How to create and share a custom generator with generator packages
There are several built-in generators, like cmake
, visual_studio
, xcode
… But what if your build system is not included or the
existing built-in ones doesn’t satisfy your needs? This how to will show you how to create a generator for
Premake build system.
Important
Check the reference of the Custom generator section to know the syntax and attributes available.
Creating a Premake generator
Create a folder with a new conanfile.py with the following contents:
$ mkdir conan-premake && cd conan-premake
from conans.model import Generator
from conans import ConanFile
class PremakeDeps(object):
def __init__(self, deps_cpp_info):
self.include_paths = ",\n".join('"%s"' % p.replace("\\", "/")
for p in deps_cpp_info.include_paths)
self.lib_paths = ",\n".join('"%s"' % p.replace("\\", "/")
for p in deps_cpp_info.lib_paths)
self.bin_paths = ",\n".join('"%s"' % p.replace("\\", "/")
for p in deps_cpp_info.bin_paths)
self.libs = ", ".join('"%s"' % p for p in deps_cpp_info.libs)
self.defines = ", ".join('"%s"' % p for p in deps_cpp_info.defines)
self.cppflags = ", ".join('"%s"' % p for p in deps_cpp_info.cppflags)
self.cflags = ", ".join('"%s"' % p for p in deps_cpp_info.cflags)
self.sharedlinkflags = ", ".join('"%s"' % p for p in deps_cpp_info.sharedlinkflags)
self.exelinkflags = ", ".join('"%s"' % p for p in deps_cpp_info.exelinkflags)
self.rootpath = "%s" % deps_cpp_info.rootpath.replace("\\", "/")
class premake(Generator):
@property
def filename(self):
return "conanpremake.lua"
@property
def content(self):
deps = PremakeDeps(self.deps_build_info)
template = ('conan_includedirs{dep} = {{{deps.include_paths}}}\n'
'conan_libdirs{dep} = {{{deps.lib_paths}}}\n'
'conan_bindirs{dep} = {{{deps.bin_paths}}}\n'
'conan_libs{dep} = {{{deps.libs}}}\n'
'conan_cppdefines{dep} = {{{deps.defines}}}\n'
'conan_cppflags{dep} = {{{deps.cppflags}}}\n'
'conan_cflags{dep} = {{{deps.cflags}}}\n'
'conan_sharedlinkflags{dep} = {{{deps.sharedlinkflags}}}\n'
'conan_exelinkflags{dep} = {{{deps.exelinkflags}}}\n')
sections = ["#!lua"]
all_flags = template.format(dep="", deps=deps)
sections.append(all_flags)
template_deps = template + 'conan_rootpath{dep} = "{deps.rootpath}"\n'
for dep_name, dep_cpp_info in self.deps_build_info.dependencies:
deps = PremakeDeps(dep_cpp_info)
dep_name = dep_name.replace("-", "_")
dep_flags = template_deps.format(dep="_" + dep_name, deps=deps)
sections.append(dep_flags)
return "\n".join(sections)
class MyPremakeGeneratorPackage(ConanFile):
name = "premakegen"
version = "0.1"
url = "https://github.com/memsharded/conan-premake"
license = "MIT"
This is a full working example. Note the PremakeDeps
class as a helper. The generator is creating Premake information for each
individual library separately, then also an aggregated information for all dependencies. This PremakeDeps
wraps a single item of such
information.
Note the name of the package will be premakegen/0.1@<user>/<channel> as that is the name given to it, while the generator name is
premake (the name of the class that inherits from Generator
). You can give the package any name you want, even the same as the
generator’s name if desired.
You export
the package recipe to the local cache, so it can be used by other projects as usual:
$ conan export . myuser/testing
Using the generator
Let’s create a test project that uses this generator. We will use a simple application that will use a “Hello World” library package as a requirement.
First, let’s create the “Hello World” library package:
$ mkdir conan-hello && cd conan-hello
$ conan new hello/0.1
$ conan create . myuser/testing
Now, let’s create a folder for the application that will use Premake as build system:
$ cd ..
$ mkdir premake-project && cd premake-project
Put the following files inside. Note the premakegen@0.1@myuser/testing
package reference in your conanfile.txt.
[requires]
hello/0.1@myuser/testing
premakegen@0.1@myuser/testing
[generators]
premake
#include "hello.h"
int main (void) {
hello();
}
-- premake4.lua
require 'conanpremake'
-- A solution contains projects, and defines the available configurations solution "MyApplication"
configurations { "Debug", "Release" }
includedirs { conan_includedirs }
libdirs { conan_libdirs }
links { conan_libs }
-- A project defines one build target
project "MyApplication"
kind "ConsoleApp"
language "C++"
files { "**.h", "**.cpp" }
configuration "Debug"
defines { "DEBUG" }
flags { "Symbols" }
configuration "Release"
defines { "NDEBUG" }
flags { "Optimize" }
Let’s install the requirements:
$ conan install . -s compiler=gcc -s compiler.version=4.9 -s compiler.libcxx=libstdc++ --build
This generates the premake4.lua file with the requirements information for building.
Now we are ready to build the project:
$ premake4 gmake
$ make (or mingw32-make if in windows-mingw)
$ ./MyApplication
Hello World Release!
Now everything works, so you might want to share your generator:
$ conan upload premakegen/0.1@myuser/testing
Tip
This is a regular Conan package, so you could create a test_package folder with a conanfile.py to test the generator as done in
the example above (invoke the Premake build in the build()
method).
Using template files for custom generators
If your generator has a lot of common, non-parameterized text, you might want to use files that contain the template. It is possible to do this as long as the template file is exported in the recipe. The following example uses a simple text file, but you could use other templating formats:
import os
from conans import ConanFile, load
from conans.model import Generator
class MyCustomGenerator(Generator):
@property
def filename(self):
return "customfile.gen"
@property
def content(self):
template = load(os.path.join(os.path.dirname(__file__), "mytemplate.txt"))
return template % "Hello"
class MyCustomGeneratorPackage(ConanFile):
name = "custom_generator"
version = "0.1"
exports = "mytemplate.txt"
Storing generators in the Conan local cache
Warning
This is an experimental feature subject to breaking changes in future releases.
In addition to distributing them using Conan packages, custom generators can be stored
in the generators folder in the Conan local cache (by default ~/.conan/generators
).
Generators stored in the local cache can be used in the same ways as the built-in generators, i.e. they can be referenced on the command line with conan install when using the --generator option, and do not require installing a package to use. Instead, these generators can be distributed using conan config install.
import json
from conans.model import Generator
# The generator name will be the literal class name (not the filename)
class custom_generator(Generator):
@property
def filename(self):
return "custom_generator_output.json"
@property
def content(self):
return json.dumps(self.deps_env_info.vars)
$ conan install <path_or_reference> --generator custom_generator
Note
Generators loaded from the local cache do not need to be accompanied by a recipe class. Additionally, more than one generator can be loaded from the same python module when loaded from the local cache.