New features

Toggles for build type, optimization and vcrt type

Since the very beginning Meson has provided different project types to use, such as debug and minsize. There is also a plain type that adds nothing by default but instead makes it the user's responsibility to add everything by hand. This works but is a bit tedious.

In this release we have added new options to manually toggle e.g. optimization levels and debug info so those can be changed independently of other options. For example by default the debug buildtype has no optimization enabled at all. If you wish to use GCC's -Og instead, you could set it with the following command:

meson configure -Doptimization=g

Similarly we have added a toggle option to select the version of Visual Studio C runtime to use. By default it uses the debug runtime DLL debug builds and release DLL for release builds but this can be manually changed with the new base option b_vscrt.

Meson warns if two calls to configure_file() write to the same file

If two calls to configure_file() write to the same file Meson will print a WARNING: message during configuration. For example:

project('configure_file', 'cpp')

configure_file(
  input: 'a.in',
  output: 'out',
  command: ['./foo.sh']
)
configure_file(
  input: 'a.in',
  output: 'out',
  command: ['./foo.sh']
)

This will output:

The Meson build system
Version: 0.47.0.dev1
Source dir: /path/to/srctree
Build dir: /path/to/buildtree
Build type: native build
Project name: configure_file
Project version: undefined
Build machine cpu family: x86_64
Build machine cpu: x86_64
Configuring out with command
WARNING: Output file out for configure_file overwritten. First time written in line 3 now in line 8
Configuring out with command
Build targets in project: 0
Found ninja-1.8.2 at /usr/bin/ninja

New kwarg console for custom_target()

This keyword argument conflicts with capture, and is meant for commands that are resource-intensive and take a long time to finish. With the Ninja backend, setting this will add this target to Ninja's console pool, which has special properties such as not buffering stdout and serializing all targets in this pool.

The primary use-case for this is to be able to run external commands that take a long time to execute. Without setting this, the user does not receive any feedback about what the program is doing.

dependency(version:) now applies to all dependency types

Previously, version constraints were only enforced for dependencies found using the pkg-config dependency provider. These constraints now apply to dependencies found using any dependency provider.

Some combinations of dependency, host and method do not currently support discovery of the version. In these cases, the dependency will not be found if a version constraint is applied, otherwise the version() method for the dependency object will return 'unknown'.

(If discovering the version in one of these combinations is important to you, and a method exists to determine the version in that case, please file an issue with as much information as possible.)

python3 module is deprecated

A generic module python has been added in Meson 0.46.0 and has a superset of the features implemented by the previous python3 module.

In most cases, it is a simple matter of renaming:

py3mod = import('python3')
python = py3mod.find_python()

becomes

pymod = import('python')
python = pymod.find_installation()

Dictionary addition

Dictionaries can now be added, values from the second dictionary overrides values from the first

d1 = {'a' : 'b'}
d3 = d1 + {'a' : 'c'}
d3 += {'d' : 'e'}

Dist scripts

You can now specify scripts that are run as part of the dist target. An example usage would go like this:

project('foo', 'c')

# other stuff here

meson.add_dist_script('dist_cleanup.py')

Fatal warnings

A new command line option has been added: --fatal-meson-warnings. When enabled, any warning message printed by Meson will be fatal and raise an exception. It is intended to be used by developers and CIs to easily catch deprecation warnings, or any other potential issues.

Helper methods added for checking GNU style attributes: __attribute__(...)

A set of new helpers have been added to the C and C++ compiler objects for checking GNU style function attributes. These are not just simpler to use, they may be optimized to return fast on compilers that don't support these attributes. Currently this is true for MSVC.

cc = meson.get_compiler('c')
if cc.has_function_attribute('aligned')
   add_project_arguments('-DHAVE_ALIGNED', language : 'c')
endif

Would replace code like:

if cc.compiles('''into foo(void) __attribute__((aligned(32)))''')
   add_project_arguments('-DHAVE_ALIGNED', language : 'c')
endif

Additionally, a multi argument version has been added:

foreach s : cc.get_supported_function_attributes(['hidden', 'alias'])
   add_project_arguments('-DHAVE_@0@'.format(s.to_upper()), language : 'c')
endforeach

gnome.generate_gir() now optionally accepts multiple libraries

The GNOME module can now generate a single gir for multiple libraries, which is something g-ir-scanner supported, but had not been exposed yet.

gnome.generate_gir() will now accept multiple positional arguments, if none of these arguments are an Executable instance.

Hotdoc module

A new module has been written to ease generation of hotdoc based documentation. It supports complex use cases such as hotdoc subprojects (to create documentation portals) and makes it straight forward to leverage full capabilities of hotdoc.

Simple usage:

hotdoc = import('hotdoc')

hotdoc.generate_doc(
  'foobar',
  c_smart_index: true,
  project_version: '0.1',
  sitemap: 'sitemap.txt',
  index: 'index.md',
  c_sources: ['path/to/file.c'],
  languages: ['c'],
  install: true,
)

i18n.merge_file() now fully supports variable substitutions defined in custom_target()

Filename substitutions like @BASENAME@ and @PLAINNAME@ were previously accepted but the name of the build target wasn't altered leading to colliding target names when using the substitution twice. i18n.merge_file() now behaves as custom_target() in this regard.

Projects args can be set separately for cross and native builds (potentially breaking change)

It has been a longstanding bug (or let's call it a "delayed bug fix") that if you do this:

add_project_arguments('-DFOO', language : 'c')

Then the flag is used both in native and cross compilations. This is very confusing and almost never what you want. To fix this a new keyword native has been added to all functions that add arguments, namely add_global_arguments, add_global_link_arguments, add_project_arguments and add_project_link_arguments that behaves like the following:

# Added to native builds when compiling natively and to cross
# compilations when doing cross compiles.
add_project_arguments(...)

# Added only to native compilations, not used in cross compilations.
add_project_arguments(..., native : true)

# Added only to cross compilations, not used in native compilations.
add_project_arguments(..., native : false)

Also remember that cross compilation is a property of each target. There can be target that are compiled with the native compiler and some which are compiled with the cross compiler.

Unfortunately this change is backwards incompatible and may cause some projects to fail building. However this should be very rare in practice.

More flexible override_find_program().

It is now possible to pass an executable to override_find_program() if the overridden program is not used during configure.

This is particularly useful for fallback dependencies like Protobuf that also provide a tool like protoc.

shared_library() now supports setting dylib compatibility and current version

Now, by default shared_library() sets -compatibility_version and -current_version of a macOS dylib using the soversion.

This can be overridden by using the darwin_versions: kwarg to shared_library(). As usual, you can also pass this kwarg to library() or build_target() and it will be used in the appropriate circumstances.

Version comparison

dependency(version:) and other version constraints now handle versions containing non-numeric characters better, comparing versions using the rpmvercmp algorithm (as using the pkg-config autoconf macro PKG_CHECK_MODULES does).

This is a breaking change for exact comparison constraints which rely on the previous comparison behaviour of extending the compared versions with '0' elements, up to the same length of '.'-separated elements.

For example, a version of '0.11.0' would previously match a version constraint of '==0.11', but no longer does, being instead considered strictly greater.

Instead, use a version constraint which exactly compares with the precise version required, e.g. '==0.11.0'.

Keyword argument for GNU symbol visibility

Build targets got a new keyword, gnu_symbol_visibility that controls how symbols are exported from shared libraries. This is most commonly used to hide implementation symbols like this:

shared_library('mylib', ...
  gnu_symbol_visibility: 'hidden')

In this case only symbols explicitly marked as visible in the source files get exported.

Git wraps can now clone submodules automatically

To enable this, the following needs to be added to the .wrap file:

clone-recursive=true

subproject() function now supports the required: kwarg

This allows you to declare an optional subproject. You can now call found() on the return value of the subproject() call to see if the subproject is available before calling get_variable() to fetch information from it.

dependency() objects now support the .name() method

You can now fetch the name of the dependency that was searched like so:

glib_dep = dependency('glib-2.0')
...
message("dependency name is " + glib_dep.name())
# This outputs `dependency name is glib-2.0`

qt_dep = dependency('qt5')
...
message("dependency name is " + qt_dep.name())
# This outputs `dependency name is qt5`

decl_dep = declare_dependency()
...
message("dependency name is " + decl_dep.name())
# This outputs `dependency name is internal`