fix(dist/linux): handle the possible unavailability of /proc in rustup-init.sh

[rami3l]

Closes #2700. I'm no shell expert, just trying to do my best...

This fix is based on @miigotu and @kerberjg's find in #2700 (comment) and #2700 (comment) respectively. It first refers to /proc/self/exe if it's accessible, then $SHELL if it's set, and finally /bin/sh. This executable is then explicitly passed to functions that previously relied on /proc/self/exe.

有備無患 yǒu bèi wú huàn

• to be prepared against adverse events

Concerns

• What about the grep '^Features' /proc/cpuinfo line which also relies on /proc? Fixed.
  • Currently, if grep fails (because e.g. /proc/cpuinfo is unavailable or malformed), we still assume armv7. This doesn't work in the case described in rustup-init.sh fails to detect platform correctly under docker buildx which lacks /proc #2700, where we're using ARMv6. OTOH, falling back to arm is much safer in this case and will resolve the aforementioned issue.
• How to test this patch?
  • The podman instance I use seems to have /proc access. I've tested the patch with a modified /proc path (/fakeproc) and the logic seems to work.

[rbtcollins]

Any reason not to use

_current_exe=get_current_exe

[rami3l]

@rbtcollins I assume you meant _current_exe=$(get_current_exe)?

The echo "Warning: ..." calls in the function body will interfere with the function's output when it's being used like so (unless the 1>&2 trick is used, as I understand it). Looking at the surrounding code, it seems like the $RETVAL trick is used exactly in this case to specify the actual value to be returned.

I referred to downloader() as an example when writing this. This is how it calls other functions via $RETVAL:

rustup/rustup-init.sh

Lines 596 to 599 in 2a5a69e

	check_curl_for_retry_support
	_retry="$RETVAL"
	get_ciphersuites_for_curl
	_ciphersuites="$RETVAL"

This is how it issues warnings:

rustup/rustup-init.sh

Line 604 in 2a5a69e

echo "Warning: Not enforcing strong cipher suites for TLS, this is potentially less secure"

Maybe we should use 1>&2 for echo commands instead? But in that case we might hit #3350, right?

[rbtcollins]

I think this looks reasonable.

There are other probes we could use, such as uname and lsb_release and gcc but they all have edge cases and caveats. One question for you then GTG IMO.

[rami3l]

Oops, a tiny typo here...

Suggested change

	echo "Warning: Unable to find /proc/self/exe. System architecture detection might be inaccurate."
	echo "Warning: Unable to find /proc/self/exe. System architecture detection might be inaccurate."