Reading back over this post, I realise I covered a fair bit of the why, but not much of the how. So I thought I’d revisit the topic and expand.
The first thing to realise about Meson is that unlike autotools, no executable part of the build system is distributed as part of a package. There is no configure script; but more importantly, no esoteric bootstrapping process to create the configure script. The second thing to realise is that separation of source & build directories is the norm, not optional, “advanced” behaviour.
Let’s take a look at a tiny C++ hello-world package. From the root, the layout would look something like this:
./meson.build ./src/meson.build ./src/hellow.cpp
(Strictly speaking you don’t need a src subdirectory, but as I said before, packaging software for use by others is an untaught art. In any real-world project, it won’t be long before the root folder grows things like a licence, a readme file, etc., and you want to start adding assets which aren’t source code.)
The top-level meson.build just defines the project, and includes the src subdirectory:
project('hellow', 'cpp', license: 'GPL3+', version: '1') subdir('src')
src/meson.build defines the executable we want to build:
executable('hellow', 'hellow.cpp', install: true)
src/hellow.cpp should be self-explanatory. :)
So if none of that is actually runnable, how do we produce our executable? Simple: invoke Meson to create a build directory, then invoke Ninja from that directory to build the project.
From inside the project root:
phil@hue meson-hellow $ meson setup . build The Meson build system Version: 0.47.2 Source dir: /home/phil/Projects/meson-hellow Build dir: /home/phil/Projects/meson-hellow/build Build type: native build Project name: hellow Project version: 1 Native C++ compiler: c++ (gcc 8.2.1 "c++ (GCC) 8.2.1 20181105 (Red Hat 8.2.1-5)") Build machine cpu family: x86_64 Build machine cpu: x86_64 Build targets in project: 1 Found ninja-1.8.2 at /usr/bin/ninja phil@hue meson-hellow $ cd build/ phil@hue build $ ninja [2/2] Linking target src/hellow. phil@hue build $ ./src/hellow Hello, world!
The build directory: configuration & options
The first line of the above is the actual Meson invocation: meson setup <source directory> <build directory>. This produces a build directory, from where you can inspect & alter the build configuration, set any project-specific compile-time options, build & install the project, and produce tarballs for distribution.
Modifying the build configuration after the initial set-up may seem counter-intuitive if, like me, you’re used to having to clean everything and re-run a configure script to change anything. Meson, however, automatically inserts enough magic into its Ninja rules to regenerate everything required on the fly. As far as I can tell, in theory, you only need to re-run meson setup if you want to change compiler, or generate something other than Ninja files for building (XCode & Visual Studio are supported; I haven’t tried them).
To list available options and their settings, just run meson configure from the build directory. To change them, use the -D argument; such as
meson configure -Dprefix=/tmp
The range of built-in options is quite impressive. Most importantly to me, with my distribution-friendly-developer hat on, the full set of installation paths seems to be present and correct, with sane defaults (bindir, libdir, datadir, etc., all relative to prefix as nature intended).
Anyone who’s tried to add a custom configure-time option to an autotools project will know that doing so is a mess. Separate macros for defining the option, and formatting its help string; a “type system” consisting of booleans and strings; the necessity of falling back on shell script to actually do anything meaningful based on the values of options.
With Meson, you can simply drop a meson_options.txt in the project root, define the options you want to add (including their types, descriptions, and default values), then use get_option() from your meson.build files to do things based on their values.
A common thing to want to do in C/C++ projects is define the values of custom preprocessor symbols at build time, based on the host system, build environment, or the values of user-specified options. Thankfully this, too, is simpler and less error-prone than with autotools, though the approach isn’t totally alien.
Within your meson.build files, you can call configuration_data() to create a configuration data object; this then has an array of helpful methods, including set() and set_quoted(). This configuration data object can then be given as an argument to configure_file(), to auto-generate a header file representing the configuration. Include that header file from your source - job done.
A note on imperative, declarative, and domain-specific languages
If you remember, in part 1, I said that sane people shouldn’t want their build system to use a full-on imperative programming language; that they should want it to be declarative. That may not sit well with all the talk of “creating objects” and “calling methods” in the previous paragraph. So what’s going on?
Well, Meson’s syntax seems to sit somewhere between purely declarative, and imperative. If you don’t need any project-specific options, or to modify your code’s behaviour based on the build environment, then you can get away with declaring the project and its artefacts (executables, shared objects, etc.) in what looks very much like a simple series of statements, bereft of logic. Once you start doing those things, you will encounter functions, objects, if statements, etc., but these are all part of a domain-specific language whose sole focus is defining build systems. It’s not quite declarative, but it’s intentionally minimalistic, and restrictive of anything outside its remit.
… and the importance of documenting them
It also has a very good manual. Here’s the official page on configuration, and the full reference for configuration_data(), the configuration data object, and configure_file().
Installation and distribution
Installation and distribution duties are performed through Ninja; hence, from a build directory.
phil@hue meson-hellow $ meson setup . build -Dprefix=/home/phil/foo # ... omitted for clarity ... phil@hue meson-hellow $ cd build/ phil@hue build $ ninja install [2/3] Installing files. Installing src/hellow to /home/phil/foo/bin phil@hue build $ find /home/phil/foo/ /home/phil/foo/ /home/phil/foo/bin /home/phil/foo/bin/hellow phil@hue build $ /home/phil/foo/bin/hellow Hello, world!
(I don’t know if I mentioned that you can set options directly from the meson setup command. Well, you can.)
phil@hue build $ ninja dist # ... shortened for clarity ... [0/1] Creating source packages Testing distribution package /home/phil/Projects/meson-hellow/build/meson-dist/hellow-1.tar.xz. [0/1] Installing files. Installing src/hellow to /tmp/tmpi2fmjgyp/usr/local/bin Distribution package /home/phil/Projects/meson-hellow/build/meson-dist/hellow-1.tar.xz tested. phil@hue build $ find meson-dist/ meson-dist/ meson-dist/hellow-1.tar.xz meson-dist/hellow-1.tar.xz.sha256sum
Nice. Note that not only has it produced a tarball, it’s also tested it, by automatically unpacking it, building & installing in a temporary prefix. It’s also given me a SHA-256 checksum as a little extra.
I also said previously that cross-compiling for Windows from Linux was trivial, then proceeded to present no evidence whatsoever. Well, here goes:
phil@hue meson-hellow $ meson setup . w64-build --cross-file /usr/share/mingw/toolchain-mingw64.meson The Meson build system Version: 0.47.2 Source dir: /home/phil/Projects/meson-hellow Build dir: /home/phil/Projects/meson-hellow/w64-build Build type: cross build Project name: hellow Project version: 1 Native C++ compiler: c++ (gcc 8.2.1 "c++ (GCC) 8.2.1 20181105 (Red Hat 8.2.1-5)") Cross C++ compiler: /usr/bin/x86_64-w64-mingw32-g++ (gcc 7.3.0) Host machine cpu family: x86_64 Host machine cpu: x86_64 Target machine cpu family: x86_64 Target machine cpu: x86_64 Build machine cpu family: x86_64 Build machine cpu: x86_64 Build targets in project: 1 Found ninja-1.8.2 at /usr/bin/ninja phil@hue meson-hellow $ cd w64-build/ phil@hue w64-build $ ninja [2/2] Linking target src/hellow.exe. phil@hue w64-build $ wine ./src/hellow.exe Hello, world!
Yep - just provide the --cross-file argument to meson setup, and away you go. Note the “Cross C++ compiler” line in the output, the compilation to hellow.exe, and the usage of Wine to run it.
In the interests of full disclosure: I’m using Fedora 28 at the moment. The MinGW-based compiler itself is from the mingw64-gcc-c++ package, and the Meson cross file from one of its dependencies, mingw64-filesystem. That said, it doesn’t look like defining your own cross files for other, more esoteric systems is rocket science (at least, not for the kind of people who are likely to attempt it); the file itself isn’t very long.
Thanks for reading. Have I convinced you yet?