Always include long long Make_fixed_precision

[jrtc27]

One of the users of this function is getStatInfo in unix_specific, which
converts a struct stat's st_nlink. However, on x32, st_nlink is a 64-bit
quantity (the system call ABI is 64-bit), so there are two problems.
Firstly, it will be truncated; secondly, since the long and int
overloads involve a cast to an integer of the same width, compilation
fails with an ambiguous call.

[dcjm]

Thanks for investigating this. I think the #if tests should be changed to
#if (SIZEOF_LONG_LONG != 0)
otherwise compilation could fail if long long isn't present. Perhaps that's not possible.

[jrtc27]

Oh I see, that's what's done for arb.h. Is there a reason why you're checking SIZEOF_LONG_LONG rather than HAVE_LONG_LONG?

[dcjm]

I don't know why it's checking SIZEOF_LONG_LONG. I've changed that in both arb.h/arb.cpp and run_time.h. Has this fixed the problem?

[jrtc27]

I will check. I noticed from that commit that you still have the #if SIZEOF_LONG_LONG <= SIZEOF_VOIDP around Make_arbitrary_precision in arb.cpp but not in arb.h. Is there a reason why that's there?

[dcjm]

Because there are different implementations depending on whether a long long value will fit in a machine word or not.

[jrtc27]

Oh right, yes, I should have looked for context. My bad.

[jrtc27]

I finally got round to trying this and your fix seems to have worked, as expected; thanks!