warn user that installation will fail if she doesn't have necessary write permissions

[Mizou04]

added the guard logic against absent write permissions in bin/ruby-install

[postmodern]

Minor changes requested. git commit --amend and force push the branch, instead of closing the PR.

Also, this needs unit-tests. Luckily, you can copy test/cli-tests/no_reinstall_test.sh and adapt the test file. The test file just needs to do chmod -w $test_install_dir in the setUp() function before the tests.

[postmodern]

Minor changes.

[Mizou04]

Can you review the last commit?

[postmodern]

Just thought of an edge-case where this would break for new users that do not have /opt/rubies or ~/.rubies on their system's yet.

[postmodern]

I just realized an edge-case. If $install_dir doesn't exist yet, such as when running ruby-install for the first time on a new system that doesn't have /opt/rubies or ~/.rubies yet, it will consider the $install_dir non-writable. We should also check for it's existence and if it's not-writable.

if [[ -d "$install_dir" ]] && [[ ! -w "$install_dir" ]]; then
  ...
fi

[postmodern]

This might cause problems if the user specifies a really long --install-dir who's parent directories do not exist yet.

ruby-instlal --install-dir `/does/not/exist/yet` ruby

Since the top-most parent directory / is not writable by a regular user, than creating /does would fail.

We could move this logic into share/ruby-install/functions.sh pre_install which already does a mkdir -p "${install_dir%/*}" || return $?.

[Mizou04]

there's an issue I think with this, for example if install_dir is of this form : /this/part/exists/this/part/exists/not, we should recursively (iteratively) return backwards to check for the existing part, before testing if it was writable or not; then we proceed with our normal operation

[postmodern]

Or we could take advantage of mkdir -p which recursively creates the directories.

[Mizou04]

it won't create them if we don't have write permissions over /this/part/exists/

[postmodern]

If the mkdir -p fails because the last existing sub-directory isn't writable then the command will fail, which could be detected, an error message printed, and an error code returned.

[postmodern]

Noticed a few issues with the code. I also still think the while loop can be replaced with mkdir -p.

[postmodern]

You're going to hate me for constantly suggesting alternative approaches, but I think recursively checking each parent directory is unnecessary:

local parent_dir="${install_dir%/*}"

if [[ ! -d "$parent_dir" ]]; then
  mkdir -p "$parent_dir" || fail "cannot create installation directory"
elif [[ ! -w "$parent_dir" ]]; then
  fail "installation directory is not writable"
fi

• If the parent directory does not exist, try creating it with mkdir -p. If mkdir-p fails, this means that one of the parent directories is not writable, and hence all sub-directories would also not be writable. This basically offloads the recursive checking logic to mkdir -p.
• If the parent directory does exist, then test if it is writable.

[postmodern]

I still think the above while loop is a bit too complex and think checking mkdir -p would have the same effect.

[postmodern]

fail calls exit -1, so status=$? has no effect. Maybe you want to use error and then return 1?

[postmodern]

I think this should be return $??

[postmodern]

Maybe to make this test more explicit, you could add another variable called non_writable_parent_dir="$test_fixtures_dir/..." and set test_install_dir="$non_writable_parent_dir/ruby".