fix startswith/endswith implementation

[bkamins]

This PR implements a check that prefix b does not match partial character in a in startswith(a, b).

[nalimilan]

@bkamins knows better than I do whether it's correct. For a serious review better ask @StefanKarpinski.

[nalimilan]

Missing space for alignment. Would probably be more readable as a if.

[bkamins]

rewritten to use if now.

[bkamins]

The logic is that b is a prefix of a if next valid index in a is just after ncodeunits(b) codeunits in string a. The reason is that we know that in b position ncodeunits(b) is the end of a character in b (but it does not have to be the end in a in general so we have to test for this).

[vtjnash]

I don't understand the original TODO: what's wrong with doing the naive byte-comparison and declaring success if they match?

[bkamins]

I guess the original TODO says that it is not implemented yet 😄.
However given your question the answer is the following.
Consider strings:

x = "α"
y = "\xb1"

Now endswith(x, y) is false currently and it is a correct information.

However if you compared bytes you would say that there is a match because codeunit(x, 2) is the same as codeunit(y, 1).

[bkamins]

I have added endswith implementation and extended both to handle SubString{String}.

[bkamins]

Is this fix good to merge?

[StefanKarpinski]

It seems like way too much of a performance hit, although these particular functions are usually not performance-critical – #26921 is more of a problem. These both need the same fast primitive: a memcmpx function that returns the index of the first byte where they differ.

[bkamins]

Actually here the performance hit is much smaller than in #26921 and cannot be avoided (that is why here I wanted to merge, and I am waiting with the other one).
The reason is that where actually we can safely use memcmp (as I do now) and the only thing we have to check is if last characters in both strings are the same, which we have to do anyway (and memcmpx will not allow us to skip this test).

[StefanKarpinski]

Ok, fair enough. Let's merge this one then.