Re: Optional static types for Lua - experimental derivative Ravi

[Richard Hundt <richardhundt@...>]

>>
>> > Well the objectives are very different I think. I am looking to
>> > improve efficiency of certain operations. So the static typing is for
>> > that purpose. Therefore the features I intend to support will be a
>> > subset of above. For example, as of now I have no intention to provide
>> > syntactic support for classes or interfaces. Also no concept of union
>> > types.
>>
>> Ah, okay, hence the additional ops. That's got me thinking: wouldn't
>> it be feasible to do run-time instruction specialization without the
>> need to add syntax (so you'd patch in your new ops dynamically)? If it
>> worked though, you could run vanilla Lua modules and perhaps get a
>> performance boost. Maybe that's a ton of work though without
>> substantial gains because recording, narrowing/widening overhead might
>> be too high to be beneficial. Dunno. Just an idle thought.
>
>
> Do you mean something like this?
>
> "Optimizing Lua using run-time type specialization"
> https://www.complang.tuwien.ac.at/anton/praktika-fertig/schroeder/thesis.pdf

Yep, exactly like that :)

It's pretty cool seeing that someone's already tried it. The gains
seem to be minimal though and platform and compiler dependent, which I
suppose is a nod to Lua's performance. Having said that, 5.3 might
benefit more from specialization than previous versions as used in
that paper (nice find btw, and thanks for pointing me at Typed Lua in
the first place).