CMake build anyone?

CMake build anyone?

[Óscar Fuentes]

For my throw-away Qt project I'm creating a cmake-based build of
Emacs. My initial plan is to implement just the stuff strictly required
for my purposes, but if there is interest from any of the build
maintainers, I could steer the development for creating a "serious"
build specification.

After reading nt/INSTALL, for instance, I got the impression that the
build process would be simplified using cmake, and maintaining the cmake
build scripts would be easier than the current w32 makefiles (no shell
wizardly required, direct support for autoconfig-like tests, etc)

CMake: http://www.cmake.org

Re: CMake build anyone?

[Juanma Barranquero]

On Sun, Jul 25, 2010 at 19:28, Óscar Fuentes <ofv@wanadoo.es> wrote:

> but if there is interest from any of the build
> maintainers, I could steer the development for creating a "serious"
> build specification.

IIUC, with CMake you write configuration files, from which you
generate "normal" makefiles.

That would be useful if it allowed us to simplify and reunify the
GNU/Linux / Unix / MSDOS / Windows makefiles, bulding them from a
single set of source configuration files. That's not a trivial
undertaking, though.

    Juanma

Re: CMake build anyone?

[Óscar Fuentes]

Juanma Barranquero <lekktu@gmail.com> writes:

> On Sun, Jul 25, 2010 at 19:28, Óscar Fuentes <ofv@wanadoo.es> wrote:
>
>> but if there is interest from any of the build
>> maintainers, I could steer the development for creating a "serious"
>> build specification.
>
> IIUC, with CMake you write configuration files, from which you
> generate "normal" makefiles.

Yes. CMake creates files for your tool of choice (`make', `nmake',
Eclipse, etc.) It is not a build tool itself, but a build generator.

> That would be useful if it allowed us to simplify and reunify the
> GNU/Linux / Unix / MSDOS / Windows makefiles, bulding them from a
> single set of source configuration files. That's not a trivial
> undertaking, though.

I did that for LLVM (1) and it is perfectly doable for Emacs,
AFAIK. Maybe there are a few configuration tests that are not trivial,
but we can "borrow" from autoconf.

Please note that I'm not proposing throwing away the autoconf-based
build (not because CMake can't replace autoconf/automake, but just
because I already created enough controversy here.) Just providing cmake
as an option where autoconf is available, and as the official method
where it isn't.

One caveat: knowing nothing about the emacs MSDOS build, I don't know
how well CMake works for it. Maybe we can use cmake's cross-compiling
feature for driving the MSDOS compiler (DJGPP?)

1: http://www.llvm.org

Re: CMake build anyone?

[Juanma Barranquero]

On Sun, Jul 25, 2010 at 19:57, Óscar Fuentes <ofv@wanadoo.es> wrote:

> I did that for LLVM (1) and it is perfectly doable for Emacs,
> AFAIK. Maybe there are a few configuration tests that are not trivial,
> but we can "borrow" from autoconf.

You're talking about emulating the current autoconf machinery, so
talking about the Unixy targets. And I'm saying that the trouble right
now is that the Windows makefiles have drifted quite a lot from the
Unix ones (IIRC they are based on very old Unix makefiles) and
unifying them will be difficult, or at least arduous.

> Just providing cmake
> as an option where autoconf is available, and as the official method
> where it isn't.

That'd be great.

> One caveat: knowing nothing about the emacs MSDOS build, I don't know
> how well CMake works for it. Maybe we can use cmake's cross-compiling
> feature for driving the MSDOS compiler (DJGPP?)

You'll have to ask Eli, but somehow I don't think the MSDOS port will
be harder than the Windows one, quite the contrary.

    Juanma

Re: CMake build anyone?

[Óscar Fuentes]

Juanma Barranquero <lekktu@gmail.com> writes:

> You're talking about emulating the current autoconf machinery, so
> talking about the Unixy targets. And I'm saying that the trouble right
> now is that the Windows makefiles have drifted quite a lot from the
> Unix ones (IIRC they are based on very old Unix makefiles) and
> unifying them will be difficult, or at least arduous.

Based on my experience with LLVM, which build process was more
convoluted than Emacs', I don't expect problems with the Windows
build. I've looked at nt/config.nt and much of the stuff there already
is implemented by my Unix prototype. AFAIK the Windows build follows the
same logical steps as the Unix build, but working around the limitations
of not having autoconf etc. That workarounds will disappear for the most
part on the cmake build. nt/config.nt, for instance, will be unnecessary
and src/config.in (or actually src/config.cmake) will be used for all
platforms.

>> Just providing cmake as an option where autoconf is available, and as
>> the official method where it isn't.
>
> That'd be great.

Okay.

Re: CMake build anyone?

[Juanma Barranquero]

On Sun, Jul 25, 2010 at 21:44, Óscar Fuentes <ofv@wanadoo.es> wrote:

> Based on my experience with LLVM, which build process was more
> convoluted than Emacs', I don't expect problems with the Windows
> build.

I'm starting to think you're an incurable optimist :-)

> AFAIK the Windows build follows the
> same logical steps as the Unix build, but working around the limitations
> of not having autoconf etc.

I think a more accurate description is "most of the same logical
steps, in slightly different ways, in sometimes quite different
order". Windows makefiles are based on old Unixy makefiles. Currently
there are many differences, some trivial, some deeper.

> Okay.

If you make it work, you'll have my gratitude. Maintaining makefiles
is near the bottom on my least-palatable-things-to-do list.

    Juanma

Re: CMake build anyone?

[Eli Zaretskii]

> From: Óscar Fuentes <ofv@wanadoo.es>
> Date: Sun, 25 Jul 2010 19:57:50 +0200
> 
> One caveat: knowing nothing about the emacs MSDOS build, I don't know
> how well CMake works for it. Maybe we can use cmake's cross-compiling
> feature for driving the MSDOS compiler (DJGPP?)

As long as you are not asking the end users to have CMake installed
(and I hope you don't), the DOS build will be okay.  It actually uses
the same Makefile.in files used by Posix platforms, but instead of
running the configure script, it edits them with a bunch of Sed
scripts in the msdos/ subdirectory.

(There's a fully functional DOS port of Bash, so in principle we could
just run configure.  However, since Emacs is one of a few tools needed
to bootstrap a working DJGPP development environment, we deliberately
keep its build system independent of Bash and all the other utilities
Bash needs to run configure, and keep the list of required utilities
to the bare minimum.)

Re: CMake build anyone?

[Eli Zaretskii]

> From: Juanma Barranquero <lekktu@gmail.com>
> Date: Sun, 25 Jul 2010 20:21:53 +0200
> Cc: emacs-devel@gnu.org
> 
> > One caveat: knowing nothing about the emacs MSDOS build, I don't know
> > how well CMake works for it. Maybe we can use cmake's cross-compiling
> > feature for driving the MSDOS compiler (DJGPP?)
> 
> You'll have to ask Eli, but somehow I don't think the MSDOS port will
> be harder than the Windows one, quite the contrary.

Yes, DJGPP is a Posix.2-compliant environment (because it comes with a
Posix-compliant standard library), while MinGW uses the system
libraries that aren't Posix.

Re: CMake build anyone?

[Óscar Fuentes]

Eli Zaretskii <eliz@gnu.org> writes:

>> One caveat: knowing nothing about the emacs MSDOS build, I don't know
>> how well CMake works for it. Maybe we can use cmake's cross-compiling
>> feature for driving the MSDOS compiler (DJGPP?)
>
> As long as you are not asking the end users to have CMake installed
> (and I hope you don't), the DOS build will be okay.

A CMake install is required. Maybe that is an issue for MSDOS users,
because there is no cmake port for MSDOS AFAIK. Hence the only option
for MSDOS users would be cross-compile, but that's not an option because
you mentioned that bootstrapping Emacs is required on a DJGPP
development environment.

If you imply that requiring cmake for Windows people is too a serious
inconvenience, it should be noted that cmake makes gnuwin32/msys
unnecessary, so you exchange one requirement for another. As an user, I
prefer to depend on cmake, because it is more compact and less intrusive
(it doesn't introduce lots of executables and you don't need it on the
PATH, for instance.)

For the time being, we can put MSDOS aside and go ahead if cmake is seen
as convenient enough by the Windows maintainers.

[snip]

Re: CMake build anyone?

[Juanma Barranquero]

On Sun, Jul 25, 2010 at 23:34, Óscar Fuentes <ofv@wanadoo.es> wrote:

> A CMake install is required.

Why? (Note: I'm talking only about MSDOS/Windows.)

You envision CMake as substituting the current Windows building
environment wholesale.

My idea is more like this:

- We have the configuration files in some place, like admin/setup.
They are common for all ports (Unix, GNU/Linux, NextStep, MSDOS,
Windows, etc.).
- Developers, or users with CMake, can run the tool and (re)generate
the makefiles. These makefiles get committed to the trunk/branch, and
contain dependencies which are always needed, like gcc / msvc, etc.
- Users or the tarballs or checkouts can run nt/configure.bat (or
equivalent) to create the additional setup info (for example, paths to
include files for the image libraries).

I.e, I think the easiest transition path is having CMake as a tool
*for the maintainers*, to regenerate the makefiles. It's not optimal,
and I don't doubt that switching to a CMake-only build system would be
advantageous; but it introduces a new dependency and it is perhaps too
radical a change. Instead, we can use it like the makeinfo stuff or
the Unicode data files from admin/unidata are used right now: the user
can regenerate info or uni-*.el files, but it is usually not required.

> For the time being, we can put MSDOS aside and go ahead if cmake is seen
> as convenient enough by the Windows maintainers.

As long as it is an alternative and not a replacement, it seems like a
good idea worth trying IMHO.

All this conditional on RMS, Stefan and Chong not opposing using CMake
as an alternate build system for political or technical reasons, of
course.

    Juanma

Re: CMake build anyone?

[Óscar Fuentes]

Juanma Barranquero <lekktu@gmail.com> writes:

>> A CMake install is required.
>
> Why? (Note: I'm talking only about MSDOS/Windows.)

[snip]

> I.e, I think the easiest transition path is having CMake as a tool
> *for the maintainers*, to regenerate the makefiles.

This is a common requirement but not possible. CMake does the job of
`configure' and hence it is sensitive to the contents of the user's
machine and to the options he sets. Then, build files for his tool of
choice are generated. They can be makefiles for make, or nmake, or
jom... They can be project files for an assortment of IDEs too.

Even if it were possible to distribute functional makefiles (it is not,
due to technial details) they would force a very stringent set of
requirements on the user wrt which libraries must be installed, etc.

As mentioned before, I think that depending on cmake is more convenient
than depending on gnuwin32/msys.

[snip]

>> For the time being, we can put MSDOS aside and go ahead if cmake is seen
>> as convenient enough by the Windows maintainers.
>
> As long as it is an alternative and not a replacement, it seems like a
> good idea worth trying IMHO.

I'm convinced that you will appreciate the simplification CMake brings
over the current Windows build system.

> All this conditional on RMS, Stefan and Chong not opposing using CMake
> as an alternate build system for political or technical reasons, of
> course.

CMake is under the BSD license. If anyone opposes from now on to
including support for cmake on the source distribution of Emacs, please
speak up. That would save me a few hours of boring work.

Re: CMake build anyone?

[Juanma Barranquero]

On Mon, Jul 26, 2010 at 01:44, Óscar Fuentes <ofv@wanadoo.es> wrote:

> Even if it were possible to distribute functional makefiles (it is not,
> due to technial details) they would force a very stringent set of
> requirements on the user wrt which libraries must be installed, etc.

Do you mean that if I use CMake to generate makefiles in my setup,
which uses the TDM-2 mingw32 4.4.1-dw2 compiler (a port of MinGW)
installed on C:/mingw, the makefiles would be dependend on that tool,
and its installation path, and couldn't be used by someone who has GCC
3.4.5 (mingw-vista special r3) on "C:/Program Files/MinGW"?

    Juanma

Re: CMake build anyone?

[Óscar Fuentes]

Juanma Barranquero <lekktu@gmail.com> writes:

> Do you mean that if I use CMake to generate makefiles in my setup,
> which uses the TDM-2 mingw32 4.4.1-dw2 compiler (a port of MinGW)
> installed on C:/mingw, the makefiles would be dependend on that tool,
> and its installation path, and couldn't be used by someone who has GCC
> 3.4.5 (mingw-vista special r3) on "C:/Program Files/MinGW"?

Yes.

There is a FAQ for it:

http://www.cmake.org/Wiki/CMake_FAQ#Why_does_CMake_use_full_paths.2C_or_can_I_copy_my_build_tree.3F

The scenario you describe is worse still, because the other user may be
using a different mingw runtime or winapi version, or some library may
be missing from his machine, etc.

You can think that cmake must be installed on the machine where emacs is
compiled for the same reason that the `configure' script must be run on
the traditional build.

Re: CMake build anyone?

[Eli Zaretskii]

> From: Óscar Fuentes <ofv@wanadoo.es>
> Date: Sun, 25 Jul 2010 23:34:07 +0200
> 
> > As long as you are not asking the end users to have CMake installed
> > (and I hope you don't), the DOS build will be okay.
> 
> A CMake install is required. Maybe that is an issue for MSDOS users,

Forget MSDOS.  It's a problem for everyone.  It means to build Emacs
the end users will have to install a tool that is not there out of the
box.  I think this is a serious argument against using CMake.

> because there is no cmake port for MSDOS AFAIK.

That should be the least of our problems.

> If you imply that requiring cmake for Windows people is too a serious
> inconvenience, it should be noted that cmake makes gnuwin32/msys
> unnecessary

MSYS is already unnecessary (and even not recommended), and GnuWin32
is just a bunch of ported programs with no relation to one another.
The Windows build requires only `cp' and `rm', and they don't have to
be from GnuWin32.

> For the time being, we can put MSDOS aside and go ahead if cmake is seen
> as convenient enough by the Windows maintainers.

I think it's inconvenient even on GNU/Linux.

Re: CMake build anyone?

[Juanma Barranquero]

On Mon, Jul 26, 2010 at 05:07, Eli Zaretskii <eliz@gnu.org> wrote:

> The Windows build requires only `cp' and `rm'

Also makeinfo, etc. if the user wants to build info files.

> I think it's inconvenient even on GNU/Linux.

Well, that means that we will never be able to switch to another build
system; all will have some external requirements.

But I think the point is that Óscar is not proposing switching to it,
but having the two side by side. That has upsides and downsides (one
of them being that you have to maintain two mutually incompatible
build systems, with the oportunity to neglect one of them).

I must confess that I would prefer what I was asking for: a tool to
make (re)generation of the makefiles easier, instead.

Now, I think I know what you will say: that tool exists, and it's
called "sed" ;-)

    Juanma

Re: CMake build anyone?

[Óscar Fuentes]

Eli Zaretskii <eliz@gnu.org> writes:

[snip]

>> If you imply that requiring cmake for Windows people is too a serious
>> inconvenience, it should be noted that cmake makes gnuwin32/msys
>> unnecessary
>
> MSYS is already unnecessary (and even not recommended), and GnuWin32
> is just a bunch of ported programs with no relation to one another.
> The Windows build requires only `cp' and `rm', and they don't have to
> be from GnuWin32.

You need to install `cp' and `rm' from somewhere, right? I'll say that
obtaining and installing cmake is faster and safer than doing the same
with those utilities.

How many problems gnuwin32 causes? How many problems are caused by
people seeing the `rm' and `cp' requirement and installing random things
from the Internet (including MSYS)?

>> For the time being, we can put MSDOS aside and go ahead if cmake is seen
>> as convenient enough by the Windows maintainers.
>
> I think it's inconvenient even on GNU/Linux.

I'm not sure we are talking about the same here. The main convenience of
cmake is for the Windows maintainers. If I can't convince you (Eli),
Jason and Juanma about its advantages, it is pointless to add it to
Emacs.

Re: CMake build anyone?

[Óscar Fuentes]

Óscar Fuentes <ofv@wanadoo.es> writes:

[snip]

> The main convenience of cmake is for the Windows maintainers. If I
> can't convince you (Eli), Jason and Juanma about its advantages, it is
> pointless to add it to Emacs.

Let's address the issue from a different POV: please name a few things
that annoys you as matinainers of the Windows build and some features
you would like to see implemented, and I'll tell how using cmake can
help with them.

Re: CMake build anyone?

[Andreas Schwab]

Juanma Barranquero <lekktu@gmail.com> writes:

> On Mon, Jul 26, 2010 at 05:07, Eli Zaretskii <eliz@gnu.org> wrote:
>
>> The Windows build requires only `cp' and `rm'
>
> Also makeinfo, etc. if the user wants to build info files.

Info files are part of the distribution.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: CMake build anyone?

[Juanma Barranquero]

On Mon, Jul 26, 2010 at 11:18, Andreas Schwab <schwab@linux-m68k.org> wrote:

> Info files are part of the distribution.

Who's talking about "distribution"?

Eli just said "[t]he Windows build", and building Emacs on Windows out
of Bazaar needs makeinfo if you want to have info files.

Of course if you restrict yourself to distributions you can get a
binary one, and then cp and rm aren't needed either.

    Juanma

Re: CMake build anyone?

[Andreas Schwab]

Juanma Barranquero <lekktu@gmail.com> writes:

> Eli just said "[t]he Windows build", and building Emacs on Windows out
> of Bazaar needs makeinfo if you want to have info files.

If you are building from the repository you are expected to have
suitable development environment anyway.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: CMake build anyone?

[Juanma Barranquero]

On Mon, Jul 26, 2010 at 12:11, Andreas Schwab <schwab@linux-m68k.org> wrote:

> If you are building from the repository you are expected to have
> suitable development environment anyway.

If you're building from a source tarball you are expected, too. Most
people on Windows do not have MSVC or MinGW, nor the image libraries'
headers, so building will require them to download and install quite a
few things.

    Juanma

Re: CMake build anyone?

[Eli Zaretskii]

> From: Óscar Fuentes <ofv@wanadoo.es>
> Date: Mon, 26 Jul 2010 06:15:13 +0200
> 
> >> For the time being, we can put MSDOS aside and go ahead if cmake is seen
> >> as convenient enough by the Windows maintainers.
> >
> > I think it's inconvenient even on GNU/Linux.
> 
> I'm not sure we are talking about the same here. The main convenience of
> cmake is for the Windows maintainers. If I can't convince you (Eli),
> Jason and Juanma about its advantages, it is pointless to add it to
> Emacs.

I thought you were suggesting to use CMake in Emacs on all platforms.
If not, I think it's a non-starter, since I doubt that the head
maintainers will agree to add something like that just for the sake of
the Windows build (and I won't blame them).

Re: CMake build anyone?

[Eli Zaretskii]

> From: Óscar Fuentes <ofv@wanadoo.es>
> Date: Mon, 26 Jul 2010 06:45:37 +0200
> 
> please name a few things that annoys you as matinainers of the
> Windows build and some features you would like to see implemented

None in particular.  The current system works pretty well, and there
are much more important jobs to do for the Windows build, if someone
has time and energy.

Re: CMake build anyone?

[Eli Zaretskii]

> From: Juanma Barranquero <lekktu@gmail.com>
> Date: Mon, 26 Jul 2010 11:54:32 +0200
> Cc: Eli Zaretskii <eliz@gnu.org>, Óscar Fuentes <ofv@wanadoo.es>, 
> 	emacs-devel@gnu.org
> 
> On Mon, Jul 26, 2010 at 11:18, Andreas Schwab <schwab@linux-m68k.org> wrote:
> 
> > Info files are part of the distribution.
> 
> Who's talking about "distribution"?
> 
> Eli just said "[t]he Windows build", and building Emacs on Windows out
> of Bazaar needs makeinfo if you want to have info files.

No, Andreas got my drift correctly.  I was thinking first and foremost
about building a release tarball.

> Of course if you restrict yourself to distributions you can get a
> binary one, and then cp and rm aren't needed either.

Which won't work if I need to fix a bug or add a feature.  This is
still Free Software, right?

Re: CMake build anyone?

[Eli Zaretskii]

> From: Juanma Barranquero <lekktu@gmail.com>
> Date: Mon, 26 Jul 2010 12:16:45 +0200
> Cc: Eli Zaretskii <eliz@gnu.org>, Óscar Fuentes <ofv@wanadoo.es>, 
> 	emacs-devel@gnu.org
> 
> On Mon, Jul 26, 2010 at 12:11, Andreas Schwab <schwab@linux-m68k.org> wrote:
> 
> > If you are building from the repository you are expected to have
> > suitable development environment anyway.
> 
> If you're building from a source tarball you are expected, too.

And having to install one more doesn't help.

> Most people on Windows do not have MSVC or MinGW, nor the image
> libraries' headers

MinGW is not just for Emacs; I find it highly unlikely that someone
who uses Emacs does not already has some development environment on
her machine.  As for image headers and libraries, the minimal set is
right there on the GNU FTP site, where you get the source tarball.

Btw, we may need to require another non-standard (and not so small
package) in the near future: Perl, since makeinfo was reimplemented as
a Perl script, and that's what we will see in the next release of
Texinfo.

Re: CMake build anyone?

[Juanma Barranquero]

On Mon, Jul 26, 2010 at 19:03, Eli Zaretskii <eliz@gnu.org> wrote:

> None in particular.  The current system works pretty well,

Updating the Windows makefiles adrift of the Unix ones is not what I
would call "pretty well". YMMV.

    Juanma

Re: CMake build anyone?

[Juanma Barranquero]

On Mon, Jul 26, 2010 at 19:07, Eli Zaretskii <eliz@gnu.org> wrote:

> No, Andreas got my drift correctly.  I was thinking first and foremost
> about building a release tarball.

Building a release tarball is not something most users do, so the
number of prerequisites is less important.

> Which won't work if I need to fix a bug or add a feature.  This is
> still Free Software, right?

I don't get your meaning. If you need to build, there are a bunch of
things you *have* to have installed. You talked about cp and rm (and
somehow considered GCC/MinGW a given) but surely free software users
also do fix the info files and will need makeinfo, won't they?

The point being: I'm not pushing for or against CMake, but at the end
of the day, it is just *one* more prerequisite among quite a few.

I think you see it as less useful because for you the makefiles wok
pretty well, as you said. Personally, I hate having to modify a
makefile or a .bat script. Not to mention having to study changes in
the Unix build system to determine whether they make sense in the
Windows side.

    Juanma

Re: CMake build anyone?

[Juanma Barranquero]

On Mon, Jul 26, 2010 at 19:12, Eli Zaretskii <eliz@gnu.org> wrote:

> MinGW is not just for Emacs; I find it highly unlikely that someone
> who uses Emacs does not already has some development environment on
> her machine.

I only have MinGW because of Emacs. I don not use it for anything
else. I do have a development environment on my system, but certainly
it is not C/C++ oriented. Ada anyone?

> Btw, we may need to require another non-standard (and not so small
> package) in the near future: Perl, since makeinfo was reimplemented as
> a Perl script, and that's what we will see in the next release of
> Texinfo.

Funny. I consider much more onerous to force someone to install Perl*
just to make info files, than CMake to have a modern, sane build
system.

(* This is not a pet peeve, BTW; Perl is one of the first things I
install in every computer I use.)

    Juanma

Re: CMake build anyone?

[Eli Zaretskii]

> From: Juanma Barranquero <lekktu@gmail.com>
> Date: Mon, 26 Jul 2010 22:02:12 +0200
> Cc: Óscar Fuentes <ofv@wanadoo.es>, emacs-devel@gnu.org
> 
> On Mon, Jul 26, 2010 at 19:03, Eli Zaretskii <eliz@gnu.org> wrote:
> 
> > None in particular.  The current system works pretty well,
> 
> Updating the Windows makefiles adrift of the Unix ones is not what I
> would call "pretty well".

It happens seldom enough.

Re: CMake build anyone?

[Eli Zaretskii]

> From: Juanma Barranquero <lekktu@gmail.com>
> Date: Mon, 26 Jul 2010 22:19:26 +0200
> Cc: schwab@linux-m68k.org, ofv@wanadoo.es, emacs-devel@gnu.org
> 
> > Btw, we may need to require another non-standard (and not so small
> > package) in the near future: Perl, since makeinfo was reimplemented as
> > a Perl script, and that's what we will see in the next release of
> > Texinfo.
> 
> Funny. I consider much more onerous to force someone to install Perl*
> just to make info files, than CMake to have a modern, sane build
> system.

It's not like I'm thrilled to see that.  But since I stopped hacking
makeinfo several years ago, I have no say in what the maintainers
decide to do.

Re: CMake build anyone?

[Miles Bader]

Juanma Barranquero <lekktu@gmail.com> writes:
> Funny. I consider much more onerous to force someone to install Perl*
> just to make info files, than CMake to have a modern, sane build
> system.

I'm no fan of perl, but it's a very widely used, generally useful
package; it isn't surprising if someone already has perl installed,
especially if they're emacs users.

Cmake, by comparison, is a fringe product.

-Miles

-- 
My spirit felt washed.  With blood.  [Eli Shin, on "The Passion of the Christ"]

Re: CMake build anyone?

[Stefan Monnier]

From what I heard, CMake is an interesting tool and we may consider
using it for Emacs.  But the way I see it, it only really makes sense to
do that if it can be used for "all" platforms and/or if it can be used
to reduce the amount of stuff to install (e.g. in Windows).

So unless CMake can be used to make it easier to install Emacs on
Windows, I think it'll have to wait until cmake is likely to be
installed on most unixy systems.  I.e. don't hold your breath.


        Stefan

Re: CMake build anyone?

[Juanma Barranquero]

On Mon, Jul 26, 2010 at 23:19, Miles Bader <miles@gnu.org> wrote:

> I'm no fan of perl, but it's a very widely used, generally useful
> package; it isn't surprising if someone already has perl installed,
> especially if they're emacs users.

Not on Windows.

> Cmake, by comparison, is a fringe product.

All modern build systems are, almost by definition.

    Juanma

Re: CMake build anyone?

[Óscar Fuentes]

Stefan Monnier <monnier@IRO.UMontreal.CA> writes:

> From what I heard, CMake is an interesting tool and we may consider
> using it for Emacs.  But the way I see it, it only really makes sense to
> do that if it can be used for "all" platforms and/or if it can be used
> to reduce the amount of stuff to install (e.g. in Windows).

CMake scripts are inherently cross-platform and in the case of Emacs the
build would work for all platforms, except MSDOS. Cross-compiling could
be used for generating MSDOS executables, but Eli already explained that
bootstrapping Emacs on MSDOS is required by DJGPP.

On Windows, we remove one requirement (`rm' and `cp') and add another
(cmake.) CMake does lots of other interesting things, though. This is a
partial list of the features and benefits provided by cmake (`cmake'
here includes `ctest' and `cpack'):

http://www.cmake.org/Wiki/Really_Cool_CMake_Features

> So unless CMake can be used to make it easier to install Emacs on
> Windows, I think it'll have to wait until cmake is likely to be
> installed on most unixy systems.  I.e. don't hold your breath.

Fair enough. But please note that I was not proposing replacing the
current autoconf build with cmake (although some users could benefit
from having a cmake build on Unix, see the feature list linked above.)

My proposal was motivated by my impression about the Windows build being
hard to maintain and I thought that the developers could be interested
on an alternative that removes most of the burden so they can focus
their efforts elsewhere. Since only Juanma expressed some interest and
other people objected, let's forget the proposal and move on.

Re: CMake build anyone?

[Stefan Monnier]

> Fair enough. But please note that I was not proposing replacing the
> current autoconf build with cmake (although some users could benefit
> from having a cmake build on Unix, see the feature list linked above.)

I understand it, but I think it'd only work if the intention is to
replace the current build system (and not just eventually), otherwise
it'd mean maintaining two very different build systems, which doesn't
sound much better than what we have now.

> My proposal was motivated by my impression about the Windows build being
> hard to maintain and I thought that the developers could be interested
> on an alternative that removes most of the burden so they can focus
> their efforts elsewhere. Since only Juanma expressed some interest and
> other people objected, let's forget the proposal and move on.

As I said, I think it's a good idea, but Emacs being what it is, it'll
probably only end up doing it many years from now (just like it took
ages for Emacs to adopt ANSI C, and to fully make use of autoconf).


        Stefan

Re: CMake build anyone?

[Miles Bader]

Stefan Monnier <monnier@IRO.UMontreal.CA> writes:
> As I said, I think it's a good idea, but Emacs being what it is, it'll
> probably only end up doing it many years from now (just like it took
> ages for Emacs to adopt ANSI C, and to fully make use of autoconf).

That's probably for the better anyway, as there's currently no obvious
winner among the "new gen build systems," and all have their problems.
Maybe in a few years, there will be more of a consensus (perhaps someone
will do what Linus did with git -- get sick of the whole sordid mess,
and finally do it right).

-Miles

-- 
Patience, n. A minor form of despair, disguised as a virtue.

Re: CMake build anyone?

[Wojciech Meyer]

Miles Bader <miles@gnu.org> writes:

> Stefan Monnier <monnier@IRO.UMontreal.CA> writes:
>> As I said, I think it's a good idea, but Emacs being what it is, it'll
>> probably only end up doing it many years from now (just like it took
>> ages for Emacs to adopt ANSI C, and to fully make use of autoconf).
>
> That's probably for the better anyway, as there's currently no obvious
> winner among the "new gen build systems," and all have their problems.
> Maybe in a few years, there will be more of a consensus (perhaps someone
> will do what Linus did with git -- get sick of the whole sordid mess,
> and finally do it right).

I think CMake got it `almost right'. The language is a bit obscure but
the idea is right. I am not sure however if in case of projects like
Emacs it is worth, it is not even about creating `not mature enough'
dependency (being dependent on something like Perl for instance is more
reasonable, as it is more mature project), but about what Autoconf
offers in terms of portability. Autoconf & Automake makes so good job on
checking if everything is in place before the actual build happens, that
it might be just sub-optimal vs work that should done to do it properly.
(but that's my private opinion). OTOH project like KDE switched from
autoconf to cmake with a big success, so possibly it could be done. (but
KDE is much bigger project and they'd had scalability problems).

Wojciech

>
> -Miles

Re: CMake build anyone?

[Tom Tromey]

>>>>> "Miles" == Miles Bader <miles@gnu.org> writes:

Miles> That's probably for the better anyway, as there's currently no obvious
Miles> winner among the "new gen build systems," and all have their problems.
Miles> Maybe in a few years, there will be more of a consensus (perhaps someone
Miles> will do what Linus did with git -- get sick of the whole sordid mess,
Miles> and finally do it right).

Why should Emacs require the GNU version control tool but not require
the GNU build system?  An irritating question, but one that I think
ought to be answered.

Tom

Re: CMake build anyone?

[Miles Bader]

Tom Tromey <tromey@redhat.com> writes:
> Miles> That's probably for the better anyway, as there's currently no obvious
> Miles> winner among the "new gen build systems," and all have their problems.
> Miles> Maybe in a few years, there will be more of a consensus (perhaps someone
> Miles> will do what Linus did with git -- get sick of the whole sordid mess,
> Miles> and finally do it right).
>
> Why should Emacs require the GNU version control tool but not require
> the GNU build system?  An irritating question, but one that I think
> ought to be answered.

I don't think Emacs should "require" either (especially when they're GNU
packages in name only) -- using GNU tools is definitely a nice thing
(for the reasons RMS often states), but it's just one factor to
consider, and should never be a rigid requirement (whereas, for
instance, using free software _is_, and should be, a rigid requirement).

That said, I actually like automake a lot (it's what I use for my
personally created software), and think something like quagmire shows
promise (if anybody worked on it... :/ ).  So I'd definitely consider
them as candidates for a possible build-system change for Emacs
(although I don't know how well they'd suit Emacs' hairy build process).

-Miles

-- 
Arrest, v. Formally to detain one accused of unusualness.

Re: CMake build anyone?

[Stefan Monnier]

> Why should Emacs require the GNU version control tool but not require
> the GNU build system?

Emacs does not require the GNU version control tool (ever heard of
"Emacs release", or "tarball"?).


        Stefan

Re: CMake build anyone?

[Alfred M. Szmidt]

I seriously hope nobody is suggesting to break "./configure && make
all install uninstall" which is so far the only sane way to build
anything; CMake as such breaks that.