Re: [Flightgear-devel] [RFC] Building one unified fgfs executable

[Hans U. N. <hu...@n-...>]

Melchior FRANZ wrote:
> * Hans Ulrich Niedermann -- Wednesday 25 July 2007:
>>  23 files changed, 698 insertions(+), 184 deletions(-)
> 
> Umm ... and this huge change only to stuff GLUT, SDL, *and*
> OSGVIEWER support into the same binary, as a minor development
> helper for two months or three, until we can rip out GLUT and
> SDL support altogether? After which we change the 23 files back?
> Sounds like a big pain for little gain to me.  :-}

To the contrary: I think easily switching between the GLUT and osgViewer
versions without having to do multiple compiles every time will greatly
accelerate the bugfixing period for osgViewer.

Then in a few months, when osgViewer has been made work even in these
problematic cases, we can remove the GLUT support and my abstraction
layer again and exclusively rely on osgViewer.

That should in the worst case not be more difficult than removing
fg_gsa.{cxx,hxx} and get rid of the resulting compile and link errors.

As to SDL...

As Hans Fugal and myself just found out the hard way while making the
patch work on OSX, SDL ist a PITA to coexist with anything else in the
same executable (use a search engine for "SDL_main").

Given that SDL has thus never worked on OSX (the proper linking flags
were missing), and Vivian is not aware of any Windows user using SDL,
the only platform where SDL seems to have ever been used is *ix.

Discussion on IRC hinted at the following consensus:
   * Completely remove the SDL support.
   * Rely exclusively on osgViewer in the long term.
   * Keep GLUT support for short term until osgViewer works as
     well or better than GLUT everywhere.

Melchior has expressed his horror at the amount of code for the
abstraction layer which allows both GLUT and osgViewer versions in the
same executable, while Hans Fugal and I think that the easier debugging
by immediate comparisons between GLUT and osgViewer versions compensates
for that.

(Note: For FG/plib, SDL the only available alternative to often fatally
buggy glut, and should thus probably be kept. Therefore, this discussion
is purely about FG/OSG.)

So, where do we go from here?

Rip out SDL from current CVS? IMHO, yes.

Provide easy switching between GLUT and osgViewer via command line
switch? IMHO: Yes, absolutely, until we remove the GLUT part.

Include the ripping out of SDL into my patch? That would make my patch
look less frightening in diffstat :) OK, doing it separately is cleaner.