On 2015/12/09 18:25, Eric Wong wrote:
>> > (Sorry if I overlooked marking)
> Wait, putobject calls (and TS_VALUE operands) do not go away when
> opt_case_dispatch insn is emitted, so marking keys in literal hash is
> redundant. since iseq_set_sequence adds marks for all TS_VALUE.
> (but maybe I'm misreading compile.c, too)

OMG. You are correct.
(I forget that the following instructions)

>> > I'm not sure it is worth or not. IMO keeping simple with Hash is better
>> > on this case with your measured improvements.
> Yes, we can revisit after 2.3 release.  Maybe this is appropriate for
> sorted array + bsearch since the table is frozen at compile time and
> st uses too much memory.

I agree. But I'm not sure we can do bsearch (can we compare with
different type values?).

-- 
// SASADA Koichi at atdot dot net