Shim memory map functions

[nguerrera]

System.IO.MemoryMappedFiles and System.Security.SecureString now have common Unix builds.

cc @sokket @stephentoub

[stephentoub]

If we've decided to use uintptr_t / UIntPtr for size_t, can we just do so in our signatures on both the managed and native sides and not do the "using size_t = System.UIntPtr;" thing?

[nguerrera]

I'm still a bit torn on my size_t approach and considering going to ulong/uint64_t everywhere.

Right now, it's actually declared as size_t in the native code and not uintptr_t, but there's a static assert to check that size_t is what we expect.

UIntPtr on managed side is a pain (no arithmetic, comparison operators, etc.) On native side, if we used uint64_t, we'd have to do a few < SIZE_MAX checks (or have debt to do so when x86 comes online), but it would make the managed code easier to maintain I think.

I guess there are the following options:

(a) Managed=UIntPtr, Native=size_t, static assert that sizeof(size_t) == sizeof(uintptr_t)
(b) Managed=UIntPtr, Native=uintptr_t, static assert that sizeof(size_t) == sizeof(uintptr_t)
(c) Managed=ulong/Native=uint64_t, native fails calls if it sees size > SIZE_MAX

(a) is what I have now, but (c) is where I'm leaning.

What do you think?

[jonmill]

I agree with (c)...I've been converting size_t (and other types like it) to long types; it seems much cleaner to me and makes an easy contract.

[stephentoub]

I like (c) best as well, followed by (b). I'd prefer not (a).

[nguerrera]

I couldn't repro the ftruncate failure on OS X, but I left the refactoring I did to make it easier to add more checks.

Also, self nit: 'work well with' (will fix before squashing).

[nguerrera]

I switched everything to ulong/uint64_t, but haven't added any range checks yet. This means things can technically overflow on 32-bit, but generally the managed code is passing in things that fit in uint.MaxValue or else checking IntPtr.Size before using them regardless. Still we should be more careful. I filed #3165 to make sure we catch narrowing implicit conversions and I'll build things on x86 with that on when I fix it to add appropriate range checks.

[nguerrera]

@stephentoub @socket anything else? I left my fixups as separate commits to see delta between iterations, but I'll squash them before merging.

[stephentoub]

LGTM