It looks like ISO-2022-JP converters (all versions) will output C0 control code bytes when using fallbacks from the JIS X 0201 converter, which currently uses the ibm-897_P100-1995.ucm table which has fallbacks for PC control image look-alikes. This could be particularly problematic for those that fall back to the bytes that also encode SO/SI/ESC:
<UFFED> \x0E |1
<U263C> \x0F |1
<U21B5> \x1B |1
The best solution seems to be to hardcode the JIS X 0201 conversion in ucnv2022.c, omitting the fallbacks to C0 control code bytes. Otherwise, it's mostly the same as ASCII (except for 5C and 7E) and halfwidth Katakana, plus fallbacks for fullwidth ASCII. This is very easy to do in a few lines of code.
Then if JIS X 0201 is not needed as a separate converter, this .cnv table file can be removed.
