INSTALL target #745
INSTALL target #745
Conversation
Hey, this makes the whole ScriptsSystem thing nearly instant! Woohoo!
Hoikas
force-pushed
the
install-target
branch
from
d93e4d2 to
7a7e4d9
Compare
I think vcpkg may be inserting generator expressions into our junk. Kewl.
cmake/DeployQt5.cmake
Outdated
| # Copy Qt DLLs to the build directory so the stupid application can be run in Visual Studio | ||
| add_custom_command( | ||
| TARGET ${_pdqt_TARGET} POST_BUILD | ||
| COMMAND ${CMAKE_COMMAND} -E remove_directory "$<TARGET_FILE_DIR:${_pdqt_TARGET}>/qt" |
There was a problem hiding this comment.
If this gets run more than once (which it does), each target will re-populate the deployment directory. Our Qt apps all use the same Qt libs, so the end result is probably still fine, but this still does more work than necessary.
Perhaps instead we can make a list of executables that need to be considered for windeployqt, and then a single target that runs it on all of them at the same time (windeployqt can take multiple executables on the command line)...
There was a problem hiding this comment.
I was initially considering doing that by doing cmake's undocumented function overloading... But vcpkg already does that for add_executable(), which means we'd introduce an inifite loop. Maybe we could have a plasma_add_qt_executable that handles the book-keeping for us? Appending to a list in the CMakeLists.txt or having a hardcoded list seems kind of dirty to me.
There was a problem hiding this comment.
I don't know why you need to overload anything... You could just switch to the normal install command for the Qt executables, and then run the deployment from the Tools directory's CMakeLists.txt, since that file already has knowledge of the full set of Qt executables to build...
There was a problem hiding this comment.
I'm not certain how to obtain the list of those executables aside from hardocding a list of them (yuck) or injecting logic to gain it.
There was a problem hiding this comment.
Another option, which still duplicates work but at least doesn't potentially delete needed files, would be to add a target that only cleans the qt staging directory once, and then still have each Qt executable run its own deployment...
There was a problem hiding this comment.
One thing to note is that I've removed the default specification of CMAKE_RUNTIME_DIRECTORY, so, with Visual Studio, each executable is now building in its own directory and has its own qt staging directory. Previously, they were all going into <builddir>/bin/<config> So, if you're debugging one of these directly from the IDE, we might not want to clear its staging directory--you'd potentially be blasting the dependencies away before a debug session. I'm not sure what the behavior is with single config generators eg makefiles. (That comment about debugging from the IDE is irrelevant because we're copying Qt into a staging directory, not into the binary directory itself. Sigh.)
| 5. Copy the **resource.dat** file from `<build_dir>\bin` to the *MOUL-OS* folder, | ||
| or from http://www.guildofwriters.org/tools/resource.dat if you did not build | ||
| your own. | ||
| 5. Download and copy http://www.guildofwriters.org/tools/resource.dat into |
There was a problem hiding this comment.
Why remove the instructions for using the locally-built copy?
This new INSTALL feature automatically copies it to the correct place, so this step is entirely optional now.
There was a problem hiding this comment.
resource.dat will only be generated if you manually pip install -r requirements.txt or otherwise have the correct packages preinstalled. I figure that's a problem for another PR 😄
|
💥 |
This does a lot of things. On a high level, it changes the resource.dat file generation such that it properly gets marked as a GENERATED dependency of plClient. Further, we correctly chain the vector graphics and scripts as dependencies of resource.dat such that resource.dat will only be regenerated by modifying these files. H-uru#745 unintentionally broke this by removing the default definition of `CMAKE_RUNTIME_OUTPUT_DIRECTORY` -- CI didn't detect it because it it defines that variable in the cache. More sneakily, we now see the beginnings of more automated dependency management in that our CMake can now install the python packages for building resource.dat. For now, it is OFF by default - it seems rude to unilaterally install packages in a system Python install. Future work to enable a submodule based vcpkg toolchain would use this facility however.
This adds an INSTALL target that is supposed to install all of the files needed for the client and tools into an organized structure. It doesn't match the GNU standard of bin, lib, etc. because we are a game, not a system library or command line executable.
This is still considered a WIP because not all DLLs are copied into the installed tree on Windows. We currently runwindeployqton all GUI tools, and I would like to do something similar for the non-Qt DLLs. Vcpkg supportsVCPKG_APPLOCAL_DEPS, which copies the DLLs into the binary directory. The installed variant of this is still experimental though and seems to error on my bare-bonesinstallcmake commands. Do we want to try to work with vcpkg's experimental feature or roll that functionality ourselves? I realize, of course, that doing such on unix-like platforms is pointless... the goal I'm working toward here is to create good distributions for Windows.The latest vcpkg master has a correctly functioning
X_VCPKG_APPLOCAL_DEPS_INSTALL, so any DLLs needed by vcpkg are correctly copied into the install prefix. Further, our cmake now runs windeployqt on all GUI tools... because not everyone wants to build Qt5 from source using vcpkg .This obsoletes the clunky "ScriptsSystem" target - now just
make installorcmake --build . --target INSTALLto get that done.Depends on #740.