Re: Arguments against switching to SCons?

As you know I'd rather welcome a switch away from autotools, but I've 
been persuaded by the arguments in the past (all being perfectly valid).
But now if even Thatcher says that the autotools 'are stretching it'...

As to Thatchers questions:
I'd say SCons/Lua would be a very good choice:
* bootstrapping the Lua interpreter is a matter od fairly standard 
Makefiles (no dlls needed etc.)
* Asko has demonstratet that a complete Lua system (LuaX) can be 
prefectly built with SCons/Lua alone, no need for Python
* SCons is written to work with _any_ version of Python higher than 
1.5.2 I think, so any Python should do if we do need it at all

One problem I do see is the fact that SCons[/Lua] doesn't produce 
standard Makefiles that could be dostributed in the source tarballs, so 
each user wishing to compile needs to use SCons/Lua...

In that vein I was looking at CMake, which seems closer to autotools... 
I haven't used it yet, but from the docs it seems rather powerfull and 
straightforward to use. Added benefit: One description file produces 
Makefiles/Prjectfiles for all platforms. THat is, vc6/7 project files 
(or gcc Makefiles too) for windows, etc.
(btw SCons also supports ms compilers on windows etc.)

One point common to both SCons and CMake would be gaining more 
flexibility compiler wise. autotools are very limited to gcc/gnu make ...

In conclusion I guess I'd tend towards using SCons/Lua. CMake would need 
to be bootstrapped on most platforms (not linux, win32 and osx) also. 
And we can nicely use a Lua interpreter and Luac/bin2c anyway during 
many build processes, so we need to build then anyway somewhere along 
the way...

So there's my .2 eur... :)

-Martin

Thatcher Ulrich wrote:

> I have a lot of patience, but autotools is really stretching it.  Just
> to review, here are the reasons I know of NOT to switch to SCons.  If
> these reasons are invalid or obsolete or just not important, I could
> be convinced to try switching.
>
> * Requires Python.  The four CF hosts that we do nightly builds on all
>  have python, although not all identical versions (I think the Alpha
>  machine has 2.1, and the others have 2.2).
>
>  Are there current luacheia users out there who can't/won't install
>  python and SCons in order to build luacheia?  Juergen Fuhrman had
>  expressed concern about this in the past, w/r/t TRU64 and/or some
>  other systems he works on.  I was persuaded by this argument before,
>  but my opinion has changed somewhat -- in over a year, we have not
>  heard a peep from any luacheia users on platforms other than Linux,
>  Windows, OSX, and OpenBSD.
>
>  Personally, My ISP runs FreeBSD, so I care about that too, but lack
>  of python is not a problem there.
>
> * Asko's SCons/Lua seems cool, but it sounds like users (or build
>  scripts we write) would have to bootstrap Lua in order to use it?
>  And fancier features would require python anyway (not sure if that's
>  true)?  That's not the end of the world; just extra complication.
>
>  Also, in general I'm not a big fan of using a non-standard version
>  of a project, when there is an actively maintained version with
>  documentation, community etc.
>
> * It's something new and involves a learning curve and effort to get
>  it going.  What we have now kinda does work.
>
> * What will the source tarball user experience be like?  I don't know
>  if I've ever installed a source tarball that relied on SCons.  Can
>  SCons smoothly support "./configure && make install" without a lot
>  of work on our part to set it up?
>
>  




-------------------------------------------------------
This SF.Net email is sponsored by: IBM Linux Tutorials
Free Linux tutorial presented by Daniel Robbins, President and CEO of
GenToo technologies. Learn everything from fundamentals to system
administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click