There might be reasons to extend number parsing to support digits with variation selectors.
However, the status quo in ICU is that variation selectors terminate numbers in both number parsing (DecimalFormat class) and in numeric collation (Collator). Numeric collation is actually much more limited than number parsing, to strictly strings of digits, not including sign (thus only non-negative), decimal, exponent, etc. More processing in the bowels of the collation code would be very complicated, and ambiguous: "file-5.txt" is probably file number 5 rather than file minus five. markus
