Re: [saag] ASN.1 vs. DER Encoding

[Nico Williams <[e-mail]>]

On Wed, Mar 27, 2019 at 09:49:09AM +0000, Peter Gutmann wrote:
> Nico Williams <[e-mail]> writes:
> 
> >Something similar can be said of SSHv2's encoding rules.
> 
> The ones that arbitrarily mix binary data and comma-delimited text strings?

You can do that in ASN.1 too if you want...  That was just lack of
discipline in using the features of the encoding.  I agree that it's
awful.

> Of all the encodings I've had to write parsers for, SSH's one scares me the
> most.

I've seen terrible DER hand-rolled codecs...  It's hard to say which is
worse for hand-coding.  But ad-hoc, informal designs like TLS' or SSH's
preclude code generation, so in that sense they are easily the worst.

XDR is one that's in the middle: you get a formal language, and the
encoding is trivial enough that you can hand-roll it with very good
chances of getting it right *if* you have suitable library.

(In Heimdal we have the krb5_{storage,store,ret}_*() functions, a subset
of which gives you XDR, and which make it easy to write safe codecs.
These should really be renamed heim_*() and moved from libkrb5 to
libheimbase, where they can do more good outside krb5 context.)

> It's also the one where I've found the most crashes in the other
> system while developing the code and getting various bits slightly
> wrong, e.g. sending "foo," or ",," as a value, or getting a binary
> length field slightly wrong in combination with text-string data.

I believe that.