Decimal representation for optional mantissa with non-constant exponent about fix-simple-binary-encoding

[donmendelson]

The spec currently says:

For optional fields, a special mantissa value is used to indicate that a field value is null.

For efficiency, the value of the exponent need not be checked by a decoder if the mantissa is null.

However, I don't like the fact that the value of the exponent is non-deterministic. In my opinion, all values should be deterministic. That is, given the same semantic values, two encoders should produce exactly the same buffer. Also, failing to overwrite previous values in a buffer could leak information and pose a security risk. So it should always write something. I would be in favor of changing the spec to say that the exponent should also be set null. Do others agree?

[kleihan]

Agree. Did you mean to write "However, I don't like the fact that the value of the exponent is non-deterministic."?

[donmendelson]

Thanks, Hanno. Corrected.

[RFrenkel]

Thank you, Don and Hanno!

So i take it the best practice is to set exponent also to null =127.

[donmendelson]

I suggest that we edit the spec to say that this is recommended practice.

(BTW the spec says that character arrays should be padded with nulls for the same reason--a deterministic output.)

[RFrenkel]

Great, thank you!