Nikodemus Siivola <nikodemus@...> writes:
> The attached patch moves error traps out of line into the elsewhere
> segment on x86 and x86-64, and reuses traps when possible (that is,
> when the elsewhere segment already constains an error trap with the
> same code and argument registers.)
> Core size goes down a bit, but no performance-vice this seems the make
> no difference at all.
Notice that functions using PPRINT-LOGICAL-BLOCK etc. will contain lots
of UNBOUND-SYMBOL-ERROR traps. For example, the disassembly of
PPRINT-FILL (a pretty simple function) contains 11 such traps. The
disassembly of PPRINT-LAMBDA-LIST contains 30!
Perhaps your patch should be reevaluated on that ground?