Tests integer overflow on 32-bit

[simonbyrne]

This line has an integer overflow on 32-bit machines, causing tests to fail.

cf https://github.com/JuliaLang/julia/issues/18741

Probably a good idea to switch on Appveyor, as that seems to be the most reliable way to test 32-bit builds

[StephenVavasis]

I am the author of that code. It sets up a large "Sieve of Eratosthenes" using SortedSets and SortedDicts to exercise the code for adding and deleting entries in BalancedTree32.jl. This code in BalancedTree23.jl has many branches because there are many cases; the leaf could be a 2-node; it could be a 3-node; there could be 3-nodes all the way to the top of the search tree, etc. I wrote my own code coverage tool (https://github.com/StephenVavasis/microcoverage) that checks coverage of every branch (i.e., every operand of ? :, && and ||) of every expression of every statement, and I discovered that I need a fairly large sieve in order to hit all of the branches. I never tested this code on a 32-bit machine.

So I propose as a fix: Line 345 of the indicated file should be replaced by:

N = myplatform_is_32_bit? 500 : 5000

Line 562 may need a similar patch.

I don't know the correct way to code the predicate: "my_platform_is_32_bit".

On a related note, is there a "standard" way to test coverage of every branch of the ?:, && and || operands?

[simonbyrne]

I don't know the correct way to code the predicate: "my_platform_is_32_bit".

Either Sys.WORD_SIZE == 32 or Int == Int32 should work.