Ubuntu is a kind of system where everything after the first space in a
shebang is provided as the first argument. Therefore, we can't allow
generating a script that has more than one argument to Ruby interpreter
in the shebang.

http://stackoverflow.com/a/4304187/11687

References ea0103fb

Fixes #569

Re-wrapping to 78 chars was dangerous since it could lead to moving
a #-reference to the beginning of a line which would then get
interpreted as a comment.

Wrapping pull request message is not useful anyway since it's Markdown.
For best results when rendered, users should unwrap their text.

Fixes #548, closes #570

This reverts commit 5a942210.

Removing "X" from `less` configuration did help to clear the screen on
exit, but also led to paginated output being swallowed up when the
output was less than one page (i.e. one line). I have no idea what
actually happens, but it's important to bring back full output.

I couldn't find a way to have paginated output cleared on exit at the
same time as having single-page output normally.

Fixes #520, reopens #368

The comment char is read from git config if it exists, but defaults to
"#" as before.

Fixes #530

`bundle exec rake standalone` in development mode would try to shell out
to `ruby --disable-gems` to detect 1.9 support, but that would fail
because Bundler tries to propagate itself through RUBYOPT and
`--disable-gems` obviously doesn't play along with it.

If the user's `LANG` environment variable doesn't include "UTF-8", the
default external encoding in Ruby will be US-ASCII and as such, string
operations will fail dealing with pull request message or changelog that
contains non-ASCII characters.

This explicitly configures external encoding in hub to default to "UTF-8".
This applies to data over ARGV, `$stdin.read`, and reading files.

Fixes #566

When looking up an existing hub token among user's authorizations, be
sure to fetch all results by following pagination links. Previously we
just stopped at the 1st page, and if the token for hub didn't appear
there we would try to create it, which would result in a 422 validation
error if it already existed.

References 1e4477b5

Fixes #531, fixes #558, fixes #563

Since the /tmp directory is readable by everybody on Unix, and since
the patch name could be public or easy to guess, a attacker could create a symlink
to a file writable by the user running hub, which would be replaced by the patch.

This has been assigned CVE-2014-0177

Use Homebrew's preferred commit message

For simple updates of version, Homebrew's preferred commit message and
PR title is "hub 1.2.3".

Closes #517 [ci skip]

Hub merge should use repo name from API

The repository name of the pull request head doesn't necessarily match
the name of the project's repository. This can happen when a user forks
a project but already has a repository of the same name. Use the name
returned in the api response for the fetch url to avoid this edge case.

It shells out to other Ruby scripts, but we don't want them to inherit
the same bundle as the hub project. So we clear RUBYOPT which in a
Bundler environment contains `-rbundler/setup`.

`less -X` prevents clearing of screen, but we want these commands to
clear themselves up on exit, i.e. to not litter people's output buffers:

    git help hub
    hub <command> --help

Fixes #368

If someone misspells `--help`, for instance:

    git pull-request -help

This aborts the command instead of continuing by interpreting `-help` as
it was the deprecated title argument. Any value that starts with a dash
will abort.

Fixes #460

Enable API debug output by setting HUB_VERBOSE

Hashes are unordered in Ruby 1.8, so we can't check for this in debug
output: `{"name": "dotfiles", "private": false}`. The keys might appear
in different order.

A proper fix would be to normalize keys order before output assertion in
tests, but I can't be bothered to make that. Here I'm simply avoiding to
make any assertions for that line.

https://travis-ci.org/github/hub/jobs/19817554

This will print information about every HTTP request and response to
stderr. Authentication info is redacted.

Example:

    $ HUB_VERBOSE=1 hub ci-status
    > GET https://api.github.com/...
    > ...

For people who are using hub for a longer time now, hub has stored an OAuth
token in Authorizations API with a `note_url` value that is now outdated
(e.g. "http://defunkt.io/hub/"). The newer `oauth_app_url` doesn't match
that, so the token is not reused. Hub tries to create a new OAuth token
with a `note` that says "hub" which fails with a 422 because it already
exists.

This fixes reusing old OAuth tokens even if the `note_url` doesn't match
anymore. Any key with `note` that says "hub" will get reused.

Fixes #496

Technically, they're going to be the same values for OAuth tokens
created via the API, but since we write to `note_url` field it makes
sense to use the same field to read the value back, for consistency.

This is so that a user-friendly message is shown instead of a useless
stack trace.

Previously, these errors would not be shown to the user, which would
lead to some very uninformative HTTP error messages.

This is so the next shell prompt isn't displayed on the same line where
the interrupted hub prompt is.

This is irrelevant in zsh, which ensures that the prompt always starts
on its own line.

Add option to open newly created pull request in the browser.

Closes #344

Allow Enterprise configured for regular HTTP

Allows using these commands for Enterprise configurations over regular HTTP.

Example `~/.config/hub`:

  my.example.org:
  - user: myuser
    oauth_token: mytoken
    protocol: http

Currently there is no way to set `protocol` other than manually editing the
configuration file.

Since we use a test server representing the GitHub API, all requests get
transformed to HTTP in tests. Make a check that the original requests
really was intended to go over HTTPS.