replacing MIT's ASN.1 code

[Ken Raeburn]

On Oct 15, 2007, at 19:39, Ezra Peisach wrote:
> While I do not really have a say in any of this I see the following  
> concerns:

Why not?  You've been contributing for years.

> a) There are some issues of signed/unsigned that are present in the  
> krb5 specs and implementations... For instance the nonce is  
> supposed to be unsigned - and we are encoding as a signed  
> integer... If the high bit is set - this is encoded as a five byte  
> integer - which heimdal cannot handle.
> There is one other discrepancy in signed/unsigned  handling - but I  
> cannot remember what it is.  I think it might be kvno.
>
> This means we will need to be careful in how things are encoded in  
> terms of interop.

Yes, I wouldn't be the least bit surprised if the ASN.1 we feed to  
the compiler is different from that in RFC 4120, for just this reason  
(and the incorrect older implementations, as you mention).

> c) We do have a  test suite for asn1 encodings/decodings - so we  
> will test some edge cases - like indefinite encodings, etc. If  
> these work w/o memory leaks, etc - that would help validate the ASN. 
> 1 implementation.

I'm actually a little concerned that our current code wouldn't stand  
up if we tried stress-testing it....

Ken