Check properly for permutations whose degree would exceed the internal limit. Document the limit.

[stevelinton]

No description provided.

[stevelinton]

Addresses #575

[ChrisJefferson]

My only comment would be are we sure we don't want to limit PERM4s to 2<<28 in both 32-bit and 64-bit. How confident are we that most code works with perms between 2<<28 and 2<<32?

[stevelinton]

Confident up to 2^31-1. The only thing that would break at 2^28 would be INTOBJ_INT in 32 bit.

Above 2^31-1 there might be a signed/unsigned mixup somewhere. I'll try and do some tests on babbage.

[frankluebeck]

Looks fine to me, thanks.

In the documentation I would use <M>2^{28}-1</M> instead of <M>2^{{28}}-1</M> such that the braces vanish in the text version of that help page (2^28-1 instead of 2^(28)-1), but that is a matter of taste.

[stevelinton]

I have done a few basic checks for permutations of degree close to 2^32-1 and they seem to work OK. Not something we can reasonably add to the test suite yet.

@frankluebeck -- fixed the formatting

[ChrisJefferson]

I tried poking this and I couldn't break it, also couldn't come up with anything to add to a test that wouldn't break anyone with < 4GB workspaces.

Therefore, I'll merge this patch for now.