[21.5] Fix crash in PNG initialiation [was: can't move cursor ...]

[Stephen J. Turnbull]

21.5 APPROVE COMMIT

Stephen J. Turnbull writes:

 > OK, yes, this is really ugly.  (But then, when isn't a setjmp/longjmp
 > ugly?)  I think the attached patch should fix the problem, but I don't
 > yet have a test case and I've never patched code involving setjmp/
 > longjmp before so a second opinion would be welcome (and a third...).
 > (I think the relevant test case for the original bug can be easily
 > constructed by substituting "You're going down, sucker!" for
 > PNG_LIBPNG_VER_STRING, but I haven't done that yet.)

I still need to do this, but

 > I'll run with it a while before committing, but would really
 > appreciate comments.

I've run, I've pulled in a bunch of PNGs including some that our PNG
library doesn't like, with no trouble yet.

 > 2007-02-17  Stephen J. Turnbull  <stephen at xemacs.org>
 > 
 > 	* glyphs-eimage.c (png_instantiate_unwind): Avoid recursion.
 > 	(png_instantiate): Initialize setjmp_buffer early, and avoid
 > 	recursive entry to error handler.