Further addenda:

ONMATCH cannot strictly be bracketed by FAIL. Here is a counter-example:

ONMATCH L1
CHAR 'a'
// fall-through
ONMATCH L2
CHAR 'b'


But note that under my previously proposed jump scoping rule this is
equivalent to:

ONMATCH L1
CHAR 'a'
JMP L3
FAIl
L3: ONMATCH L2
char 'b'


The main issue, I think, is that we want a way to hoist things
incrementally across SPLIT instructions. The challenge is to choose a VM
instruction set that facilitates this. Predicated instructions (in the
style of ONMATCH) facilitate dense encodings, but don't necessarily
facilitate simple transformations. If we really buy in to the notion of
predicated instructions, then perhaps the bracketing idea should go out the
window.


shap
_______________________________________________
bitc-dev mailing list
[email protected]
http://www.coyotos.org/mailman/listinfo/bitc-dev

Reply via email to