Given I discovered that #1946 was a bad issue (my fault) I'm not sure we need this.

Maybe not, but you may not wan't to solely rely on the existence of sw_vers to detect darwin as it can "easily" be fooled. $ARCH should be more reliable. On the other hand, it should not be that likely to have a fake sw_vers. ^^

Fair, I'll give this a bit more of a ponder. Belt and braces can't hurt I suppose.

I'm not sure I like the amount of code churn needed to have the global $ARCH but I also don't like the idea of having to call get_architecture from multiple places. It'd be non-trivial to thread $_arch through all the calls, leaving the possibility of just checking $(uname -s) in the check_help_for function. I remain unsure which approach is best over-all because we abstracted uname checking into get_architecture for a reason. I'm not going to make a call on this yet, but if you could come up with a way to minimise the change while having the same effect, I'd like to see a reduced patch.

I remain uncomfortable with this change. If a reduced scope patch is produced I'll consider it, otherwise I'll close this PR soon. Thank you for caring enough to try and fix this.

its pretty trivial really.

_arch -> downloader -> check_help_for. Its one code path, one variable.

Yes, I could just pass it through like you suggested. I'll try that soon.

I apologise for the delay, my December was crazy and then I was unwell. 👍

I've rebased your branch. Assuming this passes CI, I'll merge.