Stefan Monnier wrote:
> At least I know Richard made me undo some "Lisp_Object foo = ..." years
> ago, saying that we shouldn't do that because it fails to compile in
> some cases (presumably when you used some specific compiler and you also
> asked for Lisp_Object to be a union rather than an int).
> I did not heed his advice very eagerly, to tell you the truth.

Many pre-C89 compilers indeed did not support union initializers (i.e., 'union u 
x = { something };', so perhaps there was some confusion about what worked and 
what didn't. Anyway, that water passed under the bridge long ago.

In reviewing this patch, I noticed some other confusion in this area that I 
introduced in 2013 while pacifying gcc -Wclobbered, and which I think you asked 
about but I didn't understand your question at the time (sorry). I attempted to 
fix the problem just now by installing the attached patch, which undoes Chris's 
simplification and then reworks the setjmp calls to use a local rather than a 
global. This avoid a need to load from a static variable in the normal flow of 
execution.