Re: comments on draft-ietf-sming-reqs-03.txt

[Frank Strauss <strauss@ibr.cs.tu-bs.de>]

Hi!

>> The semicolon is of course not a requirement. But in my opinion the
>> prerequisites for an easy and unambiguous development of parsers is a
>> requirement, since we are faced these days with lots of buggy SMI
>> modules due to the fact that for a long time of more than 5 years
>> developers didn't have MIB compilers of that precision that we are all
>> accustomed to by C and Java compilers. I don't want to repeat such a
>> pain.

Jon> Frank, on this specific point, I think we still disagree. The
Jon> problems with [bad] SMI modules is that they are far less the
Jon> result of buggy compilers and much more the result of poor design
Jon> and implementation. This is due in part to the junior level of
Jon> people that work on them in most companies and the relatively low
Jon> priority the work receives. A better compiler does not solve
Jon> much.

I agree in the sense, that a single better compiler does not make all
the error tolerant compilers vanish. That's why some companies today
still claim `It works for us. You have to use our MIB compiler to make
it work for you as well.' even though there is SMICng and libsmi and
maybe other good compilers. With a new language (no matter to which
degree it is `new'), we have the chance to define the grammar and
semantics more precisely and formally.  I have the feeling that we
agree on defining the grammar in ABNF, and that's one of the things
that look really important to me.

 -frank