(possible suggestion) Speed up gisid if the dataset is already sorted about stata-gtools

[mcaceresb]

The speed gains are not there for me, so I can't assume it will be faster in every version of Stata.

I have an internal check to see if it's sorted for hashsort. Basically a recursive call to

for (i = start + 1; i < end; i++)
    if ( comp(i - 1, i) > 0 ) return (0);
return (1);

So if the previous element is greater, it is not sorted. If all elements are s.t. i - 1 <= i then it is sorted. I think I can modify it for isid:

for (i = start + 1; i < end; i++)
    if ( (rc = comp(i, i - 1)) <=  0 ) return (rc);
return (1);

If 1 then this is an id because every element is such that i > i - 1; if 0 then this is not an id because there exists an element such that i == i - 1. If -1 then this is not sorted and I can proceed with the hash normally.

[mcaceresb]

I just tried a version on this on Linux and had a 5x speedup. I'll write tests to make sure I'm not misidentifying when a set of variable is or not an id

[sergiocorreia]

Cool! Let me know if you want me to run any benchmark,

[mcaceresb]

I haven't pushed. I need to write tests to make sure I haven't mis-coded anything. I'll write those tomo and let you know to see if the speedup holds on your end.

[mcaceresb]

I have added the changes. Let me know how it benchmarks.

[sergiocorreia]

Yep, much faster!

gisid is now at half the speed of the gen + assert trick (which is 3x faster than gisid on the master branch, which is 4x faster than fisid, which is 5-10x faster than isid)

On the past I rarely used isid because it was slow, so the new way of doing it is a game changer in my opinion., as I can check almost instantly if I have unique IDs or not.