COBS: off-by one error with encoding/decoding lengths

1

Hi Blues team!

I ran into a subtle issue with the COBS codec length functions that causes max_decoded_length(max_encoded_length(X)) to sometimes return X - 1 instead of X.

The issue:

When X = 65536:

  • NoteBinaryCodecMaxEncodedLength(65536) returns 65796

  • NoteBinaryCodecMaxDecodedLength(65796) returns 65535 (not 65536)

This caused our code to reject valid 64KB chunks since it didn’t think the buffer was large enough.

Root cause:

In n_cobs.c, _cobsGuaranteedFit() computes overhead based on the encoded buffer size:

uint32_t cobsOverhead = 1 + (bufLen / 254) + COBS_EOP_OVERHEAD;

But 65796 / 254 = 259, while 65536 / 254 = 258. The extra overhead bytes push us into the next 254-byte bucket, causing a 1-byte discrepancy.

Suggested fix:

Use the algebraic inverse instead:

  uint32_t _cobsGuaranteedFit(uint32_t bufLen) {

      uint32_t fixed_overhead = 1 + COBS_EOP_OVERHEAD;

      if (bufLen <= fixed_overhead) return 0;

      return ((bufLen - fixed_overhead) * 254) / 255;

  }

This ensures the round-trip property holds for all valid inputs.

1 Like
3

Hi @devElert ,

I’m reviewing this today.

My plan is to:

  • repro your observation
  • create a test to catch it
  • fix it
  • merge the change
  • add it to the next release
  • release

Your issue seems to be clear and well documented, so I don’t anticipate needing anything further, but I’ll let you know if I have any additional questions or observations along the way.

Thank you for such a detailed/clear report,

Zak