Re: [bug-gawk] FPAT bug?

[Manuel Collado]

El 03/04/2017 a las 8:29, address@hidden escribió:
> [...]
> I will get around to investigating the issue with FPAT.

Perhaps a problem with the FPAT feature is how the parsing algorithm is 
structured. Thinking about it, I feel there is a simpler way to 
structure the code.

Attached is what can be an awk reference implementation of the FPAT 
parser. It differs from the current implementation in some ways:

- Each iteration of the parsing loop identifies exactly one
      (separator[n-1],field[n])
  pair. The last iteration may deliver just a final separator.
- Each (separator/field) pair must consume some input characters.
  Except for the first (sep[0]/field[1]) that can be both nulls.
- So a null field is only valid as a first field or after
  a non-null separator.

This avoid the use of flags to remember some state from the previous 
loop iteration.

The attached code has been tested with the (adapted) set of fpat*.awk 
tests, and some more.

Please advice if it is worth the pain of trying to restructure the 
current code, or if there is a simpler way to fix the FPAT bug.

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

refpatsplit.awk
Description: Text document