generate a hashtable for the keywords, which gives a
nice speed boost (compared to the naive strcasecmp method at least). But
since there's hardly a reason to regenerate it every time and require
people to install gperf, the output (keywords.c) is added here as well.

Here are some stats from test/rulescomp:

Before:
compiled 1000 keymaps in 4.052939625s
==22063==   total heap usage: 101,101 allocs, 101,101 frees, 11,840,834 bytes allocated

After:
compiled 1000 keymaps in 3.519665434s
==26505==   total heap usage: 99,945 allocs, 99,945 frees, 7,033,608 bytes allocated

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoutils: add {un,}map_file to read an entire file
Ran Benita [Thu, 14 Mar 2013 12:33:40 +0000 (14:33 +0200)]
utils: add {un,}map_file to read an entire file

This wraps the current mmap call and adds a fallback implementation for
systems which do not have mmap (e.g. mingw).

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoAdd scanner-utils.h for common scanner functions
Ran Benita [Thu, 14 Mar 2013 12:31:55 +0000 (14:31 +0200)]
Add scanner-utils.h for common scanner functions

We want to share the same functions for another scanner.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoBump version to 0.3.0 xkbcommon-0.3.0
Daniel Stone [Mon, 1 Apr 2013 16:50:20 +0000 (17:50 +0100)]
Bump version to 0.3.0

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
11 years agokeymap: add xkb_keymap_new_from_buffer()
David Herrmann [Mon, 11 Mar 2013 11:53:39 +0000 (12:53 +0100)]
keymap: add xkb_keymap_new_from_buffer()

The current API doesn't allow the caller to create keymaps from mmap()'ed
files. The problem is, xkb_keymap_new_from_string() requires a terminating
0 byte. However, there is no way to guarantee that when using mmap() so a
user currently has to copy the whole file just to get the terminating zero
byte (assuming they cannot use xkb_keymap_new_from_file()).

This adds a new entry xkb_keymap_new_from_buffer() which takes a memory
location and the buffer size in bytes.

Internally, we depend on yy_scan_{string,byte}() helpers. According to
flex documentation these already copy the input string because they are
wrappers around yy_scan_buffer().
yy_scan_buffer() on the other hand has some insane requirements. The
buffer must be writeable and the last two bytes must be ASCII-NUL. But the
buffer may contain other 0 bytes just fine.

Because we don't want these constraints in our public API,
xkb_keymap_new_from_buffer() needs to create a copy of the input memory.
But it then calls yy_scan_buffer() directly. Hence, we have the same
number of buffer-copies as with *_from_string() but without the
terminating 0 requirement.
The explicit yy_scan_buffer() call is preferred over yy_scan_byte() so the
buffer-copy operation is not hidden somewhere in flex.

Maybe some day we no longer depend on flex and can have a zero-copy API. A
user could mmap() a file and it would get parsed right from this buffer.
But until then, we shouldn't expose this limitation in the API but instead
provide an API that some day can work with zero-copy.

Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
[ran: rebased on top of my branch]
Conflicts:
Makefile.am
src/xkbcomp/xkbcomp.c

11 years agoxkbcomp/keymap: silence a gcc warning
Ran Benita [Sat, 30 Mar 2013 16:19:01 +0000 (19:19 +0300)]
xkbcomp/keymap: silence a gcc warning

src/xkbcomp/keymap.c:127:12: error: 'found' may be used uninitialized in this function [-Werror=maybe-uninitialized]

Not really, but why not.

Signed-off-by: Ran Benita <ran234@gmail.com>
11 years agoAdd key-sequence checking to rulescomp
Daniel Stone [Wed, 27 Mar 2013 22:40:58 +0000 (22:40 +0000)]
Add key-sequence checking to rulescomp

Make sure we're actually getting the keymaps we're hoping to
compile.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
11 years agotest: Add va_list variant of test_key_seq
Daniel Stone [Tue, 19 Mar 2013 10:59:38 +0000 (10:59 +0000)]
test: Add va_list variant of test_key_seq

For use when chaining tests.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
11 years agoAllow NULL rmlvo for xkb_keymap_new_from_names
Daniel Stone [Sat, 2 Mar 2013 07:47:59 +0000 (23:47 -0800)]
Allow NULL rmlvo for xkb_keymap_new_from_names

Previously we allowed you to pass a names struct with five NULL members,
but not just pass NULL for the struct itself.  This was pretty dumb. :(

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
11 years agotest: Add environment checking to rulescomp
Daniel Stone [Mon, 18 Mar 2013 22:15:20 +0000 (22:15 +0000)]
test: Add environment checking to rulescomp

To ensure that overriding RMLVO from the environment works.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
11 years agotest: Suppress RMLVO environment inheritance by default
Daniel Stone [Tue, 19 Mar 2013 10:29:49 +0000 (10:29 +0000)]
test: Suppress RMLVO environment inheritance by default

But add a flag to allow it for later usage.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
11 years agoAdd environment overrides for default RMLVO
Daniel Stone [Thu, 28 Feb 2013 18:48:40 +0000 (10:48 -0800)]
Add environment overrides for default RMLVO

You can now set default values in the environment, as well as a context
option to ignore the environment, e.g. for tests.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
11 years agotest: Use test_get_context() in log.c
Daniel Stone [Mon, 18 Mar 2013 21:03:00 +0000 (21:03 +0000)]
test: Use test_get_context() in log.c

Since the only behavioural change is overriding default includes.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
11 years agotest: Add flags argument to test_get_context()
Daniel Stone [Mon, 18 Mar 2013 21:02:35 +0000 (21:02 +0000)]
test: Add flags argument to test_get_context()

Allowing overriding of environment suppression, at first.

Signed-off-by: Daniel Stone <daniel@fooishbar.org>
11 years agotest: Move test_key_seq to common.c
Daniel Stone [Mon, 18 Mar 2013 20:55:18 +0000 (20:55 +0000)]
test: Move test_key_seq to common.c

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