It seems that mrbc sometimes generates a broken bytecode file on Intel Mac.
After some debugging, I found that the problem is caused by the mrb_ro_data_p function.
It sometimes returns false-positive results, which leads to using a released memory area for symbols.
For this reason, some strings in the bytecode file are sometimes broken.
After some investigation, I found that constant strings in macOS are stored in the __TEXT segment, so it is probably enough to check that the pointer is in this segment (for some reason, it currently checks if the pointer is after the _TEXT segment).

Example:

puts "Hello, World!"

The debug section of the bytecode is broken (there is no file name):

00000000: 5249 5445 3033 3030 0000 007f 4d41 545a  RITE0300....MATZ
00000010: 3030 3030 4952 4550 0000 0042 3033 3030  0000IREP...B0300
00000020: 0000 0036 0001 0004 0000 0000 0000 000a  ...6............
00000030: 5102 002d 0100 0138 0169 0001 0000 0d48  Q..-...8.i.....H
00000040: 656c 6c6f 2c20 576f 726c 6421 0000 0100  ello, World!....
00000050: 0470 7574 7300 4442 4700 0000 0021 0001  .puts.DBG....!..
00000060: 0002 9005 0000 0013 0001 0000 0000 0000  ................
00000070: 0000 0002 0200 0145 4e44 0000 0000 08    .......END.....

Expected:

00000000: 5249 5445 3033 3030 0000 0081 4d41 545a  RITE0300....MATZ
00000010: 3030 3030 4952 4550 0000 0042 3033 3030  0000IREP...B0300
00000020: 0000 0036 0001 0004 0000 0000 0000 000a  ...6............
00000030: 5102 002d 0100 0138 0169 0001 0000 0d48  Q..-...8.i.....H
00000040: 656c 6c6f 2c20 576f 726c 6421 0000 0100  ello, World!....
00000050: 0470 7574 7300 4442 4700 0000 0023 0001  .puts.DBG....#..
00000060: 0004 312e 7262 0000 0013 0001 0000 0000  ..1.rb..........
00000070: 0000 0000 0002 0200 0145 4e44 0000 0000  .........END....
00000080: 08