- remove extern declaration in source file, use header instead
- add inclusion of header defining the functions of the file to
get the compiler to cross-check them
- marked static the functions that should not be visible ouside
their file
- remove extern declaration in source file, use header instead
- add inclusion of header defining the functions of the file to
get the compiler to cross-check them
- marked static the functions that should not be visible ouside
their file
This allows the compiler to warn if the definition in the file is
no in line with what is exposed to the users of the function through
the header definition.
The declarations have been split by source file, adding the usual
separation mark. Removed these 2 prototypes:
- W_SetFocusOfToplevel: case typo, the correct prototype already existed
in the file;
- W_TextWidth: function is not defined anywhere
When a function is used as a call-back, it is dangerous to have
arguments with a type different than what is expected by the
call-back definition.
This patch sets the argument list to match what is expected by
the call-back prototype and inserts an explicit type conversion
at the beginning of the function.
It is dangerous to let the compiler know about a function without
letting him know the arguments because he won't be able to report
invalid calls.
This patch concern the cases where adding the arguments led to
problems because the functions were used as call-back. As it is
dangerous to have parameter mismatchs in call-back, setup the
args as expected by prototype and added explicit conversion inside
the concerned function, so the compiler will know and be able to
do what may be necessary.
It is dangerous to let the compiler know about a function without
letting him know the arguments because he won't be able to report
invalid calls.
This patch concern the cases where adding the arguments did not
need other code change.
This is the correct way to tell that a function takes no
arguments, because an empty parameter list tells the compiler
that it is not yet defined, and is tolerated only for
compatibility with very old C compilers for whom prototypes
were not yet a defined language element.
The functions are now grouped by source file (groups in alphabetic
order) with the traditional separation mark. This makes the file
easier to work with.
It is not a good idea to multiply the number of header files,
specially in this case where 'double.h' defined so few things
(and lacked the usual copyright notice / include guards).
We probably don't want our users to have to endure them, so they
are enabled only when Debug is activated, because they tend to
help keeping the code safe.
The equality comparison (a == b) is known to be a dangerous trap
when floating-point arithmetics are involved. In the current case
the offending operation can be done with integers directly.
The equality comparison (a == b) is known to be a dangerous trap
when floating-point arithmetics are involved. This patch changes
all the cases which try to do this to a safer check.
The original code allowed to have 0.0, but this can generate
division by zero in WScrollView. As a value of 0.0 is not realistic
anyway, use a minimum constant instead.
The equality comparison (a == b) is known to be a dangerous trap
when floating-point arithmetics are involved. This patch changes
all the cases which try to do this to a safer check.
Some header were creating variable, this is a bad practice which
is likely to not behave as expected. This creates one distinct
variable in each object file that used the header, and:
- on well behaved compiler, this ends up in a link error (see
commit 39fdb451ba for an example)
- on bad behaving compiler, this can be linked as multiple local
variable, thus having strange effects when running program
- on insouciant compiler (who said gcc?) the variables are
silently merged, hiding portability issues
Autoconf provides the necessary stuff to detect if inline keyword
is supported, and to detect special syntaxes, so let's use this
and remove the multiple local definitions, this makes code simpler.
When using the formula [sizeof(array) / sizeof( x )] to get the number
of element in a static array, it is better to use array[0] for 'x'
instead of the base type of array:
- in case the base type would change someday;
- if the compiler were deciding to insert padding somewhere
Some compilation options are actually targetting the preprocessor
instead of the compiler; using the wrong variable can have some
subtile side effects, so let's get things right.
Considering the number of headers we have, it is a good idea to
avoid possible problems. For details, you may read:
http://en.wikipedia.org/wiki/Include_guard
All headers should be ok now.
The automake documentation states that using substitution inside the
list of SOURCES will not work and calls for not doing it. The use
of 'EXTRA_xxx' made things look like they worked but is probably not
enough for corner cases.
This patches switches to the conditional method which will be safe.
To be able to do this in a clean way, it was necessary to add the
attribute also in PLData's ptr field, which is actually right
because none of the function changes its content.
The function that fills it from a file/pipe however needed small
changes to respect the const-ness of the field.
The history is actually loaded from a file into a property list that
is then converted to an array. The intermediate property list was
not freed, which led to memory leak.
It looks like it was a tentative of optimisation to avoid duplicating
an already allocated string and re-use the pointer instead, but this
means it is not possible to free the original container as it would
free the string too.
There is a better way to do this, but it requires an API change on the
WUtil library so it is left for a future improvment.
The previous code limited the number of entries that were read into
the history array, but the user is expecting a maximum on the
number of entries displayed. This can make a little difference in
two cases:
- if there are duplicate entries (dups are checked for and removed)
- if some entries are not strings (unlikely, but not impossible)
The new code just stops adding history entries when the user
specified count is reached.
When menus are read in the PropList format, they are loaded into a
temporary PropList object, which is parsed into the internal menu
structure, and the PropList object is no more used. There were two
cases where this temp object was not freed.
A common argument to all these functions is the name of the key to
operate on, and this name is never modified by the functions. Marking
it as const reflects this, and can allow compiler to generate better
results thanks to this info.
The addition of the const attributes is actually an API change, so
we have to reflect this for the next official release.
Because the change on 'wusergnusteppath' may impact users of the API,
we won't only change REVISION like it was done for WRaster lib.
The first optimisation is to compute only once the path, and then
always re-use the value which did not change anyway.
The second optimisation is to avoid a lot of excessive function
calls, including alloc+free that are not necessary and participate
in memory fragmentation.
According to the way its value is being used everywhere, that is
what would be expected, so let's make it official.
Please note that this may introduce warnings on user code using
this function ("...discard const...") but that's an opportunity
for them to check that their code is not doing anything wrong.
The internal function 'unescapestr' is used to transform strings which
may contain escape sequences (\x) into their plain representation.
There are a few cases where the function can misbehave (typically parse
after the end of string, thus writing past the end of the reserved
result area) which can be a source of problem later. The new code
should be safer.
The previous syntax used an explicit cast to remove the CONST
attribute, but the right way is to keep the attribute and store
the result in a variable which is defined properly.
This makes the WUtil API as much const-correct as possible for
the arguments being given to its functions.
This does not make it totally correct as it does not changes the
const-ness on returned values because the goal of this patch is
to make no visible change to existing program that would use this
library.
Note that the argument is also stored as-is in the PLData structure
but only for debugging purpose (warning display to user), hence the
choice to not duplicate it. As a side effect, it was 'const'-ified
too to reflect that.
As a side note, in 'wfindfileinlist' the first argument should be:
const char * const *path_list
However, due to limited support for const in plain C, that would
introduce warnings in user code. For compatibility issues, this
was not implemented.
A number of functions do not actually modify the strings given as
parameter, but only read or duplicate it. In this case it is a good
practice to mark that parameter as pointer-to-const to let the
compiler known about it, to be able to perform appropriate
optimisations.
Prototype of function changed in commit
d1e1c13fa3
but two of them were not updated in the source.
Thanks to Rodolfo García Peñas for taking the time to test and report.
The addition of 'const' to parameters of library's functions is an
API change, although this will break neither the binary interface
nor the compilation of program using the library; the other changes
also have no impact on the compiled library object.
All these functions expects agruments like color or list-of-points
that should not be modified (and are not) by the function; added
the corresponding qualifier to reflect that.
An enum is always a better idea as it allows the compiler to do
some checks, and as this info is internal only to the WRLib it
will not change the API.
The functions are declared in different files but they were called
in another file which re-declared the prototypes. This is dangerous
as it can lead to misaligned prototypes when functions changes.
They are now grouped in the library internal header 'imgformat.h'
The config-file-keyword was defined in an array, but the user displayed
string was defined at another place, thus requiring special care when
modifying the list, to keep the same number of entries, in the same
order, ...
This was an open call for subtle bugs, so the keyword and user string
are now at a single place, making updates on the list easier and the
GUI generation code smaller.
As proposed by Christian Schulte, when compiling WindowMaker on BSD
it uses the kvm library to gather information on processes (in the
os-dependant file), so we must properly link against the corresponding
library, which this patch adds.
The list of options in the Expert tab of WPrefs.app was filled using
hard-coded list of commands with static indexes and the related
usage were spread accross procedures. This was dangerous because
it could lead to mismatched indexes, the list length being also
variable, so it was hard to update.
This patch proposes to group all the options in a single place (a
static const array), so it is easy to add/remove/reorder the options,
and the code just parses this array based on its auto-calculated
size.
As reported by Amadeusz Sławiński, there were a number of debug-only
information printed by the menu parser, which had no reason to be
in a user package.
This patch removes all the hard-coded DEBUG definitions pointed
by Amadeusz and adds a new (standard-like) option '--enable-debug'
to the configure script, which activates debug stuff for the devs.
The default behaviour is now to not have them.
As a side effect, the option also disable optimisation, which are
generally annoying when trying to run a debugger.
As found by Rodolfo, it looks like there could be memory leak in the
function 'RSmoothScaleImage' because it reserveda given number of
memory blocs but used another count to free them after use.
This patch uses the same count for release as it seems this variable
is not modified in between. Took the opportunity as Rodolfo proposed
to convert a global variable to a local variable - this global
definition seems incorrect.
As reported by Amadeusz Sławiński, the support for (d)quoted string
in the menu file was not exactly the same as in previous versions
of WindowMaker, so we had a regression which is not acceptable.
This patch propose a consistent handling for (d)quoted text instead
of the previous on-the-title-only code; now all (d)quoted strings
are still kept as-is (as expected) but the (d)quotes are always
removed.
As a side note, it also improve the support for \escaped character
to work correctly as mentioned in the example of the default menu.
Note: This reverts b2e7620868 which
was an incomplete solution to the problem.
As reported by Rodolfo, there are some cases when working with icons
where a crash can occur, which is related to trying to re-release
some memory that have been already freed previously.
This patch adds a reset-to-NULL of the corresponding pointers so that
on next usage wmaker will know there's no more memory associated with
these pointers.
The code was setting some values in the window's attribute structure
which were not being used (missing the corresponding vmask flag) and
was setting some vmask bits without setting the corresponding
value in the structure.
The new function 'unpaint_app_icon' checks its argument 'wapp' aginst NULL,
however it does dereference the pointer beforehand, which will end up in
a crash if the null-pointer case occurs.
As found by Rodolfo, a crash could happen when changing an icon to
a non-existent file. From his report:
Steps To Reproduce
1. Open the Settings box, for example for WMDock. (Right click on WMDock, Settings).
2. In the "Icon Image" field, write some characters. For example "asdfasdfasd".
3. Click in "OK". You get "Could not open specified icon file: GNUstep.tiffasdfasdfasd"
4. Hit in "OK"
5. Repeat 2,3,4 some times (sometimes 1, sometimes 2, sometimes 10).
The problem came from freeing the previous icon but not updating
the pointer, so if the new file did not exist it ended up with a
fake-looking icon being used, hence possible crash.
BTS: http://www.kix.es/mantis/view.php?id=1
When generating the full path+name of file to search for a file
being #included, it was generated in a buffer that's supposedly
large enough (MAXLINE > 2*PATH_MAX). However, this limit has a few
issues (PATH_MAX seem to be able to be bigger, and worse: we can't
be sure we're given longer args).
The code was rewrote to natively include boundary checks so we're
sure we won't overflow the buffer. A few strncpy have been removed
because in this case they tend to make things harder to write.
When a #include is used, the file is searched in many places; when
the file was searched in the ':'-separated list of path it could
led to infinite loop if the list contained more than one path and
that the file was not found in the first path, the ':' separator
was not properly passed over.
The Debian script for WindowMaker generates some warnings on startup
due to 'find' being cautious about its options, updated them to be
in line with expectations.
It helps to understand why a file is present in this file, which
in turn can help to understand where a file comes from and it makes
updates to the file easier
Now that the built-in parser has support for all the feature of CPP
being used by WindowMaker's default menu, we can remove the stuff
related to calling CPP:
- code for preparing and running CPP;
- compile-time option to de-activate the call to CPP;
- command-line option
Christophe CURIS