Add aliases in AbstractMemory for size_t and ssize_t.

[Kerilk]

Hello,

This is a feature I end up re-implementing in most of my gems, so I thought maybe it warranted to be here.
This implements read_size_t and read_ssize_t and their counterparts in AbstractMemory.

The proposal is rather straightforward, but some aspects may warrant discussion:

• I kept the _t suffix as I didn't want any confusion between get_size and size methods,
• I do this at runtime, but as sizeof(size_t) is a compile time constant it shouldn't matter,
• This only support size of 32 and 64 bits, but I think I noticed this assumption in several places in the code,
• I refactored the ALIAS macro, I avoided using C99 argument macro with no token for maximum portability, but it may feel clunky.

This would also close #537

Thanks

[eregon]

Just passing by, this would need to be tested.

[Kerilk]

@eregon I though so as well, but I couldn't find where the other aliases are tested. Maybe they are and I just missed it.

[eregon]

struct_spec.rb has related tests

[Kerilk]

@eregon I ended up adding tests to pointer_spec.rb as write_array_of_type and read_array_of_type were somewhat tested there. I hope it is fine.
Truffle ruby seems to be getting those methods from https://github.com/oracle/truffleruby/blob/master/src/main/ruby/truffleruby/core/truffle/ffi/pointer_access.rb so these would have to be added there as well.

[eregon]

I just realized, actually this is already possible with the read/write/get/put methods:

irb(main):008:0> ptr = FFI::MemoryPointer.new(100)
=> #<FFI::MemoryPointer address=0x0000000001deb6c0 size=100>
irb(main):009:0> ptr.write(:size_t, 42)
=> nil
irb(main):010:0> ptr.read(:size_t)
=> 42

So I think we should not add new methods, because it's already easily possible via a general way.
Adding new methods means more time to define them (during loading), increased footprint and overhead for all 3 backends to support it (C ext, JRuby, TruffleRuby).

I'm sorry I didn't think of that earlier, would have spared you some effort.

[Kerilk]

For the sake of the argument: size_t is a native C type that is encountered very frequently, and is platform specific, so you need a bit of logic to define those aliases in your code. Especially if you want to read/write array of those as the reader/writer does not exist:
p.read_array_of_type(:size_t, XXXXX, 3)

So I agree with you in general but I think size_t should be covered the same as int or long or pointer or float or double.

Edit: Maybe ssize_t could be replaced by ptrdiff_t, but I prefer the symmetry here.

[eregon]

I think it'd be good to add something like ptr.read_array(type, length) (and write_array).
The existing read_array_of_type feels clunky due to specifying both the type and the method to read/write (redundant arguments).