vlq_base128_le incorrectrly compiled in C#

[mikhail-khalizev]

Hi.

Source file: common/vlq_base128_le.ksy

Look in screenshot below, it is source file compiled for C#.
Since by default csharp uses Int32 type, shifts by '35', '42', '49' computes incorrectly.

Possible workaround to cast values to u8 in vlq_base128_le.ksy:

  value:
    value: >-
      (groups[0].value.as<u8>
      + (len >= 2 ? (groups[1].value.as<u8> << 7) : 0)
      + (len >= 3 ? (groups[2].value.as<u8> << 14) : 0)
      + (len >= 4 ? (groups[3].value.as<u8> << 21) : 0)
      + (len >= 5 ? (groups[4].value.as<u8> << 28) : 0)
      + (len >= 6 ? (groups[5].value.as<u8> << 35) : 0)
      + (len >= 7 ? (groups[6].value.as<u8> << 42) : 0)
      + (len >= 8 ? (groups[7].value.as<u8> << 49) : 0)).as<u8>

[yav]

I just run into this issue as well, but when experimenting with the C++ backend. I can't seem to find any information on the type inference algorithm for Kaitai structs, so I am not sure if this is a bug in the compiler or the spec, but fixing it as suggested above seems reasonable to me.

The current status, where the spec looks simple, but when you try to parse some large numbers you get the wrong result is pretty unfortunate.

[generalmimon]

@yav:

I can't seem to find any information on the type inference algorithm for Kaitai structs

See kaitai-io/kaitai_struct#257 (comment).

but fixing it as suggested above seems reasonable to me.

The current status, where the spec looks simple, but when you try to parse some large numbers you get the wrong result is pretty unfortunate.

I agree. The workaround by @mikhail-khalizev looks solid to me; let me apply it...