Introduce CMake (and removing all other project-related code) #7270

TrueBrain · 2019-02-23T21:22:56Z

Stable release build
Nightly release build

This removes:

MSVC project files
Kdevelop generation files
configure / Makefile

It replaces them all with a single method: cmake. This should also remove any awk/vbs scripts we have around, meaning we will get a single method of doing stuff.

Currently tested with these 'IDEs':

VS2017 (yes, VS supports CMake)
VS2019 (yes, VS2019 supports CMake even better)
CMake, WSL Ubuntu 18.04

This is pretty much a Work In Progress, but I wanted to show others how it would look etc, so they can also test already a bit. Any feedback is welcome.

To test it out:

mkdir build
cd build
cmake ..
make

Or open the CMakeLists.txt in VS2017 / VS2019.

Things still to do:

(I am sure I am missing stuff .. but this is something to work on at least :D)

CMakeLists.txt

nielsmh · 2019-02-24T08:31:53Z

If CMake becomes the only (meta) build system, wouldn't it make sense to scrap the source.list file as well, and define that data via CMake syntax instead?

glx22 · 2019-02-24T21:01:39Z

For MinGW directmusic and xaudio2 needs to be detected. MinGW has directmusic headers, but doesn't have xaudio2.

TrueBrain · 2019-02-24T21:13:49Z

Added both @nielsmh and @glx22 points to the list at the top.

orudge · 2019-02-24T21:15:02Z

For MinGW directmusic and xaudio2 needs to be detected. MinGW has directmusic headers, but doesn't have xaudio2.

Just to clarify, if you manually copy over an xaudio2 header file then mingw should be happy with it - I did test that when I wrote it.

src/os/windows/ottdres.rc.in

glx22 · 2019-02-26T03:10:20Z

I kept the manifest in projects, but it should probably move somewhere else

glx22 · 2019-02-27T03:32:42Z

For now we rely on cmake default CMAKE_CONFIGURATION_TYPES and CMAKE_XXXXX_FLAGS_<CONFIG>, but it should be better to manage that by ourselves

src/os/windows/ottdres.rc.in

CMakeLists.txt

TrueBrain · 2019-02-27T20:38:10Z

For now we rely on cmake default CMAKE_CONFIGURATION_TYPES and CMAKE_XXXXX_FLAGS_, but it should be better to manage that by ourselves

I disagree. The closer we stay to CMake, the easier our life will be. Let's try to do this, and only add exceptions if we really need to (and mark them as such). Otherwise we will have a config.lib in no time :D

glx22 · 2019-02-27T22:43:30Z

Indeed something is wrong in the generated project file for lzo and lzma
AdditionalDependencies for Release:
D:\developpement\GitHub\vcpkg\installed\x64-windows-static\lib\libpng16.lib
D:\developpement\GitHub\vcpkg\installed\x64-windows-static\lib\zlib.lib
D:\developpement\GitHub\vcpkg\installed\x64-windows-static\debug\lib\lzma.lib
D:\developpement\GitHub\vcpkg\installed\x64-windows-static\debug\lib\lzo2.lib
D:\developpement\GitHub\vcpkg\installed\x64-windows-static\lib\freetype.lib
D:\developpement\GitHub\vcpkg\installed\x64-windows-static\lib\bz2.lib

AdditionalDependencies for Debug:
D:\developpement\GitHub\vcpkg\installed\x64-windows-static\debug\lib\libpng16d.lib
D:\developpement\GitHub\vcpkg\installed\x64-windows-static\debug\lib\zlibd.lib
D:\developpement\GitHub\vcpkg\installed\x64-windows-static\debug\lib\lzma.lib
D:\developpement\GitHub\vcpkg\installed\x64-windows-static\debug\lib\lzo2.lib
D:\developpement\GitHub\vcpkg\installed\x64-windows-static\debug\lib\freetyped.lib
D:\developpement\GitHub\vcpkg\installed\x64-windows-static\debug\lib\bz2d.lib

glx22 · 2019-03-01T01:43:07Z

hmm that's weird, OSX cmake step now pass, lzo is found, compilation fails because it doesn't find an lzo header

glx22 · 2019-03-01T22:46:30Z

Ok there's an issue with lzo detection, I can compile with MinGW by luck, as the detected

-- LZO_INCLUDE_DIRS:H:/msys64/mingw64/include/lzo

is covered by H:/msys64/mingw64/include from other libs.
But on OSX

-- LZO_INCLUDE_DIRS: /usr/local/Cellar/lzo/2.10/include/lzo

is alone

glx22 · 2019-03-03T00:54:09Z

Stupid me failing in the commit message.

BTW I think test task for MacOS should probably be "make -C build test".

cmake/AutodetectSSE.cmake

glx22 · 2019-03-12T23:20:59Z

regression works with MinGW without converting the exe, but it fails with many

6: -  SetCommandDelay: (null : 0x00000000)
6: +  SetCommandDelay: (null : 0x0000000000000000)'

I guess the lovely regex was needed :)

TrueBrain · 2019-03-13T18:21:04Z

Yeah, but I refused to do that unreadable regex; so now I am just adding them one case at the time :D

DerDakon · 2019-07-19T20:00:37Z

.gitignore

 bundle/*
 bundles/*
 docs/aidocs/*
 docs/gamedocs/*
 docs/source/*
-.kdev4
-.kdev4/*
-*.kdev4


I would keep at least *.kdev4, this would still be relevant when importing the CMake project into KDevelop.

That can be handle in the local gitignore. I see no real need to put it globally.

DerDakon · 2019-07-19T20:00:37Z

CPackProperties.cmake.in

@@ -0,0 +1,12 @@
+# Make the current version available to CPack
+set(CPACK_PACKAGE_VERSION "@REV_VERSION@")


It should be enough if you call project(OpenTTD VERSION 5.6.7) at the top level, that would make the variable OpenTTD_VERSION available, also to CPack.

The REV_VERSION is based on the git tag(project version) or the git commit and can change between builds.
This is not pretty but necessary to make CPack behave correctly.

Please not that this will usually not change between builds: CMake will not notice if your git status has changed, so it will not rerun CMake, so it will not rerun your execute_process().

DerDakon · 2019-07-19T20:00:37Z

cmake/AutodetectEditbin.cmake

+# autodetect_editbin()
+#
+function(autodetect_editbin)
+    if (DEFINED EDITBIN_FOUND)


You could just check on EDITBIN_EXECUTABLE here, CMake will automatically put that into cache in find_program(). And apart from the messages you do not need to guard this, if the variable is already set find_program() will just do nothing.

We could refactor this entire function into very small and simple find module. Then one could use the more CMakeish find_package(EDITBIN).
You can have a look at the FindCCACHE on my fork for a minimal example.
This also keeps the error messages consistent with other find modules.

DerDakon · 2019-07-19T20:00:37Z

cmake/AutodetectEditbin.cmake

+    message(STATUS "Detecting editbin - found")
+
+    # Make sure these values are cached properly
+    set(EDITBIN_FOUND YES CACHE INTERNAL "")


If you want to have this variable after all simply not cache it: the find_program() call will do nothing, EDITBIN_EXECUTABLE is set, so this variable get's set to YES anyway. No need to clutter the cache with it.

DerDakon · 2019-07-19T20:00:37Z

cmake/AutodetectLibrary.cmake

+
+    # Check if any of the argument is ENCOURAGED
+    list(FIND args ENCOURAGED ENCOURAGED)
+    if (ENCOURAGED LESS 0)


if("ENCOURAGED" IN_LIST args)

DerDakon · 2019-07-19T20:00:37Z

cmake/Options.cmake

+        set(DEFAULT_SHARED_DIR "(not set)")
+        set(DEFAULT_GLOBAL_DIR "${CMAKE_INSTALL_PREFIX}/share/games/openttd")
+    else ()
+        message(FATAL_ERROR "Unknown OS found; please consider creating a Pull Request to add support for this OS.")


Again, just leave it as is. Let the *BSD folks do their experiment…

DerDakon · 2019-07-19T20:00:37Z

cmake/SourceList.cmake

+        endif (NOT (${PARAM_CONDITION}))
+    endif (PARAM_CONDITION)
+
+    foreach(FILE ${PARAM_FILES})


DerDakon · 2019-07-19T20:00:37Z

cmake/SourceList.cmake

+    set(PARAM_FILES "${PARAM_UNPARSED_ARGUMENTS}")
+
+    if (PARAM_CONDITION)
+        if (NOT (${PARAM_CONDITION}))


Why this? I don't think you can ever reach the inner return() as an empty variable is just false.

DerDakon · 2019-07-19T20:00:37Z

cmake/SourceList.cmake

+
+    get_property(SOURCE_PROPERTIES GLOBAL PROPERTY source_properties)
+
+    foreach(FILE ${PARAM_FILES})


DerDakon · 2019-07-19T20:00:37Z

CMakeLists.txt

+
+if (CMAKE_SOURCE_DIR STREQUAL CMAKE_BINARY_DIR)
+    message(FATAL_ERROR "In-source builds not allowed. Please run \"cmake ..\" from the bin directory")
+endif (CMAKE_SOURCE_DIR STREQUAL CMAKE_BINARY_DIR)


I probably have said this before, but just remove the arguments to all end*() calls. It makes code much easier to maintain.

fsimonis · 2019-07-22T10:31:08Z

@DerDakon I have the feeling that we are talking past each other.
I am talking about using the precaches solely for explicitly en/disabling features. This matter is orthogonal to how those features are detected. I totally agree that using the precache to influence library detection is a very very bad idea.

Explicit dependencies vs autodetection

My actual concern is that we should use explicit bindings options for dependencies. Autodetecting every library makes life more difficult for packagers and developer.

Here my reasoning:

With hard autodetection, CMake will always configure the version with the most detected libraries.
CMake will configure successfully even if an installed library was not found. A developer thus always has to check the logs to validate that all available libraries were detected as expected.
As a developer you may want to debug a configuration with a specific feature-set to reproduce or locate a bug. Hard autodetection makes this impossible.
In CI you may want to run tests on OpenTTD with a fixed feature set, such as only a single video and audio driver enabled or even multiple configuration matrices.

I propose to do the following:

Add a boolean option for every external library.
Explain why we need it. (@DerDakon Thanks, I was not aware of this module!)
Always fail if the library was not found.

Example for zlib:

# Options
option(OPENTTD_ZLIB "Enable ZLib" ON`)
add_feature_info(ZLib OPENTTD_ZLIB "Required for (de)compressing of old (0.3.0-1.0.5) savegames, content downloads, and heightmaps.")

# ...

# Package discovery
if(OPENTTD_ZLIB)
  find_package(ZLIB REQUIRED)
endif()

# ...

Running CMake without a preset of features will likely fail during the first configuration attempt, but then provide a full list of options to ccmake/cmake-gui etc.

Feature set using pre-cache population

Example: lets consider the different compression libraries used in OpenTTD. Then the options and the defaults for those features could look as follows:

Library	CMake Option	Default
zlib	`OPENTTD_ZLIB`	`ON`
LIBLZO2	`OPENTTD_LIBLZO2`	`ON`
libzlma	`OPENTTD_LIBZLMA`	`ON`

So for Ubuntu 18.04 we know that all libraries are available, thus we could explicitly enable them as such:

# linux-ubuntu-1804.cmake
set(OPENTTD_ZLIB ON CACHE BOOLEAN "")
set(OPENTTD_LIBLZO2 ON CACHE BOOLEAN "")
set(OPENTTD_LIBZLMA ON CACHE BOOLEAN "")

On some other system, liblzio2 may not be available, thus we could explicitly disable it:

# some-other-system.cmake
set(OPENTTD_ZLIB ON CACHE BOOLEAN "")
set(OPENTTD_LIBLZO2 OFF CACHE BOOLEAN "")
set(OPENTTD_LIBZLMA ON CACHE BOOLEAN "")

This way, the CI configuration for a specific distro boils down to: cmake -C $SRC/.../linux-ubuntu-1804.cmake $SRC.

DerDakon · 2019-07-22T19:48:31Z

I would use options for feature sets, and based on that require on or another library. Even if Ubuntu has Lzo2, it doesn't mean that the user has it installed. What you could do is to give the user a list of packages for a given distro that should ideally be installed.

fsimonis · 2019-08-10T12:11:44Z

So a possible feature-set based on the README would look like this:

Feature option	Required libraries
`OPENTTD_BACKEND_SDL`	libsdl
`OPENTTD_BACKEND_ALLEGRO`	liballegro
`OPENTTD_SAVEGAME_ANCIENT`	liblzo2
`OPENTTD_SAVEGAME_OLD`	zlib
`OPENTTD_SAVEGAME`	liblzma
`OPENTTD_CONTENT_DOWNLOAD`	zlib
`OPENTTD_HEIGHTMAP`	zlib, libpng
`OPENTTD_SCREENSHOT`	libpng
`OPENTTD_FONTS`	libfreetype, libfontconfig
`OPENTTD_RIGHT_TO_LEFT`	libico

@DerDakon Is that what you had in mind?

DerDakon · 2019-08-10T13:36:26Z

Basically. And the code for that could look like this:

unset(zlib_required)
option(OPENTTD_SAVEGAME_OLD "foo" Off)
option(OPENTTD_HEIGHTMAP "bar" On)

if (OPENTTD_HEIGHTMAP OR OPENTTD_SAVEGAME_OLD OR …)
  set(zlib_required "REQUIRED")
endif()

find_package(ZLIB ${zlib_required})

glx22 · 2019-09-20T13:25:25Z

I think I properly handled integration of 2d27e8e and c655f89


        Remove: all existing project files

This prepares for the switch to CMake, which takes over all current exisiting forms of project files.


        Codechange: move regression outside of bin and make it work via CMake…

… script The tst_stationlist savegame had to be changed to start the correct AI. In the old setup, all regression AIs had the same name, which made it impossible to run both regressions in parallel. With the new setup this is possible. Although all files are available to run the regression, it won't really work till CMake is introduced (which will happen in a few commits from here)


        Codechange: rework how grf and ob[msg] are generated

For grfs, it now uses CMake scripts to do its job, and both grf files are split into their own folder to make more clear what is going on. Additionally, it no longer builds in-source (although the resulting grf is copied back in the source folder). For ob[msg] it now uses CMake scripts to generate the translation files; the result is no longer stored in-source (but in the build folder). Although all files are available to create the GRFs and basesets, it won't really work till CMake is introduced (which will happen in a few commits from here)


        Add: introduce CMake for project management

CMake works on all our supported platforms, like MSVC, Mingw, GCC, Clang, and many more. It allows for a single way of doing things, so no longer we need shell scripts and vbs scripts to work on all our supported platforms. Additionally, CMake allows to generate project files for like MSVC, KDevelop, etc. This heavily reduces the lines of code we need to support multiple platforms from a project perspective. Addtiionally, this heavily improves our detection of libraries, etc.


        Add: create bundles via CPack

CPack works closely together with CMake to do the right thing in terms of bundling (called 'package'). This generates all the packaging we need, and some more.


        Fix: [AzurePipelines] switch the CI / CD to CMake too

This also means dropping Debian/jessie, as it has a CMake that is too old (3.0), with no real path to upgrade.


        Codechange: remove #ifdef from .cpp files to exclude features

With CMake, these files are simply not compiled to start with.


        Fix: update COMPILING.md stating what changed with CMake


        Codechange: force static linking for MinGW builds

TrueBrain force-pushed the TrueBrain:cmake branch from 2a97e01 to f6cd1ad Feb 23, 2019

glx22 reviewed Feb 24, 2019

View changes

CMakeLists.txt Outdated Show resolved Hide resolved

TrueBrain force-pushed the TrueBrain:cmake branch from cfff2fc to a891bd6 Feb 24, 2019

TrueBrain reviewed Feb 24, 2019

View changes

src/os/windows/ottdres.rc.in Outdated Show resolved Hide resolved

TrueBrain reviewed Feb 27, 2019

View changes

src/os/windows/ottdres.rc.in Outdated Show resolved Hide resolved

TrueBrain reviewed Feb 27, 2019

View changes

CMakeLists.txt Outdated Show resolved Hide resolved

TrueBrain force-pushed the TrueBrain:cmake branch 2 times, most recently from 6164e63 to a25c911 Mar 2, 2019

TrueBrain force-pushed the TrueBrain:cmake branch 3 times, most recently from e5b7c05 to d71ac0a Mar 10, 2019

nielsmh reviewed Mar 11, 2019

View changes

cmake/AutodetectSSE.cmake Outdated Show resolved Hide resolved

TrueBrain force-pushed the TrueBrain:cmake branch 5 times, most recently from 823f76d to 027a5b3 Mar 13, 2019

DerDakon suggested changes Jul 19, 2019

View changes

nielsmh mentioned this pull request Aug 29, 2019

Fix: Allow git worktree in version detection; drop svn, hg. #7707

Merged

glx22 force-pushed the TrueBrain:cmake branch from 995bf50 to 6f09f68 Sep 20, 2019

glx22 force-pushed the TrueBrain:cmake branch from 6f09f68 to 6575eeb Nov 3, 2019

glx22 force-pushed the TrueBrain:cmake branch from 6575eeb to 80db1be Nov 16, 2019

LordAro added this to the 1.11.0 milestone Nov 23, 2019

sdcloudt mentioned this pull request Dec 16, 2019

Unittests #7796

Closed

glx22 force-pushed the TrueBrain:cmake branch from 80db1be to 1f5ecea Dec 22, 2019

glx22 force-pushed the TrueBrain:cmake branch from 1f5ecea to 83aaf98 Jan 19, 2020

glx22 mentioned this pull request Jan 21, 2020

Error when configuring OpenTTD with OpenSSL1.1 installed with homebrew #7951

Closed

glx22 force-pushed the TrueBrain:cmake branch from 83aaf98 to 36bb68f Feb 1, 2020

glx22 force-pushed the TrueBrain:cmake branch from 36bb68f to 758b8f1 Feb 8, 2020

LordAro mentioned this pull request Feb 9, 2020

Codechange: Don't fail configure if SDL2 is not detected #7990

Merged

glx22 force-pushed the TrueBrain:cmake branch from 758b8f1 to 68def2c Feb 10, 2020

glx22 mentioned this pull request Feb 11, 2020

OpenBSD: fix endianness detection #7996

Merged

TrueBrain and others added 9 commits Apr 7, 2019

Remove: all existing project files

e342415

This prepares for the switch to CMake, which takes over all current exisiting forms of project files.

Add: create bundles via CPack

ff9ab1a

CPack works closely together with CMake to do the right thing in terms of bundling (called 'package'). This generates all the packaging we need, and some more.

Fix: [AzurePipelines] switch the CI / CD to CMake too

8d10929

This also means dropping Debian/jessie, as it has a CMake that is too old (3.0), with no real path to upgrade.

Codechange: remove #ifdef from .cpp files to exclude features

e2efcf5

With CMake, these files are simply not compiled to start with.

Fix: update COMPILING.md stating what changed with CMake

fb75e69

Codechange: force static linking for MinGW builds

Loading status checks…

56c7bec

glx22 force-pushed the TrueBrain:cmake branch from 68def2c to 56c7bec Feb 23, 2020

		@@ -0,0 +1,12 @@
		# Make the current version available to CPack
		set(CPACK_PACKAGE_VERSION "@REV_VERSION@")


		get_property(SOURCE_PROPERTIES GLOBAL PROPERTY source_properties)

		foreach(FILE ${PARAM_FILES})

OpenTTD / OpenTTD

Sponsor OpenTTD/OpenTTD

Join GitHub today

Introduce CMake (and removing all other project-related code) #7270

Introduce CMake (and removing all other project-related code) #7270

Conversation

TrueBrain commented Feb 23, 2019 • edited

This comment has been minimized.

nielsmh commented Feb 24, 2019

This comment has been minimized.

glx22 commented Feb 24, 2019

This comment has been minimized.

TrueBrain commented Feb 24, 2019

This comment has been minimized.

orudge commented Feb 24, 2019

This comment has been minimized.

glx22 commented Feb 26, 2019

This comment has been minimized.

glx22 commented Feb 27, 2019 • edited

This comment has been minimized.

TrueBrain commented Feb 27, 2019

This comment has been minimized.

glx22 commented Feb 27, 2019

This comment has been minimized.

glx22 commented Mar 1, 2019

This comment has been minimized.

glx22 commented Mar 1, 2019

This comment has been minimized.

glx22 commented Mar 3, 2019

This comment has been minimized.

glx22 commented Mar 12, 2019

This comment has been minimized.

TrueBrain commented Mar 13, 2019

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

DerDakon Jul 19, 2019 • edited

This comment has been minimized.

fsimonis commented Jul 22, 2019

Explicit dependencies vs autodetection

Feature set using pre-cache population

This comment has been minimized.

DerDakon commented Jul 22, 2019

This comment has been minimized.

fsimonis commented Aug 10, 2019 • edited

This comment has been minimized.

DerDakon commented Aug 10, 2019

This comment has been minimized.

glx22 commented Sep 20, 2019

TrueBrain commented Feb 23, 2019 •

edited

glx22 commented Feb 27, 2019 •

edited

DerDakon Jul 19, 2019 •

edited

fsimonis commented Aug 10, 2019 •

edited