The "upgrades with asking for user prompts with dependants checks"
test for `cmd/upgrade` has been failing on CI. The regex in the test
doesn't match expected output because the negative lookaheads aren't
working as expected. The intention is to make sure that the names
aren't repeated (i.e., second shouldn't match first, third shouldn't
match first or second) but the negative lookaheads should be
_inside_ the second/third capture group for this to work as intended.
This updated regex should work as expected. I manually tested it
using the output from CI to make sure that it matches when no formula
names are repeated (e.g., `Formulae (3): testball, testball4,
testball5`) and does not match if formula names are repeated (e.g.,
`Formulae (3): testball5, testball4, testball5`).
A bug introduced with #19460 is that if a redirection has a file size (for example, if it is a 302 that also has HTML content), Homebrew will invalidate the cached location.
`Homebrew::SimulateSystem.current_os` may be returning the host OS or
a simulated OS and we can't be sure which in this context. At the
moment, this is expected to be the host OS but that may change in the
future. It shouldn't matter on a technical level but using "host" in
these variable names may lead to confusion.
This replaces "host" in names with "current", as it more accurately
describes the information.
This extracts the logic for generating the `system_options` array in
the `replace_version_and_checksum` method into a separate
`generate_system_options` method. This logic is becoming more complex
(after recent changes) and manually testing it is a pain, so this
change is intended to allow us to add tests. The tests added here
provide 100% coverage for the method.
This reworks the `SimulateSystem` args in the `bump-cask-pr`
`replace_version_and_checksum` method to respect `depends_on arch`
values in casks. That is to say, we shouldn't simulate Intel for a
cask using `depends_on arch: :arm64` and we shouldn't simulate ARM if
the cask uses `depends_on arch: :x86_64`.
In the process, this refactors how we collect/combine OS/arch values.
To make this approach work predictably, I removed the logic that
omits OS values matching the host OS (as `SimulateSystem` already
handles this). The `[{ os:, arch: }]` hash format only made sense when
we were omitting values, so this returns to the previous
`[[os, arch]]` array format (to align with the
`OnSystem::ALL_OS_ARCH_COMBINATIONS` array format).
This adds tests for the livecheck DSL's `version` and `arch`
delegates. This doesn't affect test coverage but it ensures that the
methods work as expected in `livecheck` blocks.
This adds the ability to specify tests that depend on a certain CPU
architecture using `:needs_arm` or `:needs_intel`, similar to the
existing `:needs_macos` and `:needs_linux` metadata for tests that
depend on a certain OS.
The `replace_version_and_checksum` method handles a `CaskInvalidError`
when loading a cask (handling casks that aren't valid on Linux) but
we can sometimes still encounter an error when bumping a cask with
on_system blocks. For example, bumping `displaylink` will produce a
`Cask 'displaylink' is unreadable: undefined method 'csv' for nil`
error when `SimulateSystem` runs as Linux, as the cask interpolates
`version.csv.first` in a `license` string but `version` isn't set on
Linux.
This adds `Cask::CaskUnreadableError` to the `rescue` arguments,
which accounts for this particular situation (allowing `displaylink`
to be bumped like before).
`bump-cask-pr` was recently updated to add Linux support but the
change to the `replace_version_and_checksum` logic has broken the
command for casks that have on_system blocks that reference specific
macOS versions (e.g., `on_monterey :or_newer` in `logi-options+`).
The previous logic only simulated the arch, so the `current_os` value
on macOS was a specific version like `:sequoia`. The current logic
uses generic `:macos` values, which work for `on_macos` blocks but
don't work for blocks like `on_sequoia`, etc. This leads to an
`undefined method 'latest?' for nil` error, as `old_cask.version` is
`nil` in this scenario (i.e., none of the on_system blocks apply to
`:macos`, so `version` is never set).
This updates the method to only specify the OS in `system_options` if
the value doesn't align with the host (e.g., `:linux` on macOS),
which restores the previous behavior.
This also replaces `:macos` values with the newest macOS version
(e.g., `:sequoia`), so that `bump-cask-pr` on Linux can update casks
with on_system blocks that reference specific macOS versions. A
generic `:macos` value doesn't work with those on_system blocks, so
`version` is never called on Linux in that scenario but it works as
expected if we use the newest macOS value instead. This may not be
perfect but it brings `bump-cask-pr` a little closer to parity with
macOS on Linux.
Lastly, this skips `system_options` values where `old_cask` has no
version, as this means the cask doesn't apply to that OS/arch. We've
been seeing a related error on homebrew/cask autobump CI and this
guard should help to avoid it.
