Re: [office-formula] Spec tweaks

["David A. Wheeler" <dwheeler@dwheeler.com>]

robert_weir:
> General question -- has anyone done a formal analysis of our expression 
> language?

I created a flex/yacc model of an older version of the grammar, and
I did a hand-analysis of the current one.  That's when I discovered
that unquoted sheet names are fine for the usual case, but if you have
subtables that are sheet names, you need to quote them.

>  Especially with recent tweaks and extensions, it would be great 
> to throw this all into LEX/YAAC (or whatever tool is used for this these 
> days) and confirm that we're not getting conflicts.

I agree, that'd be a great thing.

If anyone wants to do that, feel free to yacc/lex stuff at:
 http://www.dwheeler.com/openformula
(MIT license, do what you want to with it.)

I might do that if I don't hear from someone else soon.

Also, before we set the spec in stone I want to hear from implementors
who've already worked to implment it, anyway.

> Also, did we have any explicit target for the complexity of grammar we 
> wanted, whether LALR, LR(k), GLR, or whatever? 

Sure, "must be implementable by a Turing machine".
Not NP-Hard is a bonus :-).   Ok, seriously...

We didn't really state a particular complexity goal, but in practice I've
tried to keep this to 1-symbol lookahead.  E.G.:
* If you see an Identifier, the next char is either '(' and we have a function,
   or not and we have a named expression.
* In several cases '...' can mean more than one thing... but the NEXT
   non-whitespace character always disambiguates.

In some cases, just seeing one character is enough to tell you what's next -
that's particularly true for cell references, which are heavily used.

--- David A. Wheeler