I think I "broke" mkoctfile for 3.4.1

[John W. Eaton]

On 18-Jun-2011, Jordi Gutiérrez Hermoso wrote:

| There's a change  I pushed on the stable branch for 3.4.1
| 
|     http://hg.savannah.gnu.org/hgweb/octave/rev/2c54fde0f397
| 
| which breaks building some Octave-Forge packages. It concerns with how
| they are passing -Wall to mkoctfile. With my change, in imitation of
| how you pass flags to ld through gcc, you should now pass it as
| -W,-Wall. I couldn't figure out from mkoctfile's docstring if the
| previous way of just using -Wall worked coincidentally, but I think it
| did. I changed this method of passing flags to gcc through mkoctfile
| because I couldn't pass -std=c++0x, and it was inconvenient to do with
| an environment variable for what I was trying to do for contract work
| (which, btw, has just turned into a formal offer of employment to
| improve Octave for this company, yay!).
| 
| Should this change be backed out? I think breaking mkoctfile this way
| and the missing symlink reported in the help mailing list is
| sufficiently bad and I wish we had caught this before releasing 3.4.1.

Yes, please undo this change.

We should discuss how (or whether) we want to handle arbitrary options
for the compilers.  Maybe we should recognize CFLAGS="..",
CXXFLAGS="...", and FFLAGS="..." as options?

I think we should probably make a 3.4.2 release relatively soon.

jwe