I don’t think this is the best value to return if matchCharacters is empty. I think we want to return std::min(startOffset, source.size()) rather than 0. Please add test cases for this, and if you think 0 is better, please explain why.

But I might suggest that if startOffset is > source.size() we should return notFound, even if we are looking for the empty string. Easiest way to do that is probably to remove the special check for empty string entirely (see my suggestion below). But another way would be to write something like this:

I don’t think we should do this checking by preflighting. It’s too easy to get it wrong when the check is different from the code below that relies on the check. Instead I think we should incorporate the check into the code below, and take advantage of the clamping that is already done by the subspan function rather than redoing it.

We could either use checkedDifference to compute delta, or just write:

if (startSearchedCharacters.size() < matchCharacters.size())
    return notFound;

But an even better way to write this that I think would be clearer is to eliminate startSearchedCharacters and delta entirely, and not add an explicit check of the length:

for (size_t offset = startOffset; offset < source.size(); ++offset) {
    if (equalIgnoringASCIICaseWithLength(source.subspan(offset), matchCharacters, matchCharacters.size())
        return offset;
}
return notFound;

I don’t see any downside to writing it this more straightforward way, and if we write enough test cases we can make sure this new version is correct. I think we could also consider eliminating the explicit check for empty matchCharacters and just write this:

for (size_t offset = startOffset; offset <= source.size(); ++offset) {
    if (equalIgnoringASCIICaseWithLength(source.subspan(offset), matchCharacters, matchCharacters.size())
        return offset;
}
return notFound;

I’m pretty sure this handles edge cases like startOffset > source.size() correctly without adding any preflights or checked arithmetic, and I would prefer this version.

I think I managed to change it from "pre-flight" to "in-flight". All tests pass.

Why do we need to add this? What test fails if we don’t add it?

None.

I removed that part.

Why is this check needed? This is already what memem does when the needle size is greater than the haystack size. Did some test indicate that was not the case on some platform?

It looks like I can remove this part and leave only the check above.

I personally like it better without the parentheses: