Feature #18825: Specialized instruction for "array literal + `.hash`"

• Status: Rejected
• Assignee: -
• Target version: -

[tenderlovemaking (Aaron Patterson)]

Feature #18611 is merged. That feature encourages people to write hash methods like this:

def hash
  [@a, @b, @c].hash
end

I would like to add a specialized instruction for this case opt_newarray_hash. It's similar to opt_newarray_max and opt_newarray_min but for the hash method.

ISeqs before the optimization:

== disasm: #<ISeq:[e-mail]:1 (1,0)-(3,3)> (catch: FALSE)
0000 getinstancevariable          :@a, <is:0>                         (   2)[LiCa]
0003 getinstancevariable          :@b, <is:1>
0006 getinstancevariable          :@c, <is:2>
0009 newarray                     3
0011 opt_send_without_block       <callinfo!mid:hash, argc:0, ARGS_SIMPLE>, <callcache>
0014 leave  

ISeqs after the optimization:

== disasm: #<ISeq:[e-mail]:1 (1,0)-(3,3)> (catch: FALSE)
0000 getinstancevariable                    :@a, <is:0>               (   2)[LiCa]
0003 getinstancevariable                    :@b, <is:1>
0006 getinstancevariable                    :@c, <is:2>
0009 opt_newarray_hash                      3
0011 leave  

The new instruction allows us to avoid allocating a new array and also avoid pushing a stack frame.

The implementation is here, and I've also attached a patch.

[mame (Yusuke Endoh)]

Just an idea. Is it possible to introduce a general instruction like opt_newarray_send instead of individual opt_newarray_* instructions?

[tenderlovemaking (Aaron Patterson)]

mame (Yusuke Endoh) wrote in #note-1:

Just an idea. Is it possible to introduce a general instruction like opt_newarray_send instead of individual opt_newarray_* instructions?

It should be possible (IMO). But I guess it means the bytecode might be larger for opt_newarray_min and opt_newarray_max because we probably need to add a parameter for opt_newarray_send. Maybe it's not a big deal though. I will try making a patch.

[tenderlovemaking (Aaron Patterson)]

• Status changed from Open to Rejected

Closing in favor of #18897