Casts between incompatible function types seem to be unavoidable here,
as wglGetProcAddress() always returns generic PROC type, so just
suppress the warnings temporarily.
These warnings are unavoidable in a few places, and testing for gcc
version and suppressing them is too verbose, so define helper macros
making it a bit less painful.
BucketFromNode typedef must be defined as a function returning size_t,
as GetBucketForNode() method in the classes generated by hash map macros
does, to avoid warnings about bad function pointer casts in Win64 build,
where size_t is not the same thing as "unsigned long".
This does the same thing but avoids gcc 8 -Wclass-memaccess warning
(harmless in this case, but potentially useful elsewhere, so it seems
better to work around rather than disable it).
The comparison could never be true as wxDEFAULT is not one of
wxBORDER_XXX values and so is not included in wxBORDER_MASK and so was
useless -- and just results in -Wtautological-compare from gcc 8.
Simply remove this check, a047aff270 added
it apparently to use themed border by default for wxVListBox, but this
was done in a better way in 28319afe55.
By now all compilers/SDKs should hopefully have {Get,Set}WindowLongPtr()
and GWLP_XXX constants, so there is no reason to keep separate, and
differing by return type in wxGetWindowProc() case (oversight?),
versions of these functions for Win32 and Win64 builds.
Combine them in a single version appropriate for both cases.
The callback used with CreateDialogIndirect() must return INT_PTR, which
is 64 bits under Win64, not LONG, which is still 32 bits.
This fixes a gcc8 -Wcast-function-type warning and might fix a real bug
in Win64 build too.
The callback must return LRESULT, which is 64 bits under Win64, not
int, which is still 32 bits.
This fixes a gcc8 -Wcast-function-type warning and might fix a real bug
in Win64 build too.
WNDPROC and FARPROC are not the same thing in MSW and it's wrong to use
WXFARPROC as the type of different window procedures we use.
Introduce WXWNDPROC which is more clear and correct and use it instead.
Also get rid of a few casts which are not necessary any longer.
This parameter wasn't used anyhow and casting wxWndProc or m_oldWndProc
to WXFARPROC just resulted in gcc8 -Wcast-function-type warnings.
Get rid of them by not passing the window proc to this function at all.
Don't use the deprecated since Win16 (!) and doing nothing
MakeProcInstance() and fix the signature of DDECallback() which was
actually wrong in Win64 build but the problem was hidden due to the
casts done inside and outside MakeProcInstance() before.
Remove this macro and fix the signature to actually conform to the real
callback one.
This also fixes another gcc8 -Wcast-function-type warning.
Put the code implementing object arrays, i.e. arrays owning pointers to
the objects, into wxBaseObjectArray<> instead of _WX_DECLARE_OBJARRAY
macro.
Also simplify WX_DEFINE_OBJARRAY() by leaving only the definitions of
functions creating and destroying the objects in it (they have to be
there and not in the template itself to allow using template with
incomplete classes).
Add wxBaseSortedArray<> template instead of putting all the code inside
_WX_DEFINE_SORTED_TYPEARRAY_2 macro.
There should be no changes, but the code is now easier to read and
modify.
Note that the control needs to be created with a reasonable size for
HitTest() to work correctly (at least under MSW, but probably not only),
so change the test set up function to use some fixed size for all kinds
of controls instead of doing it only for the multi-line ones.
Right clicking didn't work under wxGTK where it just showed the context
menu, so use left click handler instead and check whether "Alt" is
pressed.
Also add a note to make this test more discoverable.
Dramatically simplify dynamic array macros by keeping only the
implementation previously used in wxUSE_STD_CONTAINERS case and dropping
the other one, as we can use the std::vector-based implementation on top
of our own wxVector<>, which is available whether wxUSE_STD_CONTAINERS
is 0 or 1.
This allows to get rid of tons of ugly macro-based code without breaking
compatibility.
After bd650ec and cad2b9c, showing a popup menu in wxTextCtrl with
wxTE_RICH2 and then dismissing it without choosing a command would
result in the cursor flickering between arrow and beam cursors until
another context menu was shown.
See https://github.com/vslavik/poedit/issues/483
Fixed by checking if the popup menu is still valid (and so shown) in
the cursor handling code.
Still use Connect() in unit tests which were written explicitly for it
and in EventConnector, which can't use Bind() as it uses a variable for
the event type.
No real changes, just use the newer and more convenient function.
These events are unexpected and inconsistent with the other platforms,
including GTK+ 2, so make sure we don't send them during the
destruction.
Closes https://github.com/wxWidgets/wxWidgets/pull/821
Don't repeat the same check for !defined(WXUSINGDLL) many times.
No real changes, this is just a simplification (which should be viewed
with "git diff -w --color-words" to see how little has really changed).
MSVC users can simplify their projects by prepending
$(WXWIN)\include\msvc to the compiler include directories, which results
in automatic including the build-specific setup.h and linking all
required wxWidgets and system libraries.
This worked only for the default multilib builds but not for monolithic
builds where wxWidgets libraries are created and named differently.
This patch allows the users (by defining wxMONOLITHIC=1 in their
project) to use it also for monolithic builds.
Somehow this compiled with the previous gcc versions (as well as MSVS),
but a static_cast from an integer wxUIntPtr type to a pointer HANDLE
type is obviously invalid and a reinterpret_cast is needed here.
This fixes compilation with g++ 8.
The standard margins, expressed in millimeters, could result in the
usable page space being much smaller than 1000px used for the DC size
when using higher DPIs, which means that the test checking that a 2400px
image took only 3 pages could fail, as it could require 4 of them in
this case.
Fix this by getting rid of the margins, as this should ensure that the
page height is exactly 1000px now, independently of the actual DPI.
The actual number of pages is given by the number of elements in
m_PageBreaks vector after CountPages() is shown and is unknown until
then, so m_NumPages is completely unnecessary and can be just removed.
Prefer the use of the standard-like template class to macro-based list.
It also makes more sense to use a vector rather than a linked list here,
as the elements are never removed from or inserted into m_Filters, so
there is no reason to prefer the list structure.
This parameter is not actually needed for any reasonable page breaking
algorithm, as it shouldn't care about any previous page breaks except
for the last one, which is already known as "*pagebreak - pageHeight" in
this function.
Removing it will allow to stop using wxArrayInt in the code using this
method and switch to an std::vector<int> instead.
This is not a backwards compatible change, but it seems that there is
very little code actually overriding this function (none could be found
in any open source repositories) and updating it should be as simple as
removing the now unneeded argument.
There is no need to look in the known page breaks array for a page break
at the position of this cell, all we care about is whether this cell is
on the current page or not: we must insert a page break if, and only if,
it is.
No real changes in behaviour, but the code is now much simpler to
understand and this change paves way for removing "known_pagebreaks"
entirely, as it was only added for the use in this function (see
f2034f1b6e from 15 years ago), which
doesn't even actually need it.
It should also make pagination code somewhat faster.
