The comment of between-code? >>27 identifies one case that yields false positives. It relies on a closing code marker immediately followed by a spoiler, which causes both m1 and m2 to match, and the 'or' will give m1 precedence over m2 in any dispute. False positives can also be obtained by putting any non-space character, such as punctuation, after a code segment:
==one==, ~~two~~, ==three==one, ~~two~~, three
This happens because both m1 and m2 look at the last potential marker but ignore everything before it, so they do not have enough information to decide whether what looks like an opening or closing marker actually is one. The solution is to replace m1 and m2 with a scan through s1 using irregex-search and (regex code) to skip valid code segments, then look for an opening marker after the last one.