This updates logic to add a `#scheme_and_version` method to be used
with `.sort_by` and `.max_by`. Using `Keg#version` by itself can be
inaccurate when different version schemes are present. This also
updates the behavior of `Formula#eligible_kegs_for_cleanup` to match
the previous behavior. We were dropping the wrong keg based on the
sort being reversed in a previous PR.
When a Keg is unlinked, brew-link gives a helpful message for how to proceed: adding the `--overwrite` flag.
For safety, it also recommends running in `--dry-run` mode first to see what would be deleted.
So a user's common flow would be:
1. run `brew link foo`
2. get error message with guidance
3. run `brew link --overwrite --dry-run foo`
4. inspect
5. run `brew link --overwrite foo`
In this flow, steps 3-5 are likely very common. Common enough that a user may use their shell history to re-populate their prompt with step 3's command, delete the `--dry-run` flag, and re-run. (The end goal, of course, is to link `foo`.)
The `--dry-run` flag needs to be removed from the command, of course. If it had been at the _end_ of the command, it would make the subsequent modification easier.
Instead of "up arrow, left-arrow a bunch, then backspace over --dry-run, hopefully not backspacing over the formula name", it would be easier for the user if the dry-run flag were already at the end of the command. Then the user can "up arrow, backspace a few times and hit enter".
What's more, if the last arg were `--dry-run`, a more advanced bash user could even use `!:-` to re-run the link command with all-but-the-last-arg.
This allows us to remove all the manual timestamp fiddling and lets
`gnu-tar` handle it for us instead (as-per the most recent
recommendations on https://reproducible-builds.org/docs/archives/).
Some formulae (e.g. `luv`) have a `lib/lua/5.1` subdirectory inside
their keg. Before this change, the `5.1` subdirectory is symlinked into
`HOMEBREW_PREFIX`.
This can result in `luarocks` installing things into a formula's keg,
which we don't want.
Let's fix that by making sure that `brew link` creates these
subdirectories instead of symlinking them. We already do this for
subdirectories of `share/lua`:
8dd96ae8ba/Library/Homebrew/keg.rb (L430)
Previously, `brew upgrade gcc@10` could get overzealous and remove the
`LinkedKeg` record for `gcc@9`. This is bad because we then think
`gcc@9` is unlinked when it is not and it causes a tonne of conflicts
when trying to link `gcc@9` again.
Instead, fix up the alias and versioned alias cleanup to be more
precise and only delete the symlinks that point to the current `rack`,
`opt_record` or `linked_keg_record` and unify the logic so it's
performed consistently.
While we're here:
- don't `remove_old_aliases` every time we try to unlink a directory
but just perform it once per `unlink` operation
- remove the linked keg record on `uninstall`
This is weird. We're seeing some installs where formulae
have no opt prefix and no linked keg record, *but* still
ended up linked into the prefix. Upgrades will blow up in
this case because the old symlinks won't get deleted and
Homebrew will report a nonsensical error message about
not being able to link over another link from the same
formula!
Since the optpath is either missing or pointing to the
wrong destination at this point, the checks for conflict
linking above will have failed. We should instead be safe
to simply blow away these conflicting symlinks and replace
them with the new targets.
Fixes https://github.com/Homebrew/homebrew-core/issues/68866.
- manually `raise Errno::ENOENT` to ensure that a keg that doesn't exist
isn't flagged as a system dependency.
- remove the inconsistent and incorrect summary messaging.
