Allow creating wxWindowPtr from wxScopedPtr<> too, to avoid having to
use .get() in the test code, and add CHECK_SAME_WINDOW() macro which
gives more useful information about the windows in case of failure.
We can't rely on HasSelection() returning true immediately after calling
SelectAll() because this operation (as almost all the other ones) is
asynchronous with WebKit and might not have completed yet when we check
for the selection existence.
There doesn't seem to be any way to wait for its completion, so just
poll the selection state for some time before deciding that it hasn't
been updated. In practice, just a couple of ms is enough on a normal
machine, but wait up to 50ms on Travis just to be on the safe side.
Note that to reliably reproduce the problem locally it's enough to run
"taskset 1 ./test_gui -c Selection WebView", i.e. pin both the main and
WebKit processes to the same CPU.
Check compilation of all wx headers with all gcc warnings enabled.
This should make it impossible to introduce problems that only appear
when -Wpedantic or -Wany-other-not-completely-unreasonable-warning is
enabled when building user code including wx headers again.
See https://github.com/wxWidgets/wxWidgets/pull/2033
We don't use "final" in our code, as very few classes in wx code have
virtual functions but are not meant to be derived from in the user code.
Ideal would be to check the existing warnings and maybe apply "final" if
it's relevant and disable it otherwise, as these warnings can be useful
to build the application code with, but for now just disable them in the
test suite.
As wx headers are included from user applications which may compile with
higher warning level than wx itself, try to check headers compilation
with almost all of gcc warning flags turned on.
This notably should prevent the headers from becoming uncompilable with
-pedantic again in the future.
For wxMSW text controls with wxTE_RICH2 style, calling SetFont() counts
as an undoable operation, resulting in CanUndo() returning true even if
no "real" changes have been made yet.
Fix this by resetting the undo stack after creating the control using
ITextDocument::Undo().
Unfortunately this interface is not available in MinGW-32, so this fix
can't be used with it.
Closes https://github.com/wxWidgets/wxWidgets/pull/2010Closes#17524.
Add wx/xrc/xmlres.h to the list of headers compiled with
wxNO_IMPLICIT_WXSTRING_ENCODING to test that they can be used even when
the implicit conversions from "char*" to wxString are disabled.
This used to be broken, see #18898, and now that it is fixed by
5a70051c7e (Avoid assertion failure in wxButton with bitmap and empty
label, see #18898, 2020-08-21) add a unit test so that it stays fixed.
Using wxRegEx in wxCmpNaturalGeneric() introduced a dependency of the
base library on the regex library.
Replace wxRegEx with character classification functions wxIsspace(),
wxIspunct(), and wxIsdigit() to remove this rather unnecessary
dependency.
Closes https://github.com/wxWidgets/wxWidgets/pull/2014
Check that deselecting a row leaves the cells outside of this row
selected.
This requires passing "true" to the previous call of SelectRow() to
prevent it from clearing the existing selection, as it does by default.
These functions got broken by the changes of cdf3187fe5 (Improve rows,
columns and cells deselection in wxGrid, 2020-03-26), but this went
unnoticed because they were not covered by the unit tests, so add the
tests for them to prevent this from happening again in the future.
This was disabled, but only didn't work because wxSlider used tiny size
in wxGTK by default previously. After correcting this in aa2d159e8c (Use
more reasonable length for wxSlider in wxGTK by default, 2020-08-04),
the unit test passes with wxGTK too and can be reenabled.
Closes https://github.com/wxWidgets/wxWidgets/pull/1992
Check that all our public headers compile with this macro defined and
that using a char string without specifying its expected encoding
results in the expected compilation failure in this case.
Make Get[MultiLine]TextExtent() behave consistently for empty strings on
all platforms, in particular return (0, 0) size for them from
GetTextExtent() in wxGTK, which is an incompatible change but is needed
to make it behave in the same way as the others.
See https://github.com/wxWidgets/wxWidgets/pull/1970
This restores the previous behaviour inadvertently changed by bfeae1922d
(Minor optimizations in GetMultiLineTextExtent(), 2020-06-10) and makes
it official by documenting it and adding tests checking for it.
It wasn't completely obviously if this was intentional or accidental
before, but at least wxStaticText itself relied on the old behaviour,
and chances are that so did some code outside the library, so make this
part of the API now.
See #18825.
This seems more logical and is compatible with wxDC in wxMSW and wxGTK2,
as well as other kinds of DC, e.g. wxPostScriptDC.
It also looks like the current behaviour was unintentional as it
happened only because wxGCDCImpl::DoGetTextExtent() always passed all
non-null parameters to wxGraphicsContext::GetTextExtent(), even if it
didn't need the values for all of them, and thus bypassed the special
case for the empty string which was already present in the latter
function.
Fix this, making DoGetTextExtent() more efficient as a side effect (we
now avoid unnecessary calls to pango_layout_iter_get_baseline() in the
most common case), and also add another test for empty string to
wxGraphicsContext itself, for non-GTK case.
Also document this behaviour and add a test checking for it.
Using "x" as separator between the components created confusion with
hexadecimal notation, resulting in confusing messages like "0x17==0x0"
that seemed to describe comparison of 2 numbers and not 2 wxSize
objects.
Using "*" is also more consistent with wxRect output format.
