Re: Question about the condition in `luaH_getint` function

[Xmilia Hermit <xmilia.hermit@...>]

Roberto Ierusalimschy:

> It is there because 'luaH_realasize' can be
> a little expensive to compute, so the redundancy avoids that call
> in a quite common case.
Wouldn't it be better to use a more efficient test such as the following 
instead?

!isrealasize(t) && ((l_castS2U(key)-1u) & 
~(((lua_Unsigned)t->alimit)-1u)) < t->alimit

This test does some bit hacks to test that key-1 is smaller than the 
next power of two of the limit.
This is done by masking of bit p from (key-1) through ~(t->alimit-1) as 
p is chosen such that it is the integer for which 2^(p+1) >= t->alimit > 
2^p holds.
This results in (key-1) & ~(t->alimit-1)  staying larger or equal to 
2^(p+1) when key-1 >= 2^(p+1) and getting smaller than 2^p when key-1 < 
2^(p+1) as bit p is masked off through ~(t->alimit-1) allowing t->alimit 
being used as the compare value.

Regards,
Xmilia