Re: [bug-gawk] FPAT bug?

[Manuel Collado]

El 02/04/2017 a las 15:55, Andrew J. Schorr escribió:
> On Sat, Apr 01, 2017 at 04:27:46PM -0500, Ed Morton wrote:
> > Might be useful until a fix is available. Any better ideas for now?
>
> I have no better ideas, although I can't imagine that such a solution would
> please anyone. Here's a crazy question: do you ever use FPAT for anything other
> than CSV parsing? If gawk had a proper CSV parser library, would we need FPAT
> at all? I don't think I have ever used FPAT (my personal CSV parsing library
> takes a brute force string scanning approach), and it seems from the observed
> bugginess that FPAT is not often used...

Any FPAT regex that matches null strings is inherently ambiguous. It 
could provide any number of null fields at any position. So FPAT must be 
extended with some implicit restrictions on field separators.

I assume that the most intuitive approach if to allow null field 
separators only between contiguous non-null fields. I assume that FPAT 
and patsplit() implementations try to achieve this policy, but may fail 
in some cases.

But even the above is not enough. What about a null record? Should it 
provide a single null field or no fields at all?

So, what is the precise unambiguous specification of the FPAT split 
feature, if it allows null fields?

Hope this helps.
--
Manuel Collado - http://lml.ls.fi.upm.es/~mcollado