- If a HEAD branch name isn't specified at all, then the user probably
wants to shortcut adding one by being told what the default branch for
the repo is. Otherwise they have to click the URL, look at the GitHub
UI, then type the branch name into `branch: "foo"` syntax.
- Since `head` must now specify a url and branch, the `head do` block
with only these stanzas can be condensed to the single-line
`head "url", branch: "branch"` format.
- There's a TODO on the "someday" list [1] to ensure that `head` Git
URLs always specify a branch.
- So I thought I'd automate this worry by adding an audit.
- Since `resource` block URLs tend to be pinned to SHAs, if indeed
they are Git URLs, this audit only applies to `head` URLs.
[1]: https://github.com/orgs/Homebrew/projects/5?pane=issue&itemId=98789749
The `Cask::Cask` objects in the `SkipConditions` tests don't
interpolate `version` in the `url` strings, so these are technically
unversioned URLs as a result and would be skipped as unversioned. This
updates the URLs accordingly, so they won't trigger the unversioned
skip as a fallback (if the intended test doesn't work as expected).
This is something I discovered while writing a test for a cask that
shouldn't be skipped.
We've been adding `disable!` calls with a future date to casks using
an unsigned app. That implicitly deprecates the cask until it reaches
the disable date, so we've been having to add simple `livecheck`
blocks to casks that use a default check to ensure that livecheck
continues to check them. It was suggested that it would be simpler to
have livecheck not skip casks that have a `disable!` call with a
`because: :unsigned` argument and I agree, so this modifies
`SkipConditions` to add a special case for this scenario.
`ensure_formula_installed!` requires the `Formula` class to be loaded
before being called to work properly.
Let's guarantee that instead by implementing it as an instance method of
the `Formula` class.
See discussion at #20358.
The `Formulary` and `system_command` debug output is incredibly verbose
by default and this is pretty annoying when all you want is to get
better backtraces when there's an error.
Instead, let's require `--verbose` and `--debug` for the noisiest output
message types.
Follow up on `DownloadQueue` for download concurrency on `brew fetch`
and `brew install` to also add support for `brew reinstall` and
`brew upgrade`.
This required a fair bit of refactoring to make this work so I've also
made `install.rb`, `reinstall.rb` and `upgrade.rb` `typed: strict` to
add some extra guardrails from Sorbet here.
Co-authored-by: Carlo Cabrera <github@carlo.cab>
This fixes the weird/broken existing behaviour which was incorrectly
creating symlinks at download time. It also defers much more logic to
the original code.
For clarity, rename the existing `API::Download` class to
`API::SourceDownload`.
While we're here:
- add a/improve the `download_type` method on all `Downloadable`
subclasses to improve download queue output format
- move some logic to `RetryDownload`
Allowing using `HOMEBREW_DOWNLOAD_CONCURRENCY` to use the
`DownloadQueue` for `brew install` by downloading and extracting
bottles in parallel.
This requires some fixes in e.g. `Dependency` and `FormulaInstaller`
to be able to front-load all downloads and handle parallelisation of
bottle pouring.
Behaviour without `HOMEBREW_DOWNLOAD_CONCURRENCY` set should be
unchanged.
Attestations are not handled for now and the UI should be improved
before we roll this out to users.
Post-install upgrades are not yet parallelised.
Co-authored-by: Carlo Cabrera <github@carlo.cab>
- Use undocumented (for now) `HOMEBREW_DOWNLOAD_CONCURRENCY` instead
of `--concurrency` flag and avoid passing around `concurrency`
- Create and use `Formula#enqueue_resources_and_patches` helper method
- Rename some method calls to be more obvious
- Use `Downloadable` type to simplify type checks
- General refactoring
- I found a few occurrences of this pattern from
https://github.com/orgs/Homebrew/projects/5?pane=issue&itemId=97021840,
that is an automated style request for:
`core: use / instead of + operator in e.g. (lib+"lv").install "lv.hlp"`.
- Upon adding tests I realised that there's also the `prefix + "bin"`
case that's already handled differently, so let's combine the handling
given it's the same `+` that's wrong.
Add a new RuboCop to detect the use of 0.0.0.0 in formulae which
indicates binding to all network interfaces, internally or externally,
so is a bad default and potentially a security risk.
Co-authored-by: Issy Long <me@issylong.com>
This adds support for Cask old tokens used for renames of Casks.
We'll now correctly check these at installation time to avoid repeatedly
installing renamed Casks and dump them in the Brewfile. We also use this
logic to avoid cleaning up renamed Casks.
- Homebrew Bundle referred to formulae as "brews". But it referred to
casks as "casks" and taps as "taps".
- Let's use the same terminology everywhere.
- (I know that `brew "hello"` is the formula syntax in the Brewfile, so
I'm not changing that (though would be up for it, in a backwards
compatible manner), just making the code more consistent.)
- This file was _massive_ - over 60k lines and we had to bump the file
size limit for pushes to the repo!
- This was because by default Tapioca, when it encounters a
`require "rubocop"` during RBI generation, loads all of the cops ever
because they're all classes inside `RuboCop::Cop`.
- There wasn't an easy way to control this at Tapioca generation time
(we tried), so now we parse the generated RBI file and delete classes
and method definitions that we don't use.
- I regenerated the RBIs (`brew tc --update rubocop`) and added new
things to the allowlist until Sorbet came back green.
- Now the file is ~7k lines and 240K - much better!
