#1466 New fit option "errors {x,y,...}" errors or segfaults

[Ethan Merritt]

I just noticed that the documentation has not a word to say about the
new "fit" options to handle columns containing data error estimates,
added back in Feb 2014.

https://sourceforge.net/p/gnuplot/patches/585/?page=1

The file fit.c itself contains the following comment:

• Bastian Maerkisch, Feb 2014: New syntax to specify errors. The new
• parameter 'errors' accepts a comma separated list of (dummy) variables
• to specify which (in-)dependent variable has associated errors. 'z' always
• denotes the indep. variable. 'noerrors' tells fit to use equal (1) weights
• for the fit. The new syntax removes the ambiguity between x:y:z:(1)
• and x:z:s. The old syntax is still accepted but deprecated.

But I don't think that comment is correct. I gather from looking at the code
and the fit demo that the expected syntax is
fit $DATA using <using-spec> {xerror | yerror | zerror | xyerror}

The keyword "errors" is also parsed, but nothing I put after that is accepted
so I obviously don't understand how to use this option. For instance
(from the demo)

# this works  
   fit fxy(x) $PearsonYork using 2:4:(sqrt(1./$3)):(sqrt(1./$5)) xyerr via a1xy, a2xy

# this doesn't work; clearly the variable accounting is bad
   fit fxy(x) $PearsonYork using 2:4:(sqrt(1./$3)):(sqrt(1./$5)) errors x,y via a1xy, a2xy
   >>> Not enough columns in using spec.  At least 7 are required for this error spec
   fit fxy(x) $PearsonYork using 2:4:(sqrt(1./$3)):(sqrt(1./$5)) errors x,y via a1xy, a2xy
   >>> Not enough columns in using spec.  At least 9 are required for this error spec

Clearly num_errors is not getting re-initialized to 0, but even if I add that to the
code I still can't make the "errors x,y" to work. It either gives a non-helpful message
or segfaults.

   fit fxy(x) $PearsonYork using 2:4:(sqrt(1./$3)):(sqrt(1./$5)) errors x,y via a1xy, a2xy
   >>> Not enough columns in using spec.  At least 13 are required for this error spec
   fit fxy(x) $PearsonYork using 2:4:(sqrt(1./$3)):(sqrt(1./$5)):(1) errors x,y via a1xy, a2xy
   Segmentation fault

I'll mark this as an unresolved issue in version 5.0.rc2 rather than holding it up,
but it would be nice to get a fix and some documentation.