The codecvt header has been deprecatd for C++17, but no replacement has
been specified in the standard. It is possible to use the Windows
specific MultiByteToWideChar() and WideCharToMultiByte() from
<Windows.h> instead, however the codecvt header still works so there's
no reason to change it until a proper replacement has been specified in
the standard.
It is additionally necessary to manually set the standard for clang-msvc
to C++17.
Having the CI on their own line prevents the initial line from getting
way too long.
The preview version link should not show in development alpha branches,
but only the relatively tested master.
The badge had not been updated for C++14 when we made the change. It had
also not been updated to reflect the dependency on Python during the
creation of the build script.
Running with "-bindingsinterface" does not allow for continually
updating the bindings or playing around with them. While changing the
unix time in the vrappconfig is slightly harder for casual users, it
will also give them a way better user experience.
All bindings are disabled by default. This is to prevent confusing
behaviour from users who do not know about the new bindings. Forcing
users to manually set up their bindings will also enable them to play
around with the SteamVR bindings interface, which allows for much more
versatile binding options that our current offerings.
All actions are curently in main action set. If the selection of
bindings was greatly expanded it could be beneficial to encapsulate the
options more. The current code is more intended as a proof of concept
until our full binding needs have been uncovered.
The processInputBindings function should probably also be split into
several functions of related bindings, as the amount of bindings grows.
openvr_init was not in a namespace. This led to it being globally
available. This was not intended.
The solution to overlays not showing up in the bindings interface was
chosen for ease of implementation and relative ease of end user use.
The Kinect2VR project are using a solution where there is a button to
shutdown background mode and use scene mode temporarily. This did not
work for our project for unknown reasons. The IVRInput system would
become non responsive and spam error messages when VR was shut down.
Implementation is here:
https://github.com/sharkyh20/KinectToVR/blob/12aeb8b33ab20eeb588c72c5d5186dd4d87784d7/SFMLProject/inc/GUIHandler.h#L700
The command line flag will be represented as a .bat file for the end
user, like desktop mode and restarting the VR server.
When Valve solves the issue of overlays not showing up in the bindings
menu, the entire function can essentially be deleted.
Official issues here:
https://github.com/ValveSoftware/openvr/issues/914https://github.com/ValveSoftware/openvr/issues/926https://github.com/ValveSoftware/openvr/issues/879
It is also possible to make the app appear by changing the
C:\Program Files (x86)\Steam\config\vrappconfig\*.vrappconfig file for
the project to have a "last_launch_time" of something greater than zero.
From experimentation it seems that there's a cutoff time around 2 weeks
before the current time. Setting it to the day before seems to be a good bet.
(In unix time).
Keeping the action name, handle and type together allows for better
error messages and limited run time type checking, to ensure that the
wrong function isn't called.
It is necessary to initialize the manifest in a class, since the Actions
depend on this being set.
Several files have been moved around to be more consistent.
Putting them in the new directory makes the automatically get included
for deployment.
legacy_actions.json was not a good name for the file, since it actually
contained default bindings for Vive controllers. vive_defaults.json
better explains what it is.
In order to test new system:
Build and deploy application.
Copy both files from IVRInput/ dir into the bin/win64 folder.
Change const std::string m_actionManifestPath in steamivrinput.h to be
directly to your action_manifest.json file.
Run and turn on a Vive controller. Press the menu button. If logs say
"NEXT SONG SET!" it works.
The OpenVR init code is currently being run in the OverlayController
constructor. Member objects that are initialized before the
OverlayController (ideally all of them) can't use OpenVR APIs before the
first line of the OverlayController constructor. Implementing the OpenVR
Init code as a class means that it is able to initialize before all the
other objects and allow them to use OpenVR APIs in their constructors.
The class has been placed in a folder called "overlaycontroller" to
signify that it should only be used by the OverlayController.
openvr_init.h uses #pragma once despite it being unofficial since it is
supported on all modern compilers.
The copy constructor, copy assignment operator, move constructor and
move assignment operator have been explicitly deleted in order to make
sure nobody tries to use the object in an unintended way, and to
document that these functions should not be automatically created.
The version error messages have been replaced with a single lambda,
since they were all identical.
The version error text should maybe tell the user to update Advanced
Settings instead of OpenVR, since Advanced Settings contains the version
strings in the binary.
No other exit functions were documented on the official docs besides
VR_Shutdown().
Since version strings are displayed in logs and on the main screen, this
will prevent situations where the master branch is confused with the
2.8.0-release branch.
The 2-8-0-release file was a one time thing because the version numbers
were already changed, it should not be created for future releases.
The separate pause and play buttons necessitated changing between them
to signal that state had changed. Replacing it with a combined
pause/play button sidesteps all the issues in syncing up when which icon
should be shown.
The old QML was a giant mess of implementation details and logic.
Reducing the buttons to only 2 essential properties allows the
MediaControllerKeys object to be more easily understood.
Moving the MediaControllerKeys into its own QML file allows the
resulting page to be more easily understood and individual pieces
swapped in and out.
The current QML directory structure is non-existent, breaking parts down
into cohesive folders will allow for easier understanding of the bigger
picture.
.pngs were chosen because the .svgs were only available in black. Qt
currently does not have an easy way of changing colors. The rest of the
interface icons are white, so getting them in white was necessary.
The sendKeyboardBackspace function was used with a non-const parameter
throughout the entire inheritance hierachy. Const has been added to
every parameter for the reasons you add const.
The original functions had the API details of sending keys absolutely
correct, but they were cumbersome to use due to not abstracting away the
press and release aspect and manual creation of INPUT structs.
The manual management of objects also saved a few operations, but with
the frequency that these functions are called it's probably better to
have more abstract code than preventing a 28 byte object from being stack
copied twice every 2 seconds or so.
* Update OpenVR from an unknown version to 1.1.3b
There shouldn't be any breaking changes.
* Add removed variables from old OpenVR version
The strings are the names of variables in steamvr.vrsettings in
C:\Program Files
(x86)\Steam\steamapps\common\SteamVR\resources\settings\
The string allowReprojection leads to warning log spam on some
occasions.
For more information see this github issue
https://github.com/OpenVR-Advanced-Settings/OpenVR-AdvancedSettings/issues/25.
* Move logging set up into own function
This does not affect the ability to log, the logging macros are
available globally.
* Clean up logging logic and add comments
The logic to find the logging location was broken, in that it always
placed the logs in the directory of the executable.
It was supposed to place the logs in
AppData/Roaming/matzman666/OpenVRAdvancedSettings/AdvancedSettings.log,
but it hasn't done this since at least git tag "v2.7". Since the change
hasn't been documented anywhere it is assumed to be an unintentional
regression.
The settings file is also currently placed in
AppData/Roaming/matzman666/OpenVRAdvancedSettings.ini, so this commit
undoes the probable regression until a strategy can be developed for
changing all paths at once.
* Move command line arg logging calls and add to manifest calls
The LOG_IF macros are cleaner than if blocks, and they do not need to be
in the middle of the Qt logic inside a giant try block.
* Remove unused QCoreApplication constructor in manifest logic
The object is not used in any of the manifest logic functions and
running without the remove/install manifest calls does not use it
either.
* Unindent logic in installManifest
The error checking was unnecessarily done in an else block. Moving it
into an if block and negating the result does the same but leaves the
rest of the logic at a much smaller indentation level.
* Change array in installManifest to be 0 intialized
Not zero initializing could lead to strange bugs if the terminating null
isn't there.
* Move manifest install/removal to separate function
The manifest handling is tangential to the actual main operation of the
program, hence moving it to a separate function improves readability of
the actual program.
The enum ReturnErrorCode has also been moved to the top in order to be
available to all functions. On introspection it is revealed that it
should always have been placed at the top.
* Unindent logic in removeManifest
The logic is unnecessarily indented, leading to harder to follow logic.
While the current logic isn't that hard to understand, it's better if it
follows the same rhythm as installManifest, which errors checks the same
way with a !QFile::exists( manifestQPath ).
* Move manifestPath out of install/removeManifest
Moving the variable outside of the functions reduces code duplication
and increases the versatility of the functions.
Variable was renamed from manifestQPath to manifestPath, because the Q
does not add anything to the name except (maybe wrongfully in the
future) typing a variable that has been typed with 'auto'.
The default variable has been removed, as it did not improve readability
at all.
* Rename installManifest to handleManifestInstallLogic
The new name better reflects the overall purpose. installManifest should
be a function that only installs the manifest.
* Move manifest related things into manifest namespace
* Move manifestInstallation logic into functions
The logic becomes clearer when replaced by function names instead of
just blocks of error checking.
* Refactor installManifest into separate functions
The uses for the removeManifest and installManifest functions could be
boiled down to three separate use cases:
* If the application was passed the "-installmanifest" arg:
Reinstall the manifest and enable application autostart.
* If the application was passed thed "-removemanifest" arg:
Remove the manifest.
* If the application was not passed "-desktop" and also not passed
"-nomanifest" (basically every start):
Install the manifest.
Where reinstalling means to remove the manifest if installed and then
install it.
The relatively complex logic in installManifest hid this logic behind
several if statements and changing what the "alreadyInstalled" bool
signified.
The refactored functions should make everything easier to understand.
* Add variable for manifest name
* Rename variables in main to camelCase
This appears to be the style chosen for the majority of the project.
* Change cerr to be LOG(ERROR) and add logging for manifest completion
The cerr calls are not likely to be seen by anybody. Logging them as
ERROR will make sure they are available for diagnostics.
* Add comments for new manifest functions
* Reorder namespaces and functions in main
Namespaces and functions are now ordered as they appear in main.
* Add static_cast to height and width in func call in main
width and height are "qreal" (typedef for double). Having them
implicitly cast to int leads to warnings on some compilers, and can lead
to doubt whether they're supposed to be converted.
* Move organizationName and applicationName into overlaycontroller.h
The string have been moved to static constexpr autos inside the
overlaycontroller.
Them remaining static means that they are included in the final binary,
even if they don't need to be included as static.
They should probably also not be directly gotten, instead being offered
up as static getters, but it's better to do them all at the same time in
a specific commit for that.
* Add comment to installApplicationManifest
Describing that OpenVR must be initialized before the function is
called.
* Change QApplication var name and add comments
The previous QApplication was named "a". This is obviously not
sufficient for such a large part of the application, even if the class
is rather boilerplate heavy.
* Add comments to Qt portions of main
Some things have been left uncommented. That's because I don't know what
it does.
* Change documentation for command line arguments, logging config
The information under the Desktop Mode header could be rolled under the
"-desktop" command line arg.
Support for the command line args of Easylogging++ and custom logging
files have not been supported for the original developer. Both should
probably be dropped in favor of having the same logging behavior for all
users.
* Remove error causing property from QML MyStackViewPage and PlayspacePage
The properties were causing an error message to be spammed in the log
QML QQuickText: Detected anchors on an item that is managed by a layout.
This is undefined behavior; use Layout.alignment instead.
and
QML RowLayout: Detected anchors on an item that is managed by a layout.
This is undefined behavior; use Layout.alignment instead.
Removing them fixed the spam and didn't noticeably change the appearance
of the pages.
win32-clang-msvc has an error when compiling with /WX. The moc_predefs.h
file is not being built correctly, while without /WX it works. The
change should be reverted to enabling /WX for all msvc compilers when
the clang-msvc issue is fixed.
Update path to clang-format in appveyor file
Retry CI on Python script
Python script failed on CI due to run-clang-format.py: error: [Error 5]
Access is denied. Research says attempting again might solve the
problem.
The additional files packaged with the installer have been moved to
their own dir to separate them from the C++ source.
All licenses are now also included in the installer, as per the terms.
The version string in overlaycontroller.h has been updated to 2.8.0.
CLANG_FORMAT_LOC
The path to the bin dir for LLVM can be used for both clang-cl.exe and
clang-format.exe. Changing this will reduce the amount of environment
variables in play, leading to better long term quality.
Updating from easylogging++ 9.89 to 9.96.7 turns it from a header only
library to one requiring a source file. It has been added to sources
near the other third party includes in the QMAKE to signify that it
isn't a first party file.
Updating easylogging++ also means that clang-msvc can compile without
any warnings, hence turning on warnings as errors for that compiler as
well, to avoid any regressions.
logging.h previously had logic to enable debug logging even on release
builds. This logic was moved to the QMAKE file, hence the logging.h file
is no longer necessary.
Defines done outside of easylogging++.h does not work according to
official documentation
(https://github.com/zuhd-org/easyloggingpp/blob/master/README.md#configuration-macros)
"NOTE: All the macros can be defined in one of the following ways:
1. Define macros using -D option of compiler, for example in case of
g++ you will do g++ source.cpp ... -DELPP_SYSLOG -DELPP_THREAD_SAFE
... (recommended way)
2. Define macros inside "easylogging++.h" (defining macros in other
files won't work)"
Linking to this issue:
https://github.com/zuhd-org/easyloggingpp/issues/590#issuecomment-346753951
We are using the developer recommended way of defining them by command
line.
What is assumed to be the intention of enabling debug logs before
including easylogging++ never actually did anything, hence this define
is dropped. Newer versions of easylogging++ uses the macro
ELPP_DISABLE_DEBUG_LOGS to _disable_ debug logs. Leaving that macro
undefined satisfies the original intent.
This is the newest version that doesn't lead to compile errors. It has
also reduced the amount of "-Wexpansion-to-defined" errors on clang-msvc
by about 80 (from ~400 to 320).
A CONTRIBUTING file is considered special by Github, being automatically
linked to whenever an attempted pull request is made. Writing the
formatting changes along with how to submit bug reports might seem
fractured, but the special status of the CONTRIBUTING file means that
users are likely to find it. Additionally, links will be added to the
Readme and building_on_windows files.
The appveyor script can not use a single quoted double quotes string,
like '"C:\Spaces are trouble\"', hence the solution of defining an
environment variable.
appveyor.yml has been renamed to .appveyor.yml to keep consistent with
.travis.yml and to be default hidden.
Done in batch script because Powershell requires execution policy to be
changed to run scripts. Therefore batch files are considered to have a
lower barrier for entry.
username223