Ruby

normalperson@yhbt.net wrote:

https://bugs.ruby-lang.org/issues/13355

This is similar in spirit to opt_case_dispatch as the literal
Range here is guaranteed to be immutable when used for
checkmatch.

Along with the current behavior of using putobject instead of
putstring for literal strings, when there are non-literals.

Anyways, this only puts extra code into compile.c (not insns.def),
so it has less negative impact at runtime.

I will commit this in a few days if no response.

Eric Wong normalperson@yhbt.net wrote:

normalperson@yhbt.net wrote:

https://bugs.ruby-lang.org/issues/13355

I will commit this in a few days if no response.

Caught during self review:

• Use more descriptive variable names (avoid potential "send" conflict)

• s/iseq_add_mark_object_compile_time/iseq_add_mark_object/
  since the range lifetime is tied to the overall iseq lifetime.

Will commit with following changes (after tests run...)

--- a/compile.c
+++ b/compile.c
@@ -2162,12 +2162,12 @@ iseq_peephole_optimize(rb_iseq_t *iseq, LINK_ELEMENT *list, const int do_tailcal
IS_INSN_ID(end, putstring) &&
(beg = (INSN *)get_prev_insn(end)) != 0 &&
IS_INSN_ID(beg, putstring)) {

•  VALUE sbeg = OPERAND_AT(beg, 0); 

•  VALUE send = OPERAND_AT(end, 0); 

•  VALUE str_beg = OPERAND_AT(beg, 0); 

•  VALUE str_end = OPERAND_AT(end, 0); int excl = FIX2INT(OPERAND_AT(range, 0)); 

•  VALUE lit_range = rb_range_new(sbeg, send, excl); 

•  VALUE lit_range = rb_range_new(str_beg, str_end, excl); 

•  iseq_add_mark_object_compile_time(iseq, lit_range); 

•  iseq_add_mark_object(iseq, lit_range); REMOVE_ELEM(&beg->link); REMOVE_ELEM(&end->link); range->insn_id = BIN(putobject); 

Eric Wong normalperson@yhbt.net wrote:

• s/iseq_add_mark_object_compile_time/iseq_add_mark_object/
  since the range lifetime is tied to the overall iseq lifetime.

Brrr.. I was wrong the second time :x

iseq_add_mark_object is called in iseq_set_sequence later on,
so we only need to call iseq_add_mark_object_compile_time
in iseq_peephole_optimize.