Re: The build system

[John Swensen]

On Jul 8, 2010, at 12:24 PM, Jordi Gutiérrez Hermoso wrote:

> On 8 July 2010 10:38, forkandwait <address@hidden> wrote:
>> Jordi Gutiérrez Hermoso <jordigh <at> gmail.com> writes:
>> 
>>> I wish we could use something else to build Octave,
>> 
>> I am talking way out of my expertise here, but I think cmake is considered a
>> very reasonable replacement for autotools:
> 
> I was sorely tempted to drop the c-word myself, but I doubt that
> Octave could use it. Partly, because it's not GNU, but also, because I
> think just the way the sources themselves are written, they're tightly
> integrated with an assumption of the GNU build system. Furthermore,
> I'm not sure cmake meets Octave's requirements, and it doesn't aim to
> solve the problem of maximum *nix portability (Octave gets compiled a
> lot on Solaris and AIX machines and other industrial Unices, let's not
> forget), but instead it aims to solve the problem of
> Windows/MacOSX/general-Unix portability.
> 
> Maybe if I amass enough hubris, I'll try anyways to see what building
> Octave with cmake is like. :-) But not soon.
> 

Looking at the CMake website, it looks like there are binary distributions of 
CMake for Windows, Linux, OSX, SUNOS, IRIX, HPUX, and AIX.  I have worked with 
CMake on a couple of projects at school (mostly because dependent libraries 
also used CMake), but I found CMake quite easy to use.  On the other hand, I 
was only compiling ~100 source files with only a few configuration-time 
options, so I am sure there would be a whole slew of issues if someone tried to 
switch over.

I think the real questions that need answered are:
1) Does CMake (or your other favorite build system) have a smaller learning 
curve than Autotools?  If the answer is yes, then this may allow people to 
contribute to this part of Octave more easily.
2) Will switching to CMake (or your other favorite build system) be any less 
complex than the current autotools system?  If the answer is yes, then it would 
allow more people to understand the whole build process.
3) Would switching to CMake (or your other favorite build system) actually 
provide fixes for any of the build complaints people have (e.g. better 
dependency management, faster build times, better modularization of builds)?
And of course, you still have to weigh the answers to these against the other 
problems that would crop up if you switched.

John Swensen