Eval equivalent in XS functions

I'm looking for a way to have an EVAL like wrapper around a piece of code
in my XS module (XML::Parser::Expat.) I know about the G_EVAL flag to use
with perl_call_sv, but it looks like it probably adds significantly more
overhead to perl_call_sv calls, right where XML::Parser can least afford it.

And I don't think I need to catch exceptions after each handler. If a parse
fails, the low level parser object should be reclaimed anyway. Currently
exceptions are caught in the XML::Parser parse method with an eval wrapped
around a call to the XML::Parser::Expat parse method.

There's a problem, though, if an exception occurs while parsing an external
entity. Expat requires you to create a subsidiary parser for external entities
and recursively parse them. This parser should be reclaimed after you're
through with the external entity. Currently this won't happen when an
exception occurs in the subsidiary parse.

Also I've realized that I need to provide a pair of callbacks for external
entities to the user in order for them to cleanly use an IO::Handle to read
from an external entity. One to open the Handle and one to close it. (Also
useful for pushing and popping the current basename.) If an exception happens
during the parse, you still want the corresponding finishing callback to made.

I was considering experimenting with cutting and pasting the
"if (G_EVAL & flag) {...}" blocks from perl_call_sv and adapting them to
the externalEntity handler in Expat.xs, but this seems exceedingly fragile.

I was also thinking about using perl_call_sv on a sub that just calls an
XS function that wraps the code in question. The only purpose of the sub
is for me to use perl_call_sv with G_EVAL. This is what I'll probably do
unless I'm advised otherwise by you folks.

If you have advice for me, please CC: it to me as well as the list.

Thanks,
Clark

-- 
Clark Cooper		Software Engineer	Home:   coopercc@netheaven.com
			Schenectady, NY	USA	Work:   cccooper@ltionline.com

• Eval equivalent in XS functions by Clark Cooper
• Re: Eval equivalent in XS functions by Nick Ing-Simmons