The `|&` is new in Bash 4, which is not the default on Macs.
Given that Laptop is meant to be run on a fresh machine, `2>&1 |` is preferable,
since it'll work on the Bash 3.x, which is what comes installed on new machines.
* Snow Leopard was released in August, 2009 (about five years ago).
* Yosemite will be released in a couple of months.
* We support the last three versions of Ubuntu, so this is a little more
consistent.
* Use the same format as the Linux "Requirements" section for consistency.
We can future-proof the script little more by setting `node_version` to 0.10.
NVM will automatically install the latest 0.10.x version:
https://github.com/creationix/nvm#usage
Fixes https://github.com/thoughtbot/laptop/issues/260
This is a soft error that occurs when Laptop is run and PostgreSQL isn't
currently loaded by `launchclt`, e.g. on the first run of Laptop:
Starting Postgres ...
/Users/wellmade/Library/LaunchAgents/homebrew.mxcl.postgresql.plist ->
/usr/local/opt/postgresql/homebrew.mxcl.postgresql.plist
launchctl: Error unloading: homebrew.mxcl.postgresql
The fix would probably be to check `launchctl list` to see if the service is
already loaded before calling `launchctl unload`.
This small added complexity probably justifies a function for running
appropriate `launchctl` commands. Homebrew has a strong convention for the
naming of `launchd` plists: "homebrew.mxcl.FORMULA_NAME.plist", e.g.
"homebrew.mxcl.postgresql.plist" and "homebrew.mxcl.redis.plist". This likely
makes a function to do the `launchctl` work fairly straightforward.
Fixes https://github.com/thoughtbot/laptop/issues/259
Laptop raises the following warning on `brew link openssl ` when OpenSSL has
already been linked:
Warning: Already linked: /usr/local/Cellar/openssl/1.0.1h
To relink: brew unlink openssl && brew link openssl
I think the resolution is straightforward: follow Homebrew's recommendation and
`brew unlink` before `brew link`. I verified that `brew unlink` exits with
success even if the formula wasn't previously linked as would be the case the
first time Laptop is installed.
Fixes https://github.com/thoughtbot/laptop/issues/258
The following doesn't behave as intended:
brew_install_or_upgrade 'postgres'
because "postgres" is an alias, not the full name of the formula. The actual
formula name is "postgresql".
The first time `brew_install_or_upgrade 'postgres'` is invoked it works as
expected — `brew install postgres` is run. Additional invocations result in
`brew install` instead of the expected `brew upgrade`.
The `brew_install_or_upgrade` function uses `brew list -1` to obtain a complete
list of installed packages, and these package names will be the actual package
names. So, grepping for an alias (with the `-x` option) won't match:
$ brew list -1 | grep -Fx postgres
Note that there's no output, even though PostgreSQL is installed. Using the full
package name behaves as expected:
$ brew list -1 | grep -Fx postgresql
postgresql
Rather than passing on the first argument to `brew_install_or_upgrade` to the
`brew list` commands, the argument should first be expanded to the actual
package name.
Fixes https://github.com/thoughtbot/laptop/issues/257
On a clean OS X Mavericks, Laptop outputs an error message towards the end of
the install script when `brew_install_or_upgrade` is invoked for OpenSSL:
Upgrading and linking OpenSSL ...
Error: openssl-1.0.1h already installed
Linking /usr/local/Cellar/openssl/1.0.1h... 1139 symlinks created
The current version of OpenSSL was installed earlier as a dependency of
PostgreSQL. In general, a similar error will be logged whenever
`brew_install_or_upgrade` is called for an installed package that's already up
to date.
Currently, Laptop discards the error exit code of `brew upgrade` as follows:
(brew upgrade "$@") || true
In addition to logging an error when there isn't truly an error, this approach
can cause real `brew upgrade` errors and potential bugs in Laptop to be masked
(e.g. Homebrew formula build failures or Laptop mistakenly invoking `brew
upgrade` for a package that isn't actually installed).
I think a better approach would be to check whether the installed package is the
same version as the current Homebrew formula before attempting `brew upgrade`.
Homebrew's install script checks whether the command line tools are installed,
and, only if necessary, will run `xcode-select --install`.
https://github.com/Homebrew/homebrew/blob/go/install#L162
The test they are using depends in part on a heuristic:
afa45493af
I confirmed that it works on a clean 10.9 install.
We also should not instruct the user to run `sudo xcodebuild -license` at all.
Here's why:
The `xcodebuild program` isn't included in OS X Mavericks. When you run
`xcodebuild`, you're actually finding one of 83 shims found in `/usr/bin` that
are included in Mavericks. These shims are an important part of how the
`xcode-select` mechanism works. When the command line developer tools have not
been installed, invoking any of these shims won't do anything other than prompt
you to install the command line developer tools:
$ xcodebuild -license
xcode-select: note: no developer tools were found at
'/Applications/Xcode.app', requesting install.
Choose an option in the dialog to download the command line developer tools.
(And, the GUI install dialog is presented.)
Of course, since the real `xcodebuild` program isn't actually available, it
doesn't make sense to try and use it to accept the Xcode License Agreement.
Instead, go straight to installing with `xcode-select --install` (xcode-select
comes with OS X Mavericks and is not a shim):
$ xcode-select --install
xcode-select: note: install requested for command line developer tools
And, the exact same GUI install dialog is presented (prompting the user to
either "Get Xcode" from the App Store or immediately "Install" the command line
developer tools).
For more about `xcode-select` and it's shims, see `man xcode-select`.
This also removes redundant Vagrant setup instructions and capitalize Vagrant.
This uses installer(1).
On GNU we already install `heroku-toolbelt` using apt, and that package
has a dependency on the `foreman` package.
We install this via installer(1) instead of gem(1) so that it remains
regardless of the development Ruby setup.
We install this via installer(1) instead of from the `Gemfile` so that
it remains regardless of what other developers do on the project (as
described by David Dollar).
Clean install on OSX Mavericks 10.9.4, I encountered an error similar to this:
https://github.com/sstephenson/rbenv/issues/487
Implemented same zsh fix, seemed to work alright. I think this is what needs to
be changed in the code.
Includes:
* Fix the "brew_install_or_upgrade" function, use it for all components
* Remove executable perms
* Skip installing an already installed ruby
Technically re-running laptop is not idempotent in that it will upgrade
brew-installed items and potentially install a new ruby. I feel this
is expected and wanted.
Includes:
* A minimal gem, rake, and rspec environment
* The Distro class wraps up commands to test and provision vagrant boxes
* Publishing boxes to s3 has been simplified and improved
* Significant documentation updates
New laptop specs should now be significantly easier to write and understand.
If postgresql was installed and started in a previous run, start causes:
Error: Service `postgresql` already started, use `brew services restart postgresql`
When developing changes to these scripts, it's important to be able to run and rerun this script without removing applications.
Vagrant boxes can be created automatically after a successful run of the
laptop test suite. These vagrant boxes are published to
[vagrantcloud](http://vagrantcloud.com/thoughtbot/) and should be a
solid start on a vagrant dev box suitable for modern ruby and
ruby-on-rails development.
Improvements include:
* Vagrantfiles fixed to have predictable names
* test/runner.sh now knows how to render vagrant boxes after tests are
successful
* Error reporting improvements
* Full documentation on creating new base boxes
We now require vagrant >= 1.5.0 to use the automated test suite built
into laptop.
If the file ~/.laptop.local doesn't exist, the laptop script returns a
failing exit code. This doesn't really make sense, and gets in the way of our
test runner. Wrap this in slightly more verbose `if` to ensure a
successful exit code.