Using is_uniform to avoid repeated checks

[mmatera]

Now we have the is_uniform attribute, this PR uses it to make faster the match test with SequenceBlank and SequenceNullBlank. Also, ensure that this property be computed and stored in list/expression creation operations (Table, Sum, Product).

• using is_uniform property in matching SequenceBlank and SequenceNullBlank patterns
• get_result now accepts an optional attribute is_uniform to set this property in its result.
• Table, Sum and Product now set the element_property is_uniform to provide a faster evaluation of large lists.
• convert_expression_elements now handles is_uniform

[rocky]

LGTM, although I confess I don't see how this is going to help. If you want to give a specific example, I'd be grateful.

[mmatera]

@rocky, suppose that you define a function
F[vals__Real]=Sum[x,{x,vals}]
and a list of values A=Table[BesselJ[3,x],{x,0,10,1000}]. Now, to evaluate

F[A]

first you need to check if the result of Table[BesselJ[3,x],{x,0,10,1000}] has all real entries. So, the match function (as is in master) has to check the head of each element in A. But, by construction, A has all their elements of the same type. So check the first is enough.
Same happens with the evaluation: since A is uniform, when you evaluate it, it is not necesary to check for upvalues on each element, you need just to look at the first one.

In this way,
Table[A=Table[1.*i^2,{i,0,n}]; {n, Timing[MatchQ[A, {__Real}]]},{n,{2,10,100,10000}}]
should show that the time required by MatchQ should be constant instead of linear in the size of A.

However, for some reason, it seems we are missing the element_properties after the assigment, so the improvement is not shown.

[rocky]

@mmatera Thanks for the explanaition.