"\N{LATIN SMALL LIGATURE IJ}" =~ /ij/i
matches, as the fold of the ligature is 'ij'. But if you simply add
capturing parentheses, as in this post's subject line, it becomes
somewhat nonsensical, as each captured group should match some part of
the indivisible character LATIN SMALL LIGATURE IJ. And the problem is
not restricted to ligatures, but when comparing using the NFD
normalizations of a string and pattern, the captured portion matched may
not be in the original string.
I didn't see any reference to this in the ICU documentation.
