When invoking a `SystemCommand` with `:must_succeed => false`, the `SystemCommand::Result` class would mistake a `Process.Status` object for a `Fixnum`.
This commit fixes this by instantiating `Result` with the actual status code as a number.
from Cask::SystemCommand.run
* stderr and stdout are now separated, though both are available
* :print_stderr is made the default, closer to previous behavior
* exit status can be read from the result object
* plist parsing is more naturally handled in the result object.
The :plist argument to the run method was removed.
* whitespace
This is possibly the cause of other glitches such as random
failures on Travis.
Ruby 1.8 popen3 does set global $? for the exit status, Ruby
1.9 and above does not. For Ruby 2.0 users (Mavericks and
above with recent Homebrew), success or failure of the current
external command was determined by the exit status of some
previous command.
Closescaskroom/homebrew-fonts#186
- Raise `CaskError` instead of `Plist::ParseError` from module
- Improve error message when parse result is empty
- remove leading garbage text and emit it to stderr (seen in #5060)
- remove trailing garbage text and emit it to stderr (seen in #4819)
This has the incidental effect of emitting DMG licenses during
installation, which seems desirable as permanent functionality.
If not permanent, the warnings to STDERR should still be kept
temporarily to help get better bug reports on `hdiutil`.
A bug wrt DMG licenses must have been introduced in one of
#4892, #4887, #4889, #4900, #4975, #4978, or #4857. Presumably,
the cause is that STDERR was previously silenced when running
`hdiutil`. It would be cleaner (and more reliable) to redirect
STDERR and examine it separately, rather than clean up the
merged outputs.
closes#4819closes#5060
As parameter to the `run` method, but also exposed as interface to Cask authors
in eg `uninstall` `:script`. The existing DSL design: `:stderr => :silence`
is more flexible/extensible.
References: #4688
This may not be strictly necessary. But it should help track
down the `hdiutil` errors, which are persistent, but occasional,
and hard to reproduce. (We pass a flag to `hdiutil` to make
respond in XML.)
Fixes#4677. This change is necessitated by Homebrew's recent
switch to Ruby 2.x. These changes may be incomplete and/or
may not work well with Ruby 1.8. Travis should test Ruby 1.8
compatibility.
Previously, SystemCommand commands and arguments were processed as
strings which caused problems during shell interpretation if the
arguments were not escaped properly. Now all commands are expressed
as arrays and no longer require their arguments to be escaped.
Additionally, stderr and stdout could have been interleaved in the
past and now they are always separated.
- add new file "cask/utils.rb" analogous to "utils.rb" in Homebrew
- define odebug and odumpcask, analogs of ohai and friends, but
which only give output when --debug is in effect
- move the debug setting from an instance variable in Cask::CLI
to a method Cask.debug, defined in "lib/cask/options.rb", which
was added in #2276. (Perhaps options.rb should be merged back
into Cask::CLI).
- sprinkle odebug statements liberally throughout the codebase
- update tests
For safety.
- This is a step toward reworking system_command.rb so that
invocations are done in list form on the back end, avoiding
surprises from quoting and shell metacharacters.
- There is one transitional hack here: the _quote method in
Cask::SystemCommand is modified to avoid double-quoting. The
_quote method itself will go away in a future revision when
only list-forms are used.
- Casks using system are not touched. It seems natural to
address that when creating the DSL for after_install/before_install.
hdiutil can output DMG agreement information before the output plist
xml, and our plist parser was choking on that text before the xml
started.
so now we scan to the beginning of an xml document before trying to
parse the xml.
also added much more explicit error handling around the plist parsing,
to hopefully catch related and future plist-related errors.
refs #914
- add `run!` method which raises if command does not succeed
- use `run!` when the command we are running must succeed for things to
move forward. this should help produce clearer error messages in
failure scenarios.
- move caveats earlier in the install process so reports can be made
about potential failures
- remove the destination tree on cask install failure, so the cask will
not be considered installed
This cask install the batch version of Prey package, as detailed in:
http://support.preyproject.com/kb/installation/how-to-deploy-prey-in-batch-mode-mac-os
As it requires an API key to be installed, a warning is displayed to the
user with an explanation on how to fix the issue.
Also added -E option to sudo invocation so environment variables can be
passed to the installer.
closes#953
Signed-off-by: phinze <paul.t.hinze@gmail.com>
- re-added a lost nil guard on `Dmg` containers
- `FakeSystemCommand` was still returning an array of split lines
instead of a string, even though its real counterpart switched to
string when install/uninstall landed
- flushed out an alfred cli bug
- moved plist parsing down to SystemCommand layer
`Cask::Installer` was already much too complex, so I took this
opportunity to throw a `Cask::Container` abstraction around the
extraction part of the package install step.
It goes like this: a Cask's URL points to a Container of some sort. The
containers we currently support are: dmg, zip, tar, and (new) naked.
Naked refers to a raw file that just needs to be copied in place. This
currently just means a pkg file, but in the future it may expand.
A Container knows how to do two things: identify a path as being its
type (`Container.me?`) and extracting the contents of its container to
the proper destination for a Cask (`Container#extract`).
The first Cask we have that supports the naked pkg type is
`heroku-toolbelt`. (Thanks to @sheerun for the Cask definition.)
Other miscellania batched in with this refactor:
- switched to an explicit require strategy rather than globbing
- `Cask::Installer` is instantiated now to match its interface with
other similar collaorators
- simplified zip and tar identification to shorter strings rather than
exact matches of full `file -Izb` output
- `Cask::SystemCommand` gets explicit output redirection options
- many rogue backticks replaced to properly use `SystemCommand`
- fixed misnamed test file `link_checker_spec.rb`
- remove some extraneous `after` clauses in tests; leaning more on
`test/support/cleanup.rb` to uninstall for us
- pkg uninstall `:files` gets a `-rf` to support removing dirs
refs #839 and #1043
accepts a single argument, which is a relative path to a pkg
inside the extracted Cask; homebrew-cask will attempt to install this
pkg after the Cask is extracted via `installer`
because of the many different ways uninstallers work, this
has several features:
- `:script`: a script in the Cask which serves as an uninstaller (e.g.
Vagrant, VirtualBox), uses `:args`, and `:input` keys to interact
with said script
- `:pkgutil`: a regexp which captures all package_ids installed by this
cask; homebrew-cask will list all files installed under these ids and
remove them
- `:launchctl`: a list of bundle_ids for services that should be
removed by homebrew-cask
- `:files`: a fallback list of files to manually remove; helps when
uninstallers miss something
refs #661
