RE: [Firebird-devel] The metadata crash

[Ann W. H. <aha...@ib...>]

At 04:08 AM 4/22/2001 -0400, Claudio Valderrama C. wrote:

>Hmm, the only problem is that I see one point to put a flag to say "I
>already processed a default" without going into global variables. It has to
>be in the loop that modify_domain does in DDL.C and now the same issue: do
>we ignore a repeated option or do we shout?

Again, the logic belongs in DDL.C, where it should be very obvious
that there are two default nodes ... or whatever.  Parsers parse.
Lexers lex.  Both should be dumb as stumps.

I'm working on a white (well, dirty grey) paper on error handling
and will get an excerpt here soon.  Basically, you go into msgs.gdb
(I use qli), create a new message, then use codes.exe to generate
a new codes file.


> > > So do you say that we don't need to protect DYN just in
> > > case there're other syntax bugs in the this definition file?

Mike said:

> > Absolutely not. We should add sanity checks wherever they are reasonable.

I'm lost in the negatives.  DYN is a public interface.  Every public
interface has to expect bad input and handle it gracefully.  Well, it
should, even if lots of them don't.  DSQL shouldn't generate bad dyn,
but dyn shouldn't assume it's talking to a well-behaved subsystem.


Regards,

Ann
www.ibphoenix.com
We have answers.  Parsers parse.