Fix bugs due to empty set in wildcard restriction and + signs

[jodavies]

This PR fixes #554 and #599 .

The behaviour that I noted at the end of #554 (comment) is intended. When creating the {} sets, it finds that they exist already and doesn't create a copy:
https://github.com/vermaseren/form/blob/8abec49f876deae1d13ad159519b301f0b140dc5/sources/names.c#L2471-L2474

Thus the following defines only xs and xs2:

#-
off stats;
On names;
CF f;
S x,y;
set xs: ;
set xs2: 1,2;
L F = f(1)+f(2)+f(3);
id f(x?{}) = x;
id f(y?{1,2}) = x^y;
print;
.end

There is one point for discussion: the second commit implies that id f(x?!{}) = x; also does not match. In principle one could claim that x is indeed not a member of the empty set, so that this should match, but to my knowledge there will never be a case where x? does match an empty set (x? can not be "nothing" in FORM currently) so it feels a bit odd to match in this case to me.

One could even claim that x?{} could be a syntax error and terminate. In that case the x?!{} question is irrelevant.

Any thoughts?

[coveralls]

coverage: 50.819% (+0.08%) from 50.74%
when pulling 160f9d4 on jodavies:fix-sets
into 1d5e40b on vermaseren:master.

[tueda]

In principle, the set for the pattern matching can be changed at compile-time, for example,

CF f,g;
S x;

#$x = f(1,3,5);

#inside $x
  if (match(f(?a$a)));
  endif;
#endinside

L F = f(1,2,3,4,5,6);
L G = g(1,2,3,4,5,6);

repeat id f(?a,x? {`$a',},?b) = x * f(?a,?b);
repeat id g(?a,x?!{`$a',},?b) = x * g(?a,?b);

P;
.end

I think it should work even if $x = f() such that {`$a',} is expanded as an empty set.

[jodavies]

Do you mean that you would like all of the args of g to match !{} for $x=f()?

Otherwise, this example currently "works as intended" for $x = f();.

[tueda]

Do you mean that you would like all of the args of g to match !{} for $x=f()?

Yes, I think this is the natural interpretation of this pattern, x?!{} (x_?(Not[MemberQ[{}, #]]&) in the Wolfram language).

[jodavies]

OK. It is not quite as simple as I thought (just goto NoMnot;) because the code at the NoM label wants to set AN.oldvalue = w[3]; and here w=0. Probably one needs AN.oldtype = old2; AN.oldvalue = oldval; goto NoMatch; as happens also further down in the set restrictions code. It looks like oldval = w[3]; is set at TestSet: for this purpose.

Generally it is not clear in what circumstances various "old" values need to be put back or not.

[jodavies]

Rebased, and implements the match for !{} with oldtype etc put back as I understand it.

[tueda]

Looks fine.