Serialize them to strings in wxFileConfig, just as we always did for
long, but use wxRegKey support for storing them directly to the registry
in wxRegConfig.
This uses the native registry support for 64-bit values available since
Windows XP.
Note that SetValue64() can't be just called SetValue() as this would
introduce ambiguities when writing an int or an enum to the registry.
This could be avoided by providing overloads for all the arithmetic
types, but it's arguably better to be more clear about what exactly is
being written to the registry in this low-level class and using a
different name is definitely simpler.
Don't repeat exactly the same code for 3 different types, just use a
template function instead.
Change the existing TryGetValue() helpers to take all output parameters
consistently by pointer, rather than taking only long by pointer and the
other by reference. This allows to isolate the inconsistency of the type
of wxRegKey::QueryValue() parameter in a single place here, hiding it
from the rest of wxRegConfig code.
Add an extra unit test for writing/reading longs to/from wxRegConfig.
This internal function will be useful to check if the modules are
already initialized, i.e. if the library is in the "steady state"
between the end of the initialization and the beginning of the cleanup
phases.
Simply use wxVector instead, this shouldn't be less efficient (we rarely
remove the modules from the list and iterating over a vector should
actually be faster, as well as consuming less memory), but it avoids
ugly macros, is simpler to use and to debug and will be trivial to
replace with std::vector<> in the future.
No real changes, this is just pure cleanup.
The way wxBU_EXACTFIT is implemented in commit c1bb80987f
("Improve implementation of wxBU_EXACTFIT style for wxButton under wxGTK2",
2020-04-12) is too intrusive and disrupts button's appearance in some
themes. It should work better with themes if we reduce inner border in
a more GTK-compliant way by applying a dedicated GTK style to the button.
Closes#19081.
Replace "equal" parameter with the (opposite) "lowerBound" one with a
more descriptive name, add a comment describing it and also remove the
default value to force explaining how it is used in the caller.
No real changes.
Just the error code is not very useful as it doesn't say anything about
what exactly failed, e.g. seeing "The buffers supplied to a function was
too small." doesn't help understanding which function was passed a too
small buffer, so add an extra parameter to SetFailed[WithLastError]() to
log this information too.
Also log the error code itself, because SEC_E_BUFFER_TOO_SMALL is
arguably more clear than its ungrammatical error message.
No real changes.
Closes https://github.com/wxWidgets/wxWidgets/pull/2247
Use the actual type of the value, not LongestValueType, for storing
m_min and m_max as this is necessary for the comparisons between the
value and them to work correctly for unsigned types.
Also check for precision loss when converting from the bigger
LongestValueType to the actual type.
Add new unit tests, that failed before, but pass now.
This is necessary in order to deal with the numbers greater than
wxINT64_MAX that can't be represented in just long long.
It also allows to implement the intuitive handling of minus sign for the
unsigned numbers, i.e. not to accept it in FromString(), unlike the
standard functions which do (and parse -1 as 0xffff...fff).
Also extend the tests to check for more boundary cases.
Changes of d245dc9e1f (Fix drawing of dotted lines with wxDC in wxMSW,
2020-03-27) improved the appearance of dotted and dashed lines in wxMSW
but at the expense of significant (up to a factor of 300) slowdown.
Allow the applications for which the drawing performance is important to
explicitly request the old behaviour, with uglier, but faster, pens by
choosing to use low quality pens.
Update the graphics benchmark to allow specifying the pen quality and
verify that the performance when using it is the same as before 3.1.4.
See https://github.com/wxWidgets/wxWidgets/pull/2218
See #7097.
Closes#18875.
Actual implementation of IsCharOk() didn't correspond to the comments in
it or near the function declaration which stated that it's never called
with ch='-' as argument -- it was called with it and called IsMinusOk()
right before a comment saying that it doesn't need to do it, which was
very confusing.
Fix this by making the behaviour really correspond to the comments and
handling '-' at the base class level. This required introducing a new
pure virtual CanBeNegative() function, but it's going to be useful for
other things later too.
Still keep IsMinusOk() helper, but make it private now because it
doesn't need to be called from the derived class IsCharOk() any longer.
