#s semantics (was Re: Lua 5.1-work6 cumulative PATCH #1)

[Mike Pall <mikelu-0508@...>]

Hi,

Roberto Ierusalimschy wrote:
> Except for the "#s performance improvement", all those corrections are
> incorporated in the next release;

Ok, then here is the official plea to change these semantics:
I'd like #s to be tied down in the core, just like #t is.
I do not see a point in being able to override it. I explained
my rationale for this back then in the mailing list thread after
the announcement.

This is really more about consistency than just performance.
When I read some random binary stuff from a file and pass it
around as a string, then I definitely want *s to return the
length in bytes and not some strange other measure.

And about performance: you can bet on it that everyone will
use #s instead of string.len(s) in the future and then wonder
why this is slower. Bad PR.

I have one additional rationale now: It's exceptionally easy
for a JIT engine optimizer to infer that #s is a string length
operation and this inlines to only 2 instructions. But it's
exceptionally difficult (or very unsafe) to do so if the
behaviour can be changed dynamically.

IMHO if anyone wants different string 'lengths' (# of chars,
# of glyphs etc.) they can provide extra function calls or
extra types. But #s should really consistently return the
string 'size' aka 'length in bytes'.

Bye,
     Mike