Add option to allow users to inspect the system-specific install script in `$tempDir` after failed installations

[ambirdsall]

the error which prompted this issue

While attempting to install on macOS x86, I ran into a strange edge case of my own making: as a personal in-joke, I had an emacsclient wrapper script at $HOME/bin/ed, which was invoked instead of /bin/ed by printf "\$a\nLABEL=%s /nix apfs rw,nobrowse\n.\nwq\n" "$label" | EDITOR=ed sudo vifs, derailing my installation. While this specific error was extremely idiosyncratic and not worth fixing per se (I suppose it could also download a statically-linked ed binary alongside the system-specific install script, but that seems excessive), the install script's unconditional invocation of the cleanup function automatically erased the information I needed to diagnose the problem at all; it would be lovely if there were a way to opt out of the automatic rm -rf $tempDir on failure, to enable debugging issues with the inner install script.

To debug this as-is, I had to

1. run something like curl -L https://nixos.org/nix/install > /tmp/nix-installer
2. manually edit the script to disable the cleanup function
3. run /tmp/nix-installer --darwin-use-unencrypted-nix-store-volume --daemon to fetch the nested install script, wait for it to fail again
4. read through the files in $tempDir (side note: thanks to whoever wrote the statement that printed the tempDir path to my terminal) until I found the issue

something which would have made my life a bit nicer in that moment

This is a naive and partial sketch of the approach that occurred to me; I don't know the ins and outs of trapping posix signals or script lifecycles well enough to write a fully correct or robust version.

SCRIPT_ARGS="$@"
PRESERVE_STATE_ON_EXIT=0

while [ $# -gt 0 ]; do
    case $1 in
        --preserve-state-on-exit)
            PRESERVE_STATE_ON_EXIT=1
            ;;
        *)
            ;;
    esac
    shift
done

# define oops, umask, create $tmpDir

cleanup() {
    rm -rf "$tmpDir"
}
maybe-cleanup() {
    if [ $? -eq 0 ]; then
        cleanup
    else
        exit $?
    fi
}

trap cleanup INT QUIT TERM

if [ $PRESERVE_STATE_ON_EXIT -eq 1 ]; then
    trap maybe-cleanup EXIT
else
    trap cleanup EXIT
fi

# introspect system and fetch/verify/unpack script

"$script" "$SCRIPT_ARGS"

The a possibly lighterweight alternative would be to print the generated curl command for the user's system to STDOUT (possibly only on error) so they can more easily dig in if things go wrong with the inner install script.

[ambirdsall]

I couldn't tell if the expectation was to use the headings in the issue template and add my content into those predefined slots or to just make sure I hit all relevant topics from it; I went with the latter, but I'm happy to reformat if that helps to keep issue triage efficient for the maintainers.

[abathur]

I agree with the general sentiment here.

---

For anyone else looking at this: The underlying error here (user script named ed taking precedence) should be coincidentally fixed in master (unreleased) via #4289, which used absolute system paths for each instance of the EDITOR env.

[stale]

I marked this as stale due to inactivity. → More info