CMake generator.

CMake generator.

Designed to create projects for IDEs like kdevelop, qtcreator, code::blocks, or eclipse cdt.

Generated CMakeLists.txt can also be used to build (currently limited to *nix).

Requires CMake 2.8.8 or later. With patch set 9 (which builds CMake 2.8.8 on the bot) passes tests on the bots.

R=thakis@chromium.org

Committed: https://code.google.com/p/gyp/source/detail?r=1796

[bungeman-chromium, 2013-02-20 16:24:13]

This is a continuation of https://codereview.appspot.com/5996048/ .

[bungeman-chromium, 2013-02-20 20:43:09]

Fixed some pylint.

[bungeman-chromium, 2013-02-28 04:24:25]

Patch Set 3 puts inputs common to all rule commands into a variable and
references that variable instead of listing all of the common inputs in each
build rule. This trims ~1.5MB from the CMakeLists.txt when generating for
chromium, as well as making the generated CMakeLists.txt more clear.

[bungeman-chromium, 2013-02-28 19:34:53]

Patch Set 4 keeps all shared objects out of --start-group --end-group. This
prevents more places where CMake might repeat --start-group without also
repeating --end-group. It also makes more sense, and should be closer to what
other generators do.

[bungeman-chromium, 2013-02-28 23:43:12]

Patch Set 5 handles toplevel_dir like ninja. This gives relative paths when the
gyp file names resolve to absolute paths. With this change it is possible to use
this generator with gyp_chromium without much fuss.

[bungeman-chromium, 2013-03-04 19:07:24]

With Patch Set 6, inputs to Actions, Rules, and Copies are specified as
informative sources (sources to add_custom_target). These sources are not built
and are mostly ignored, but specifying these sources allows IDEs to see them.

In order to see these sources in KDevelop, it is necessary to use 4.5+.

[bungeman-chromium, 2013-03-06 22:38:05]

Just wanted to point out that after Patch Set 5 this generator is now much
easier to use. Because it is mirroring the ninja generator in layout, setting
'GYP_GENERATORS=ninja,cmake' works and has very little overhead to additionally
generate the CMakeLists.txt file.

[bungeman-chromium, 2013-03-18 16:04:07]

With Patch set 7, 'GYP_GENERATORS=ninja,cmake' works and places the
CMakeLists.txt in the correct place no matter the cwd.

It also adds all non-compilable sources to the sources list and marks them as
headers (this seems to be the blessed way for CMake to mark a file as
non-compilable). This means that all sources can be seen in an editor, but CMake
won't try to compile things we don't want compiled. The downside (upside?) to
this is that it means CMake is doing more linting, see
https://chromiumcodereview.appspot.com/12745003/#msg6 for something caught by
this change.

[Evan Martin, 2013-03-18 16:47:11]

I'm really curious how gyp -> ninja differs from gyp -> cmake -> ninja.

[bungeman-chromium, 2013-03-18 17:03:35]

On 2013/03/18 16:47:11, Evan Martin wrote:
> I'm really curious how gyp -> ninja differs from gyp -> cmake -> ninja.

Well, you can try it out :-) Note that with PS7 you'll need to comment out
'cloud_print/service/service.gyp' line like
''<(SHARED_INTERMEDIATE_DIR)/cloud_print/cloud_print_service_exe_version.rc','
on Linux if testing with Chromium. I've been building Chromium by setting
'GYP_GENERATORS=ninja,cmake' and then creating an 'out/ReleaseCMake' directory,
cd-ing to that and running '~/bin/cmake/cmake -G Ninja ../Release' so that the
CMake build is separate. Note that I have my own build of cmake 2.8.8 since my
system cmake is too old (2.8.7, arg).

Since this currently just creates a single monolithic CMakeLists.txt, with
Chromium one ends up with a CMakeLists.txt ~11.5MB and a single monolithic
build.ninja file, which is ~150MB. Generating the ninja files from CMake takes
like five minutes, but after that ninja builds like ninja. The ninja build time
seems comparable to the gyp ninja build (though I haven't really measured).

[bungeman-chromium, 2013-04-01 21:47:13]

Patch Set 8 is a small change to write the include directories out to a variable
and append that to the target's include directories. Doing things this way makes
the resulting CMakeLists.txt a bit easier to read, and shaves ~400KB off it size
when generating for Chromium.

[bungeman-chromium, 2013-04-03 23:08:45]

Patch Set 9 adds the CMake generator to the tests by pulling a shallow clone of
the v2.8.8 tag of the CMake repository, bootstrapping it, and using it (on an as
needed basis, similar to Android). Passes tests on the trybot.

[bungeman-chromium, 2013-04-18 18:44:08]

Updated to apply cleanly against current.

[bungeman-chromium, 2013-11-18 01:35:37]

This is now (at patch set 15) once again passing tests, has a few users, is
capable of building real projects, and adds 1337 lines.

https://codereview.chromium.org/12314014/diff/251001/gyptest.py
File gyptest.py (right):

https://codereview.chromium.org/12314014/diff/251001/gyptest.py#newcode179
gyptest.py:179: os.environ['PATH'] = extra_path + os.pathsep +
os.environ['PATH']
Not sure about how to handle this. I need my newer CMake to come first; it
appears the bots now have an older CMake installed.

[Nico, 2013-11-19 16:35:28]

Thanks for the ping!

Basically lgtm, a few stylistic issues below (consistent function naming, and
s/#(\w)/# \1/)

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
File pylib/gyp/generator/cmake.py (right):

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:1: # Copyright (c) 2012 Google Inc. All rights
reserved.
2013

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:74: return a[len(prefix):] if a.startswith(prefix)
else a
(nit: 2 newlines between toplevel things)

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:103: #TODO: do we need to check base_path for
absolute variables as well?
space after #

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:119: def ensure_directory_exists(path):
EnsureDirectoryExists? Or rename above methods to foo_style?

(also below)

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:285: #add_custom_command
space (also below)

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:291: for directory in dirs:
I think sets don't have deterministic iteration order (not sure if it matters)

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:377: for output_count, out in enumerate(outputs):
output_index?

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:579: class CMakeNamer:
(object)

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:682: target_type = 'none'
Do other generators support this?

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:696: #CMake is opposed to setting linker
directories as dangerous;
doesn't parse

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:1014: 
(nit: newline)

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:1061: #XXX: Undocumented and unsupported (the CMake
Java generator depends on it).
"TODO"

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:1064: #This should be irrelevent, but some targets
extract .o files from .a
irrelevant

https://codereview.chromium.org/12314014/diff/251001/test/builddir/gyptest-al...
File test/builddir/gyptest-all.py (right):

https://codereview.chromium.org/12314014/diff/251001/test/builddir/gyptest-al...
test/builddir/gyptest-all.py:26: # Android, Ninja, and CMake do not support
setting the build directory.
Ninja has -Goutput_dir, but not sure what "build directory" means in this file.

https://codereview.chromium.org/12314014/diff/251001/test/defines/gyptest-def...
File test/defines/gyptest-define-override.py (right):

https://codereview.chromium.org/12314014/diff/251001/test/defines/gyptest-def...
test/defines/gyptest-define-override.py:16: #CMake loudly warns about passing
'#' to the compiler and drops the define.
nit: space after #

https://codereview.chromium.org/12314014/diff/251001/test/lib/TestGyp.py
File test/lib/TestGyp.py (right):

https://codereview.chromium.org/12314014/diff/251001/test/lib/TestGyp.py#newc...
test/lib/TestGyp.py:540: Subclass for testing the GYP CMake generator.
", using cmake's ninja backend"?

[bungeman-chromium, 2013-11-20 20:54:51]

Addressed comments.

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
File pylib/gyp/generator/cmake.py (right):

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:1: # Copyright (c) 2012 Google Inc. All rights
reserved.
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> 2013

Done. Sigh... it has been a long time since I started working on this.

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:74: return a[len(prefix):] if a.startswith(prefix)
else a
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> (nit: 2 newlines between toplevel things)

Done.

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:103: #TODO: do we need to check base_path for
absolute variables as well?
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> space after #

Done. Here and many other places.

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:119: def ensure_directory_exists(path):
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> EnsureDirectoryExists? Or rename above methods to foo_style?
> 
> (also below)

Done. Yes, why are these different? It must be what frame of mind I was in at
the time. It looks like the rest of gyp is using CamelCase for top level defs,
so I'll go with that.

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:285: #add_custom_command
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> space (also below)

Done.

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:291: for directory in dirs:
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> I think sets don't have deterministic iteration order (not sure if it matters)

Don't care about the order here, just want to ensure that all of the required
directories exist before we run the action.

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:377: for output_count, out in enumerate(outputs):
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> output_index?

Done.

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:579: class CMakeNamer:
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> (object)

Done.

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:682: target_type = 'none'
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> Do other generators support this?

No, I don't think the other generators do this.  I've added a comment as to why
this is here. When generating build files from the CMakeLists.txt, CMake says
you did something wrong and refuses to generate build files if you claim to be
creating an executable but provide no sources. This actually happens in Chromium
with cloud_print_service at the moment. CMake is actually right; gyp-ing to
ninja generates the ninja build files for cloud_print_service on Linux, but the
target cannot actually be built. In theory someone might weasel an entry point
into a lib and somehow manage to link a working executable from just libs, but
this seems like a somewhat odd thing to do.

It might be possible to fake it out and provide an empty source (if one really
wants this to work), but this hasn't come up yet.

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:696: #CMake is opposed to setting linker
directories as dangerous;
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> doesn't parse

The tyranny of short lines. Updated so that it should be easier to follow.

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:1014: 
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> (nit: newline)

Done.

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:1061: #XXX: Undocumented and unsupported (the CMake
Java generator depends on it).
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> "TODO"

Done.

https://codereview.chromium.org/12314014/diff/251001/pylib/gyp/generator/cmak...
pylib/gyp/generator/cmake.py:1064: #This should be irrelevent, but some targets
extract .o files from .a
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> irrelevant

Done.

https://codereview.chromium.org/12314014/diff/251001/test/builddir/gyptest-al...
File test/builddir/gyptest-all.py (right):

https://codereview.chromium.org/12314014/diff/251001/test/builddir/gyptest-al...
test/builddir/gyptest-all.py:26: # Android, Ninja, and CMake do not support
setting the build directory.
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> Ninja has -Goutput_dir, but not sure what "build directory" means in this
file.

Heh, I had the same issue with 'relocation'. From the test, this is just testing
what's in builddir.gypi, that is, msvs
msvs_configuration_attributes/OutputDirectory and xcode xcode_settings/SYMROOT.
Maybe it would be better to simply have this test just run with xcode and msvs
instead of using exclusion of everything which isn't. So by 'build directory' I
mean overriding the build directory with configuration specific hacks.

https://codereview.chromium.org/12314014/diff/251001/test/defines/gyptest-def...
File test/defines/gyptest-define-override.py (right):

https://codereview.chromium.org/12314014/diff/251001/test/defines/gyptest-def...
test/defines/gyptest-define-override.py:16: #CMake loudly warns about passing
'#' to the compiler and drops the define.
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> nit: space after #

Done.

https://codereview.chromium.org/12314014/diff/251001/test/lib/TestGyp.py
File test/lib/TestGyp.py (right):

https://codereview.chromium.org/12314014/diff/251001/test/lib/TestGyp.py#newc...
test/lib/TestGyp.py:540: Subclass for testing the GYP CMake generator.
On 2013/11/19 16:35:28, Nico (in Tokyo until Nov 25) wrote:
> ", using cmake's ninja backend"?

Done. Way back when I had this set up so it could be configured to use the
'make' backend as well, but that's just a pain and I don't use 'make' anyway.

[bungeman-chromium, 2013-11-20 22:31:24]

Committed patchset #16 manually as r1796 (presubmit successful).