SELECT POSITION(('å' COLLATE "en_US.utf8") IN 'yeah'); -- should return 3 instead of 0 !?! SELECT POSITION(('o' COLLATE "en_US.utf8") IN 'ångström'); -- should return 7 instead of 0 !?!
==> up to here, this seems pretty enough to conclude that POSITION() doesn't care at all about COLLATION and always perform a byte search.
IIUC "Performing a byte search" is what happens when you use a deterministic collation where the only test is for equality. So your examples are not useful in distinguishing whether a collation agnostic byte search or a deterministic collation search is happening.
I would like to have something in the doc about that... i.e. either some examples showing how the COLLATION is impacting the results of the POSITION() function
I don't disagree, but lack of documentation regarding string functions and collations is not just limited to the position function. I find the fact the word collation doesn't appear anywhere on the string functions page to be potentially worthy of change.
How collations behave is documented, in particular:
"A collation is either deterministic or nondeterministic. A deterministic collation uses deterministic comparisons, which means that it considers strings to be equal only if they consist of the same byte sequence."
I'll admit I'm definitely an unfavorably biased observer here and don't deal with these nuances on a daily basis. I have the general impression that our documentation is correct and sufficient but could be made more user-friendly. Updating a single function doesn't do that though and in some ways makes things worse when other related elements, and the presentation of the material as a whole, doesn't go along with that single change. Based on this, and the above observation about your test cases, I don't see much motivation for change here. The effort seems to outweigh the reward. But for someone who feels differently and submits a patch there is, IMO, room enough for improvement that a well-written one is likely to be welcomed.
