RE: [Bastard-libdisasm] Data kept for invalid instructions

[<kri...@as...>]

> real instruction could start anywhere in that range. A 1-byte invalid
> insn could at least be useful when dealing with off-by-one anti-disasm
> code [though even then, only if it is improperly written].

gcc 3.3.5 compiles pipe_read from linux 2.4.27 with an invalid opcode somewhere
in between:

pipe_read:
...
 1a4:   0f 84 be fe ff ff       je     0x68
 1aa:   0f 0b                   ud2a
 1ac:   78 00                   js     0x1ae
 1ae:   20 84 22 c0 e9 b1 fe    and    %al,0xfeb1e9c0(%edx)
 1b5:   ff                      (bad)
 1b6:   ff 90 8d 74 26 00       call   *0x26748d(%eax)
 1bc:   b9 01 00 00 00          mov    $0x1,%ecx

Could be somewhat related to their BUG() macro and mixing inline assembly with C
source. Improperly generated, sure, but still it happens. Oh well...