Fixed-width hexadecimal encodings perform unaligned writes

[andrewthad]

word8FixedHex and friends perform unaligned writes to their buffer. While this is fine on x86, I believe this would cause crashes on ARM.

[hvr]

There's certainly merit in having additional aligned-write codepaths for platforms which may run into unalignment-faults which can either crash the process as you suggest or suffer a significant performance penalty. I do recall that Duncan has been suggesting for quite some time for GHC to be taught about having more primitives in its APIs dealing with unaligned/aligned writes more explicitly.

[andrewthad]

ghc-prim already provides primitives for unaligned writes, both to Addr# and the ByteArray#. So, nothing new is needed there. I can prepare something to address this issue.

[sjakobi]

#133 seems somewhat related, since it's also about unaligned writes.

[Bodigrim]

@andrewthad a patch would be highly appreciated, especially considering an advent of Apple M1.

[Bodigrim]

Given that all our CI builds, including ARM both 32 and 64 bits, work fine, I'd like to close this, unless there is a reproducible evidence of the issue.

[andrewthad]

I looked at the code again. I had misread when I originally opened this issue. I do not believe this is actually a problem. Closing.