Also change the logic a bit to iterate over the individual files per
directory, as having a directory without bash commands will otherwise
pass a literal `*.sh` to `bash -n`, causing it to fail.
ClosesHomebrew/homebrew#48323.
Signed-off-by: Martin Afanasjew <martin@afanasjew.de>
This should help to avoid collisions with external commands and other
shell functions in the future and is closer to what we do in Ruby, where
commands are namespaced by being methods of the `Homebrew` module.
`bin/brew` already sets up a bunch of environment variables. There's no
need to re-export them for external commands. (`HOMEBREW_LIBRARY_PATH`
and `HOMEBREW_CACHE` continue to be determined later in the Ruby code.)
We have asserted before that the 1st argument is the command name. No
need to pass it to the bash command, which will make the argument
handling for the command itself a bit easier.
Remove the executable bit from the file to make it clear it is not
supposed to be executed directly. This should make the shebang line and
the early check also unnecessary.
Commands implemented in shell (bash) are supposed to be sourced from
`bin/bash` instead of being executed directly. Consequently, don't
expect the implementation files to be executable.
Allow people to run this command (so we can ask people to test it)
without having to set `HOMEBREW_DEVELOPER`.
ClosesHomebrew/homebrew#48260.
Signed-off-by: Mike McQuaid <mike@mikemcquaid.com>
* only set HOMEBREW_UPDATE_BEFORE inside pull instead of fetch.
* fix HOMEBREW_UPDATE_BEFORE/AFTER variable settings. They should be set
to INITIAL_REVISION and CURRENT_REVISION correspondingly.
* avoid unnecessary duplicated shellout.
* remove unused variable.
* make sure exit after reset.
* more wide guardian range. We should trap reset_on_interrupt as soon as
we are about to change any files.
* check INITIAL_BRANCH against UPSTREAM_BRANCH
* use HOMEBREW_REPOSITORY instead of HOMEBREW_PREFIX
* better performance:
* update-bash: avoid shellout
* update-report: use strip_prefix and tr
* more robust:
* explicitly handle the case when repo is HOMEBREW_REPOSITORY to avoid
to handle the trailing backslash.
* handle both lower case and upper case when stripping non alpha and
digital characters.
While running a command from the integration tests, invoke `.result` to
trigger a save of the result set, but don't call `.format!` to avoid
(re)generating the HTML output with every run. The final output will
still be written once the main unit test run completes.
This significantly speeds up the integration tests, that take about 1/3
less time to complete when not generating the intermediate output.
ClosesHomebrew/homebrew#48280.
Signed-off-by: Martin Afanasjew <martin@afanasjew.de>
Commands executed during integration testing are executed in a separate
process and thus generate a new result set for each command. To avoid
that these results override each other, they need to have a unique
`command_name`. Derive this name from the test class/name and the index
of the command inside that test, resulting in identifiers like
`IntegrationCommandTests#test_prefix.1 brew --prefix`.
Also replaces `TEST_TMPDIR` in the arguments with `"@TMPDIR@"` to get a
cleaner command identifier that is independent of the temporary
directory that changes with every run.
Simplify running a single test (or even a single method from a single
test), which is useful when developing or debugging a test. Instead of
`brew test TEST=test_diagnostic.rb TESTOPTS=--name=test_check_DYLD_vars`
it is now possible to run `brew tests --only=diagnostic/check_DYLD_vars`
to get the same effect. Similarly, `brew tests --only=diagnostic` maps
to `brew test TEST=test_diagnostic.rb`.
ClosesHomebrew/homebrew#48279.
Signed-off-by: Martin Afanasjew <martin@afanasjew.de>
This parameter is printed by every `brew tests` run and feeding it back
into the test suite allows to get an identical run. Make it easier to
do that by accepting the argument as it is printed during `brew tests`.
This requires a slight adjustment to named-argument passing as the
number that follows `--seed` is also seen as a named argument, but we
usually only want to pass `TEST=<something>` and `TESTOPTS=<something>`.
The filter for `vendor/bundle/` is useless because this directory is
located in `Homebrew/test/` and that one is already filtered. Moreover,
SimpleCov already loads the `bundler_filter` profile (that installs
basically the same filter) in its default configuration.
This came up in https://github.com/Homebrew/homebrew-emacs/pull/158#issuecomment-172031003; `install_metafiles` is passed the current directory, which can be changed during installation by `Dir.chdir`.
There may be cases where the metafiles are in a subdirectory, but my
guess is those are rare and this is brittle and undocumented behavior
anyway.
