Re: Cray C-90 (UNICOS 8.0.3) support?

[Paul Haahr <haahr@netcom.com>]

> I have been trying to get es-0.84 to compile on our Cray C-90, running
> Unicos 8.0.3 and I keep getting the following message:
> 
>   (vn.~/usr/build/es-0.84) make
>   cc -hstdc -O -I/u/vb/gray/usr/include/readline   -c closure.c -o closure.o
>   cc-408 cc: ERROR File = closure.c, Line = 126
>     The macro "setjmp" appears in an invalid context.
> 
> The problem is in the pushhandler macro from es.h.  Has anyone else had
> difficulty with this?

I believe that there's no simple workaround.  I got this note from the
last person who I knew of who tried to bring es up on a Cray.  (I don't
see how Cray can say they ship ANSI C while imposing these restrictions,
but I'm not in the supercomputer business.)

| From: djc@niwot.scd.ucar.EDU (Dennis Colarelli)
| To: haahr (Paul Haahr)
| Subject: Re: for better or for worse,
| Date: Fri, 30 Apr 93 15:36:26 MDT
| 
| > can you tell me what the restrictions on setjmp are?  i'll try to make
| > pushhandler() more portable if you can tell me what has to be done.
| 
| Attached are the restrictions from the UNICOS setjmp man page.  Note
| that if you want to run es on the Cray 3, CSOS has, I believe, even further 
| restrictions.  I'll know more when it gets here.
| 
|    An invocation of the setjmp macro can appear only in one of the following
|    contexts:
| 
|      o  The entire controlling expression of a selection or iteration
|         statement (if, for, while, do, switch)
| 
|      o  One operand of a relational or equality operator with the other
|         operand an integral constant expression, and with the resulting
|         expression being the entire controlling expression of a selection
|         or iteration statement
| 
|      o  The operand of a unary ! operator with the resulting expression
|         being the entire controlling expression of a selection or
| 
|      o  The entire expression of an expression statement (possibly cast
|         to void)

The definition of pushhandler is:

	#define	pushhandler(hp)	( \
			((hp)->rootlist = rootlist), \
			((hp)->pushlist = pushlist), \
			((hp)->evaldepth = evaldepth), \
			((hp)->up = tophandler), \
			(tophandler = (hp)), \
			(setjmp((hp)->label) ? raised(exception) : NULL) \
		)

which clearly violates those assumptions.  On the other hand, all but
one of its uses are of the form

	List *e;
	Handler h;
	...
	if ((e = pushhandler(&h)) != NULL) {
		... catching code using e ...
	}
	... main body ...
	pophandler(&h);

and the other use has a while instead of an if and could be rewritten
to be an if inside a loop with an extra test, so macrology of the form

	#define exception_handler \
		{ \
			Handler _localhandler;
			_localhandler.rootlist = rootlist; \
			_localhandler.pushlist = pushlist; \
			_localhandler.evaldepth = evaldepth; \
			_localhandler.up = tophandler; \
			tophandler = &_localhandler; \
			if (!setjmp(_localhandler.label)) {
	
	#define catch_exception(e) \
				pophandler(&_localhandler); \
			} else { \
				List *e = raised(exception); \
	
	#define end_exception_handler \
			} \
		}

combined with rewriting all the uses as

	...
	exception_handler
		... main body ...
	catch_exception (e)
		... catching code using e ...
	end_exception_handler

would probably be acceptable to Cray.

Pardon me if my C code is somewhat rusty.  The ever-forthcoming next
release of es may include something like this.  For now, you'll have to
roll it yourself.

Paul