Reduce the x86-specificity of the crate

[tgross35]

Currently a lot of things are gated on cfg(target_arch = "x86"), which is used to indicate systems with 32-bit words. Apply the following changes to make this less specific:

• Replace x86-32 configuration with configuration based on target_pointer_width.
• Where possible, replace #[cfg] with cfg! or eliminate the check, to increase the percentage of code that gets validated on any platform.
• Remove some configuration that was unneeded, for the same reason. This includes things like #[cfg(target_arch = "x86")] on comparisons to EXPONENT_MIN or EXPONENT_MAX. The checks are only useful on 32-bit platforms, but this is a trivial compiler optimization when these values are the same as Exponent::{MIN,MAX} so not much is gained by keeping the config.

I have verified that the crate builds on armv7 targets (but have not tested).

Fixes: #26

[tgross35]

Would something like from_raw_parts_u64 or a AsWords trait be useful/welcome? In order to allow using the same words on both 32- and 64-bit platforms.

[stencillogic]

Thank you for raising the PR.

[stencillogic]

Would something like from_raw_parts_u64 or a AsWords trait be useful/welcome? In order to allow using the same words on both 32- and 64-bit platforms.

I don't quite understand the idea. What problem would it solve? Platform independent serialization / deserialization?

[tgross35]

I was thinking of making it simpler for cases like this:

astro-float/astro-float-num/src/ops/consts/e.rs

Lines 143 to 156 in dd15e90

	#[cfg(target_pointer_width = "64")]
	let words = [
	14741299514016743161,
	12241124915312604155,
	15616730352774362773,
	12672098147611000049,
	12535862302449814170,
	];
	#[cfg(not(target_pointer_width = "64"))]
	let words = [
	614153977, 3432226254, 342111227, 2850108993, 3459069589, 3636053379, 658324721,
	2950452768, 2730183322, 2918732888,
	];

The arrays contain the same data, the 32-bit version just has to be split in half. A helper function could accept a u64 array on either platform and just reinterpret as needed.

I guess this might not be too useful outside of writing platform-agnostic tests?

[stencillogic]

I was thinking of making it simpler for cases like this:
...
I guess this might not be too useful outside of writing platform-agnostic tests?

Yes, it makes sense. There are several functions like this in the crate marked with #[cfg(test)] and used solely in tests.

[stencillogic]

What is the reason for marking the code unreachable? Wouldn't it be better to keep it as it was before, since the code could be reachable for architectures different from x86, x86_64?

[stencillogic]

Don't it cause a warning from compiler about nonsense check if target_pointer_width = "64" ? If target_pointer_width = "64" then the condition is always true since EXPONENT_MAX == Exponent::MAX, and EXPONENT_MIN == Exponent::MIN. But for target_pointer_width = "32" it can return false.

This statement is fair for all the places in the code where the line "#[cfg(target_arch = "x86")]" has been removed.