Disallow .local declarations in long-style call statement

[kjs]

the current implementation for PIR allows one to declare .locals in a .begin_call/.end_call sequence, after the .call directive, like so:

.begin_call .arg 1 .arg 2 .call $P0 .local int a, b, c .result a .result b .result c .end_call

(this is short for (a, b, c) = $P0(1, 2) )

IMHO, the whole sequence of directives from .begin_call to .end_call should be considered 1 statement; having .local declarations intermixed with .result directives is a bit strange, because the .local directive itself is not part of the sequence of directives to make the call.

Furthermore, the locals stay in scope after the .end_call directive.

I propose to disallow this language 'feature', and force the programmer to write the above example as follows:

.local int a, b, c .begin_call .arg 1 .arg 2 .call $P0 .result a .result b .result c .end_call

comments welcome.

kjs

[coke]

This was originally opened as  http://rt.perl.org/rt3/Ticket/Display.html?id=58070

[plobsing]

A PIR compiler isn't in the business of enforcing good style. The current PIR compiler copes fine with this sequence and there is no technical basis for dissallowing this syntax.

If this makes an alternate PIR compiler more complicated, it can be ommitted along with other misfeatures and bugs. I doubt anyone wants a bug-for-bug compatible remake of IMCC.