Use the wxString-returning form of wxRegex.GetMatch() because the bool
form always returns true if the initial match succeeded, causing every
hat direction in the config to be interpreted as north, the first
condition in the if statement.
Fix#1192
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Start Network Link is not allowed while Local mode is selected, but
state change of Local mode was ignored until an option which triggers
EnableNetworkMenu() was selected.
Removed EnableNetworkMenu() from the Configuration option because it
doesn't change anything related to Link menu state.
Add references to frameworks being linked to `buildInputs` in
`default.nix`. Remove `-framework System` from `wxWidgets_LIBRARIES` in
the cmake code as it's not necessary and nix does it have it in
`darwin.apple_sdk.frameworks`.
TODO:
Debug build currently builds and runs. Release build builds but does not
run because of a problem with the dylib bundling.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
On initialization, if wxWidgets was compiled without EGL support, check
in environment variables if running under Wayland and if so tell GDK to
prefer using an X11 backend, causing the program to launch under
xwayland.
Fix#1028
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
In doMirroring() do nothing if the ROM size is greater than 32MB. This
fixes these ROMs automatically working without disabling mirroring in
settings.
Fix#1046
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
The NO_LINK build had bitrot and was no longer building. This fixes the
issue by properly disabling Link-related code, including unreachable
functions when the project is built with ENABLE_LINK=FALSE
Previously, MBC7 EEPROM was saved at address 0xa000 in memory, rather
than inside the `gbRam` data buffer. This was inconsistent with other
mappers, resulting in issues like the EEPROM data being cleared on
`gbReset`.
Fixes#1173
Since Wii and Wii U are both big-endian systems, we need to call in the Makefile for Libretro cores that these must use big endian code for be able to work, otherwise the emulation will just give a white screen.
Fix by Ploggy, many thanks to him.
Separate writing the version into the Info.plist on macOS into a
separate action from the online updates Sparkle framework into a
separate POST_BUILD action because we are not using Sparkle for macOS
builds right now due to codesigning issues.
Also sort tags by reverse tag name in the cmake git version extractor
because a tag was rewritten.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
CMAKE_CROSSCOMPILING is true when building arm64 on an x64 host, so
ignore this on MSVC.
Eventually we want to run osslsigncode through wine when cross compiling
for win32 on other platforms, or use the native signtool.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Check if the OpenGL context is valid on every operation, and if not
re-initialize it. Also rebind the textures every time the frame is
drawn.
Also rename the SetCurrent method to SetContext, which is clearer.
Fix#643Fix#767Fix#840Fix#843
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Make a bool SetCurrent() wrapper over checking the wxGL_IMPLICIT_CONTEXT
define and executing the appropriate behavior to set the OpenGL context.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Change the cmake code to install the translations gettext .mo files into
the .app bundle for non-release builds too.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
First launch would fail because wx does not create the configuration
directory by itself if it does not exist.
This also fixes an incorrect menu option.
The lz77 uncompresssion software BIOS implementation was exiting early
when uncompressing data, if the overall length was larger than
advertised in the function parameter. However, real GBA BIOS does read
further than the advertised length, so we do here too. This fixes
Advance Wars title screen.
This also adds a SoundDriverVSyncOn implementation, silencing an error
for some Shrek games, though they still cannot properly boot with an
emulated BIOS.
Fixes#789
* Removes wxAcceleratorEntryUnicode and assorted arrays in favor of a
new class, `config::Shortcuts`, which handles UserInput assignment to
commands and resolution at runtime. `config::Shortcuts` also handles the
INI user configuration in a backwards-compatible way. Runtime
resolution of UserInput to command is also now logarithmic rather than
linear.
* The same shortcut can no longer be assigned to 2 different commands,
which fixes#158.
* Moves the `AccelConfig` dialog to its own dedicated class.
This replaces all uses of wxJoyKeyCtrl with a new custom widget,
UserInputControl. Internally, this class keeps track of UserInputs
associated with the current control, allowing direct access to the
UserInputs, rather than going through string conversions.
Acceleration handling is simplified by going through UserInput -
converted to the key, mod, joy triplet for now - rather than handling
string comparisons.
The `wxAcceleratorTable` for the `GameArea` panel was initialized with an
array of `wxAcceleratorEntryUnicode`, while the API calls for an array
of `wxAcceleratorEntry`. This resulted in the array not being properly
interpreted by the `wxAcceleratorEntry`. This fixes the issue by using
`wxAcceleratorEntry` to instantiate the array.
These can be accessed via the global gbCartData object. In addition,
this cleans up gbMemory to remove dead code that was used as a
workaround for ROM hacks.
We rely on the rom size in the cartridge header as a proxy for the RAM
size. Only 3 cartridges exist that use MBC7, Korokoro Kirby / Kirby
Tilt'n'Rumble and Command Master. Both versions of Kirby use a 256 bytes
EEPROM while Command Master uses a 512 bytes EEPROM. There does not seem
to be any other way to reliably get the EEPROM size for MBC7 cartridges.
Some MBC7 cartridges use 512 bytes EEPROM rather than 256 bytes. For
compatibility with older versions, we keep saving and loading 256 bytes
of EEPROM for every MBC7 cartridge. TODOs have been left to figure out
how to identify the EEPROM capacity properly.
* Removes all of the per-mapper helper functions in favor of an
iovec-like structure that is set on ROM load and ROM patch.
* Changes the MBC7 RAM size to 256 bytes, which is what the cart
actually uses.
* Works around issues with homebrews by overriding some of the header
data. For instance, battery is disabled if a cart specifies a battery
but no RAM and no RTC.
Test: This was tested with every known RTC/RAM/Battery variations by
creating a battery file before these changes and checking that the
battery file is still loaded properly with these changes applied. The
reverse was also tested to ensure compatibility.
* Simplifies most of the GB save/load state code and changes many
of the global variable uses to go through gbCartData instead.
* Changes all battery reads to use gzFiles. This seems to have been an
oversight in an earlier refactor of the code.
* Removes some unused variables and stops exporting internal-only
functions in GB.cpp/gb.h.
Breaking change: Cartridges with no mapper, but a save battery should
now properly save and load changes. However, this type of cartridge has
not been used in any commercially released software for Game Boy so the
exact intended behavior is speculative.
Add MOVE_START/MOVE_END events to detect when the window is moved or
resized and pause sound on Windows to prevent the DirectSound driver
from looping.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Make some tweaks to the cmake files to support
CMAKE_BUILD_TYPE=RelWithDebInfo, release with debug information for MSVC
vcpkg builds.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
* Moves most directory settings to g_owned_opts.
* Converts the DirectoriesConfig dialog to the new structure.
* Various includes clean-ups.
Breaking change: Loading a GB/GBC/GBA ROM when the ROM directory setting
is unset for that platform will now populate the per-platform ROM
directory setting.
Further testing showed some limitations with the current GB header
parsing. Many homebrew and some licensed cartridges do not follow the
expected format. In particular, the manufacturer code used in the "new
format" is not always used properly.
Some mapper flags seem to not actually exist in practice. These have
been documented as comments.
The header parsing code was creating string objects with '\0' bytes,
this is now properly sanitized.
The destination code flag was not being set properly. This has been
fixed.
This adds checks for every allocation failure in the core Game Boy
emulator. In addition, this replaces some magic values defined
constant expressions and removes an unused function.
This also removes asserts for failed allocations in GB.cpp, they now
report a failure to the caller.
Pause only sound on menu open on Windows, and resume on menu close if
emulator is not paused.
This is to avoid DirectSound looping on menu open.
Fix#788Fix#1077
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
On Linux, when not running under Wayland and the app was built with XSS
(X ScreenSaver) support, optionally suspend the X11 screensaver when a
ROM is loaded and not paused.
Add a "Suspend ScreenSaver" checkbox in UI Settings bound to an option
to enable this feature, visible only under X11.
Defaults to off, as there is already a call to suspend the screensaver
on joystick input, which may be sufficient for the user.
TODO: Add necessary support for other platforms.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
This adds a new class, gbCartData to do all of the GB/GBC header parsing
in a single location.
Breaking change: ROMs advertising a 2 KiB RAM size now properly only
have 2 KiB allocated for RAM, whereas it used to be 8 KiB. This was
tested with the one aftermarket ROM making use of this RAM size
(Quartet), with no issue. Save files and save state files should still
be compatible since the RAM was saved with the correct 2 KiB size.
Some software (like the EU version of Digimon Racing) relies on the
SIOCNT register being properly updated when no cable is connected.
Otherwise, they risk getting stuck in a loop waiting for an update on
the SIOCNT register.
This changes all of the options handled by the GameBoyConfig dialog to
use the new Option type. This also fixes a number of small issues
related to palette handling and GB emulation:
* Modifying the custom GB palette no longer incorrectly applies to GBC
and SGB modes.
* Loading a GB save state (not GBC or SGB) properly applies the
user-defined palettte and not the one in the save state.
* The GBC core now accepts .gbc BIOS files.
Finally, this modifies and renames wxFarRadio to widgets::GroupCheckBox.
In addition GroupCheckBoxes can now be loaded and fully configured from
the XRC file rather than having to be manually configured on dialog
initialization.
This game cart may have a flashcontroller but no flashram onboard, the game automatically goes down the Flash Type path and hits a stackoverflow at cpuSaveGameFunc after cpuSramEnabled returns as false
I'm not sure of the specifics, but with this change it seems to avoid this condition.
I have tested a number of titles that create Sram, 512 and 1024Mbit flash and they all still seem to just fine.
The previously used algorithm was too aggressive. This fixes the issue
by using the last second data and using a more gentle adjustment curve
to prevent huge changes in frame skipping.
Fix some more strings marked for translation in the source code to be
more correct and easier to localize.
References #1071#1029
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
This introduces a new INI config option called "IniVersion". Whenever
default values change or options are renmaed, the latest version number is
incremented and an update path is run.
This change also rewrites part of the initialization routine by adding
more safety checks before creating or reading the configuration file.
This also removes limits on the main window position so it can now be
negative. This is necessary on Windows where multiple screens to the
left and top of the main screen have negative coordinates. A check at
startup ensures we always restore the window within the drawable area.
This introduces an OPTION macro, as well as assignment operators and
convenience conversion operators to make it easier to treat the
OptionProxy object (renamed from Proxy) as a variable.
This removes the global accessors for individual options in favor of a
proxy template class with a simpler per-type API. This adds compile-time
checks to validate that the right option is used with the right type
when accessing it directly.
Options used by the core emulator were set with global variables. This
moves all of the options used by the core emulator into a new struct,
shared by all 3 frontends.
Fix some source strings in the source code that are marked for
translation to be more formally/grammatically correct and easier to
localize.
References #1029
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
The DisplayConfig dialog had a few issues:
* The OK/Cancel buttons were not working as expected. The configuration
options would be immediately written whenever some control values were
changed. The configuration options are now properly being set on a
click to OK, and ignored on a click to Cancel.
* The X button works again.
* There is now proper validation for the Display Scale option,
validation ensures that the value is in the right range and incorrect
data can no longer be input in that field.
* The Direct3D option is now properly enabled on Direct3D builds.
* Clean-up ConfigManager options
Many of the options in ConfigManager were not actually used.
* Some options are only used by wx, they have been moved to gopts. In
addition, the type of the option has been changed to "bool" when
applicable.
* Some options are only used by SDL and do not have a corresponding
configuration option, they have been moved to the SDL implementation.
* Some options are not used at all, they have been removed.
Fix rvalue variable declarations in be1e5ca1, the && associates with the
variable, not the type.
This has no effect on functionality.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
When executing the EGL subsurface move backport fix, check for loaded
library version < 3.2.2 so as not to execute an unnecessary EGL
subsurface move operation on versions which will already have the fix.
References #1028
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
A regression from #1008 was identified as the root cause for #1050 and
fixed in #1051. This is a follow-up that changes the constants used
for `sineTable` to be unsigned hex, which fixes warnings in MSVC
without breaking changes.
Bug: #1003
Silence output of nuget commands used to get gettext tools for the
Windows build.
Remove Objective-C++ source files (.mm extension) from the list of all
sources passed to xgettext to silence the warning about the file type
being unrecognized. These do not have any strings that need to be
translated.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Adjust cmake code to support building a Windows arm64 binary on a
Windows x86/x64 host.
Add /EHsc to MSVC compiler flags for exception unwind support.
Check VCPKG_TARGET_TRIPLET in Architecture.cmake to set the appropriate
cmake variables for cross-compiling for arm64.
Add MSVC_x86_Host_Compile.cmake script, used from HostCompile.cmake, to
adjust the path of the cl.exe compiler and ENV{LIB} to build a C program
for x86 when cross-compiling, necessary for the bin2c utility.
Use wxrc from dependencies submodule when cross-compiling, because the
cross wxrc executable cannot be run on the host.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
This reverts commit 1ac7ac999b.
The %% sequence turned out to not be dealt with specially by gettext
tools.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Check for the condition where the windowing system has wayland support
but not wayland EGL support for wxGLCanvas in wx, in which case the
Simple renderer must be forced.
This is currently how the wx packages are built on Fedora.
Refer #1028
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Disable the EGL subsurface move backport for Wayland GLCanvas if the wx
config sets wxUSE_GLCANVAS_EGL=0 (off.)
This fixes the build failure on Fedora. To fix the functionality,
however, they need to enable the EGL GLCanvas in their wx builds.
Also use a different hack to access the private wxGLCanvasEGL member
using template instantiation instead of redefining a keyword as a macro.
Refer #1028
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Add a few translation markers for strings that are translated in the
wxWidgets catalogs, which we do not pull yet.
Also change the string "% of normal:" to "percent of normal:" because
the percent sign was causing an error in poedit.
Fix#1029.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
cmake:
Add cmake option `DISABLE_MACOS_PACKAGE_MANAGERS` to disable
automatically using macOS Homebrew, MacPorts or Fink.
Use `-Ofast` instead of `-O2` for default compiler optimization flags,
which enables `-ffast-math` among a few other tweaks. There is a lot of
controversy about this, but in our testing on MSVC with `/fp:fast` the
performance increase for this specific app is dramatic.
Update version regexps on the SFML headers in `FindSFML.cmake` to allow
for a variable number of spaces between the macro and its definition.
Update the version of the Sparkle update checker framework for macOS
from 1.23.0 to 2.3.0-beta.2, and add a command to the cmake commands for
copying the framework to the `.app` to delete it first if already
copied.
Builder dist changes:
Updates:
dist old v new v
==============================================================
xz 5.2.4 5.2.6
openssl 1.1.1f 1.1.1g
curl 7.65.3 7.85.0
ccache 3.4.3 4.6.3
cmake 3.19.4 3.24.1
autoconf 2.69 2.71
autoconf-archive 2017.09.28 2022.09.03
libtool 2.4.6 2.4.7
help2man 1.47.5 1.49.2
libiconv 1.16 1.17
gsed 4.4 4.8
bison 3.0.5 3.8.2
texinfo 6.5 6.8
flex e7d45afc ea6493d9
libicu 65_1 71_1
nasm 2.14.02 2.15.05
pcre 8.41 8.45
libffi 3.2.1 3.4.2
libxml2 2.9.12 2.10.2
libxslt 1.1.33-rc2 1.1.37
XML-SAX 0.99 1.02
expat 2.2.4 2.4.8
libjpeg-turbo 2.0.4 2.1.4
libtiff 4.1.0 4.4.0
freetype 2.10.0 2.12.1
libgd 2.2.5 2.3.3
liberation 2.0.1 2.1.5
urw 91edd6ec 20200910
graphviz 4bf0ec9b 5.0.1
python2 2.7.15 2.7.18
python3 3.9.1 3.10.6
swig 3.0.12 4.0.2
doxygen 1.8.14 1.9.5
bakefile 33c270a5 7a218d89
XML-Parser 2.44 2.46
ninja 1.9.0 1.11.1
meson 0.51.2 0.63.2
glib 2.62.2 51251940
libgpg-error 1.41 1.45
libgcrypt 1.8.5 1.10.1
libsecret 0.19.1 0.20.5
sdl2 2.0.10 2.24.0
flac 1.3.3 1.3.4
libogg 1.3.4 1.3.5
libvorbis 1.3.6 1.3.7
harfbuzz 1.7.5 5.1.0
sfml 03375688 2.5.1
shared-mime-info 1.9 2.2
wxwidgets 9cc0c9a0 31720e1e
graphite2 ce0e793f 425da3d0
xvidcore 1.3.5 1.3.7
fribidi 1.0.1 1.0.12
libgsm 1.0.18 1.0.22
libmodplug 5a39f591 d1b97ed0
libopencore-amrnb 0.1.5 0.1.6
snappy 1.1.7 af720f9a
speex 1.2.0 1.2.1
vidstab 1.1.0 90c76aca
libvo-amrwbenc 0.1.3 3b3fcd0d
libass 0.14.0 0.16.0
libbluray 1.1.2 1.3.2
libvpx 1.8.1 1.12.0
libx264 20190913-2245 f7074e12
libxavs distrotech-xavs-git 8300fbf4
libzmq 4.2.2 4.3.4
ffmpeg 4.3.1 5.1.1
Removed dists:
dist ver
=========================
zlib 1.2.11
pkgconfig 0.29.2
docbook5.0 5.0
libwavpack 5.1.0
Added dists:
dist ver
========================
zlib-ng ce01b1e4
zstd 1.5.2
hiredis 1.0.2
pkgconf 1.9.3
pcre2 10.40
docbook5.0.1 5.0.1
docbook5.1 5.1
Changes:
Update all non-working dist URLs.
Replace zlib 1.2.11 with a recent zlib-ng built with `--zlib-compat`.
Add zstd and hiredis, new deps of ccache.
Replace pkgconfig with pkgconf.
Use both pcre1 and pcre2.
Remove the libxml2-python dist alias and rebuild libxml2 `--with-python`
and `--with-python3` after python 2 and 3 have built.
Remove libwavpack as ffmpeg 5 no longer uses it as a dependency.
Builder core:
Rename some of the directories and files to be a bit nicer.
Add `-Ofast -march=core2 -mtune=generic -pipe` to the global
`{C,CXX,OBJC,LD}FLAGS`. These are the optimization settings used by
default in the project as well, while `-pipe` significantly reduces
build time by making the compiler use pipes internally.
Add `-DFLOAT_APPROX` to the global `{CPP,C,CXX,OBJC}FLAGS` because we
are now using `-Ofast` and consequently, `-ffast-math`.
Set the `LIBRARY_PATH` variable in the global environment.
Make some adjustments for the new dist versions and make some other
improvements.
macOS builder fixes (tested on High Sierra 10.13 and Catalina 10.15):
Add support for predefined BUILD_ROOT and append BUILD_ROOT_SUFFIX.
Allow pre-defined CHECKOUT directory for building the project sources.
Sanitize `PATH` before build to remove `/usr/local` etc., use
/usr/local/bin/brew explicitly for deps instead.
If llvm and clang is installed from nix or brew, prepend the path of the
highest version found to `PATH`.
Remove `-lm` from build environment as macOS has never had an `m` math
library and this causes problems in some config checks.
Add LLVM `libunwind` from the LLVM archive.
Add a patch for `glib`'s `meson.build` to statically link
`gettext`/`libintl`, adds `-liconv -framework CoreFoundation`.
Add an extra location to check for `stdint.h` in XCode.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
* Add convenience direct accessors to individual Options.
* Move BasicOptionObserver to a separate header and allow it to
observer multiple options. This also removes the include for
"config/option.h" for using this observer, reducing build time.
* Never skip the "Plugin" option when changing filters. This is
already handled properly in the GameArea class. If the plugin is
unusable, GameArea resets the filter to the default value.
* Make the OptionID names slightly less long and fix the inconsistent
naming in the enum.
Fix regression in ea2a929f, which broke x11 support by not checking for
Wayland when making the Wayland EGL subsurface move call.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Backport 800d6ed69b from wxWidgets git to fix resizing the
`GLDrawingPanel` in Wayland using the EGL `wxGLCanvas`.
Also rename `IsItWayland()` to `IsWayland()`.
Fix#1028
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
When processing the `OnSize` event in the `GameArea` and calling the
embedded `DrawingPanel` `OnSize` event, call `Layout()` afterwards.
This seems to improve the resizing behavior of the EGL `wxGLCanvas`
`DrawingPanel` on Wayland.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
This adds a generic Observer interface to config::Option. To
demonstrate its intended usage, the Display Configuration dialog and
the options it controls have been updated to be entirely handled via
the config::Option class.
Implementations for wxValidator are used to validate the flow between
the UI and the underlying Option. In turn, modifying the Option value
triggers all of its observers that should then do what they need to
do.
Rather than explicitly calling all of the needed methods after
modifying a global option value, the UI elements that need to be
notified when an Option value is modified will be notified via their
observers. Runtime assert checks are put in place to prevent infinite
recursion if an observer attempts to modify an Option while handling
the observer callback.
Once all uses of Options have been updated, we should get into a state
where the following will be true:
* cmdevents.cpp will no longer rely on the application state.
* All dialogs will have been moved to specific implementations,
reducing the size of guiinit.cpp
* update_opts() will no longer need to be called and will be removed.
This will then make it easier to update accelerator handling to be
done with config::UserInput.
Bug: #745
Destroy the DrawingPanel on vsync checkbox toggle so that it is
re-created with the vsync option in effect.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Add support for `eglSwapInterval()` to enable or disable VSync on
Wayland.
Also add status messages for enabling or disabling VSync, for both EGL
and GLX.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
wxWidgets 3.2+ has experimental `wxGLCanvas` support for Wayland EGL,
enable it for those versions.
The `glXQueryExtensionsString()` method for enabling or disabling vsync
does not work on Wayland currently, so it's disabled.
TODO: Enable/disable vsync on Wayland EGL.
Fix#1028.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
* Remove the custom handling for pre-3.1.4 builds on Windows since
both the mingw and msvc builds now use 3.2.
* Remove wxDPIChangedEvent handling, since this triggers a resize,
which has the same effect.
* Implement main/WinMain ourselves, rather than relying on wxWidgets
macros. This allows us to initialize the console earlier in the
process.
* Add an explicit dependency on the manifest file to trigger a link
when it is modified.
This converts the HiDPIAware into a series of helper functions, using
wxWidgets native support starting with wxWidgets 3.1.4. Custom
implementations are provided for Mac and Windows.
In addition, this fixes a few issues with the Windows build:
* Use the same defines variables in the MSVC and MinGW builds, with
g++ and Clang.
* Add a custom manifest for the application, indicating full DPI
support and support for recent Windows versions so the application
will no longer run in compatibility mode by default.
* Manually spawn a console in debug builds if none is attached and
always use the parent console if it is available otherwise, even in
release builds. This also removes the /subsystem:console linker
argument.
Remove constexpr in user-input.h as LLVM 13 complains that it is
illegal.
Use variant-lite as nonstd::variant and related types as some variant
functions are macOS 10.14 Mojave only.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
* Remove the wx prefix and make UserInput a class in the config
namespace.
* Make UserInput constexpr-constructible. This should reduce program
startup time by having the compiler pre-allocate UserInput objects.
* Change initialization of UserInput objects to be made via
constructor. A previous approach would have limited creation of
UserInput objects to a few places but that proved unworkable in the
long run.
* Changed the string to enum conversions to be built at runtime from
the existing constant array rather than redefining the strings.
* Fixed an issue in the initialization of the `checkable_mi` array.
A test had been mistakenly removed during the refactor.
* Fixed an issue where changing a bool or int option would cause an
assertion error at runtime. Changed helper methods definitions to
use a pointer to a variable rather than take a parameter by
reference, clarifying the intent from the caller perspective.
* Fixed the `renderer` enum definition to properly exclude Direct3D or
Quartz2D depending on the platform.
* Various comment fixes.
This change introduces a new class to handle all of the INI options
data, VbamOption. A VbamOption represents a single option in the INI
file. It is not constructible outside of its implementation, which
prevents the initialization of incorrectly formatted options.
Internally, a VbamOption points to a global variable in memory, which
holds the value used at runtime. This is because various parts of the
codebase edit the global variable rather than going through a central
registry. This also means we need to separately update the INI values.
In the future, we may be able to convert all existing reads and writes
to the global variable to go through VbamOption. Individual UX elements
could link directly to getters/setters for a specific VbamOption rather
than have duplicate data.
VbamOption replaces the opt_desc struct and the global opts array. All
users of opt_desc and the global opts array have been updated.
This is a necessary preliminary change to better support a refactor of
accelerators, through wxUserInput.
Issue: #745, #158
Add default menu accelerator `CTRL+I` to `ChangeIFB` which is the menu
item to change the interframe blending algorithm.
CMD+I on macOS.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
When changing the interframe blending algorithm via:
`Options -> Video -> Change interframe blending`
, show the short name of the algorithm instead of the list number in the
OSD.
Store a pointer to the wxChoice of interframe algorithms read from the
XRC for easy access.
Shorten the names of the algorithms in the XRC selection list.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Use namespace strutils:: instead of the `str_` prefix for the functions.
Remove `vec_find` as it's just an alias for `wxArrayString::Index(str)`
and is not a function on a string.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
In the warning for detected translations source .pot changes, state that
the .pot will be automatically pushed to Transifex, instead of asking
the developer to do that, as they may not have access or the utilities.
The Transifex pull job has been updated to do this.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Make the wxvbam.pot output a top-level build target and the check and
warning message for developers a POST_BUILD command for it.
Broken by 60caad86.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Save the wxChoice* for the pixel filter name selection list in the XRC
and use it to display the filter name in the status message on pixel
filter cycle instead of the filter number.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Make translations.zip a dependency of the app and remove it as a
dependency of other commands/targets that do not use it as a source.
Silence the very spammy output of the zip program.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Add static GetX11Display() function to get the app X11 Display* instead
of making the GDK/GTK calls in two places.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Pass DefaultScreen(display) as the screen number to
glXQueryExtensionsString(display, screen), this may work more reliably
for some configurations than always assuming screen 0.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Update doctest.h because the old version causes build failures on Linux.
Also move it from third_party/include/doctest to third_party/include.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Previously, wxJoyEvent surfaced the internal state of the joystick
changes, forcing consumers of this class to convert the values to the
mod and key values used in the rest of the code base. This streamlines
the wxJoyEvent API so the events sent to the consumers can be used
as-is. In particular, this allows us to remove manual generation of
"ghost" events in GameArea.
Breaking change: This has the side effect of disabling setting
diagonals from a joystick HAT as a discrete control. However, this only
ever worked with game controllers not compatible with the newer SDL
GameController API, since it handles HATs as 4 discrete buttons. For
compatibility purposes, these have been moved 1/8 turn
counter-clockwise - i.e. NE is now N.
Issue: #745
Fix the state of the File -> Play -> Start/Stop menu after playing a
movie by calling systemStopGamePlayback() when the playback stops.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Add new format for recording VBA Movies that records inputs since the
time of the last input instead of the beginning of the movie.
Keep the extension `.vmv` the same, the format is determined from the
header of the file.
Make this the new default format for recording VBA Movies.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
On MSYS2, stop using the MINGW64 windres.exe Windows resource file
compiler instead of the CLANG64 provided one, because the CLANG64
windres.exe works correctly now, while using the MINGW64 windres.exe no
longer works.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Switch usage of std::optional to nonstd::optional for now because of an
error from the Mac builder script, I cannot reproduce the problem yet
but will try to fix it later, it may not be fixable when targeting 10.7.
The header is from:
https://github.com/martinmoene/optional-lite
.
Usage is about the same, to include:
`#include "nonstd/optional.hpp"`
Then use nonstd::optional and nonstd::nullopt instead of the std::
counterparts.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
This introduces abstractions for a game key (`wxGameKey`), a game
control (`wxGameControl`) and a common global handler for in-game
user input processing (`wxGameControlState`).
User configuration is changed from a vector to a map of `wxGameControl`
to a set of `wxUserInput`, which simplifies input configuration updates.
User input processing for in-game controls is now unified between
keyboard and joypad input, and is much faster in general since access
to game control state is now always logarithmic rather than linear.
This comes at the expense of slightly slower user input configuration
updates. However, in the worst case scenario, this is still done in
O(log(n)).
This removes all uses of `wxJoyKeyBinding`. However, some uses of the
key, mod, joy triplets remain and will be cleaned up in follow-up PRs.
Issue: #745
Previously, a manual polling of every joystick was performed every 25ms,
resulting in many CPU resources being wasted. This seems to have been
put in place to deal with issues in SDL where joysticks connected after
SDL was initialized did not fire SDL events properly. From manual
testing, this issue seems to have been fixed.
This also fixes a bug in handling of legacy joystick hats, where an
incorrect conversion for the event value was performed.
Finally, this fixes minor typoes and renames
`wxSDLJoyState::ProcessEvent()` to `wxSDLJoyState::ProcessSDLEvent()` to
suppress a warning about a method override.
This introduces a new abstraction for any user input. The long-term goal
is to replace every usage of "custom" {key, mod, joy} triplets in the
code base with this new class.
This class implements comparison operators, allowing for faster access
in a set or as a key in a map, compared to the vectors currently used.
Issue: #745
* Add wxJoystick to abstract what was previously referred to as
|player_index|. This is a first step towards a larger refactor of
input handling.
* Remove "SDL" from types that are not directly SDL-related, namely
"wxSDLJoyEvent", "wxSDLControl".
* Rename "wxSDLJoy" into "wxJoyPoller". This clarifies the use of this
class.
Issue: #745
Use file(GLOB ${lib_name}*.lib) to detect the appropriate suffixes for
wx dependency libs, because the suffix keeps changing, sometimes there
is a 'd' added for debug builds and sometimes it changes to not adding
the suffix, this will make the code more robust and lead to fewer build
failures.
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Stop setting the build_prefix* variables as they are no longer used and
relying on build trees being available would break /installed caches
etc..
Signed-off-by: Rafael Kitover <rkitover@gmail.com>
Rafael Kitover