Release 1.28.0

This release includes quite a few changes including those to how we
autoname branches based on bug or blueprint. Git review won't do that
anymore.

Many other changes have to do with improvements to testing.

If usepushurl is set, then it's likely that the pushurl is pointing
to the code review system while the normal url is pointing to a read
only mirror. If the read-only mirror in question doesn't contain
the gerrit refs (like opendev.org currently) then git fetch origin
won't work properly. However, we already calculate the correct
URL at the top of the function for use in query_reviews, and fetch
will happily work with a full url rather than a named remote.

Update the function to use the remote_url so that usepushurl works
even if the mirror url does not contain refs/changes.

Change-Id: Ib72afe97e65cb1dcaf95e28450dfe6e7d5f88965

Our company-internal Gerrit uses federated logins (Shibboleth/SAML), and
the login names happen to include the at-signs. This needed a patch to
Gerrit ([1], released in 2.13.10) to allow them. Cloning using regular
git tools as well as through the gertty work, but `git-review --list`
complained because it couldn't parse these URLs.

[1] https://gerrit-review.googlesource.com/c/94914/

Change-Id: I6c1c75a585184ee3fb2847c1e6d30802e53f8b4c

Adopts use of newer stestr for running the test suite.

Passing TERM is needed or stestr will ignore the --color option when
run under tox.

Change-Id: Id2d4ce36cdfedd5a847e0428854753d2cf4140ee

This allows user to define JAVA_HOME in order to control which version
of Java will be used by Gerrit during tox testing.

Avoid Gerrit startup failure due to use of wrong version of Java,
even if `java -version` would have reported the right version.

On systems with multiple versions of Java 8/9/10 it was impossible to
run tox because the launcher would have always picked the latest
version which is clearly not compatible with Gerrit war archive used
for testing.

In the future we will start using a gerrit container for testing and
avoid dependency on local Java version.

Change-Id: I05b6f7125393a009e1fdcf556b17df158c10eea0

When something is not found in git config, the --verbose option logs the
wrong value "None", example:

2018-12-12  Running: git config --get gitreview.branch
2018-12-12  using default: None
2018-12-12  Running: git config --get gitreview.scheme
2018-12-12  using default: None
2018-12-12  Running: git config --get remote.gerrit.pushurl
2018-12-12  using default: None
2018-12-12  Running: git config --get remote.gerrit.url
2018-12-12  result: ssh://marc@review.openstack.org:29418/openstack-infra/git-review

From a --verbose user perspective this is plain wrong for (at least) all the
options defined in the DEFAULTS list.

Change --verbose message to look like this instead:

2018-12-12  Running: git config --get gitreview.branch
2018-12-12  Config['branch'] = master
2018-12-12  Running: git config --get gitreview.scheme
2018-12-12  Config['scheme'] = ssh
2018-12-12  Running: git config --get remote.gerrit.pushurl
2018-12-12  Running: git config --get remote.gerrit.url
2018-12-12  ... remote.gerrit.url = ssh://marc@review.openstack.org:29418/openstack-infra/git-review

If git_config_get_value('new_option',... ) is ever invoked in the future
with a not None, default="fubar" parameter then --verbose will print these
lines:

2018-12-19  Running: git config --get gitreview.new_option
2018-12-19  ... nothing in git config, returning func parameter: fubar

This logging issue is especially misleading considering the many levels of
defaults and fallbacks: git config x3; .gitreview; DEFAULTS list,
git_config_get_value(default=...) parameter, options parser logic, etc.

Change-Id: I6cee46e88b90b8f11689be3875d64ec5e577f12f

... because it Just Works with absolutely zero configuration when cloning
with ssh://

Change-Id: I7a413704c4a567e397b5bb6075677175b8997704

Add: 'passenv = http_proxy https_proxy' whitelist to tox.ini because there's
basically never any good reason to throw away proxy configuration.

The addition of http_proxy allows downloading
http://tarballs.openstack.org/ci/gerrit/gerrit-<version>.war
behind a proxy.

https_proxy is not needed right now but added for consistency and
"future-proofing"

_JAVA_OPTIONS is added but only as a clue and commented out by default
because it could hold non-proxy configuration which would defeat tox'
environment cleaning and introduce test variability.

This is just the tox.ini source code part of the larger
I764ed51314c8e2d0572d1f01c9792bee0a40c7e6
includes documentation of workarounds and troubleshooting tips.

Change-Id: I764ed51314c8e2d0572d1f01c9792bee0a40c7e7

Paraphrasing parts of a private email from Stephen Finucane.

Change-Id: I2b66af40ca16e311bde3b482229ef00b70dcd1f7

Newer versions of git quote branch names in their output; at least git
2.16.3 does. This causes all test_uploads_with_nondefault_rebase tests to
fail like this:

testtools.matchers._impl.MismatchError:
 'Branch test_branch set up to track remote branch maint from origin.'
 not in
  u"Switched to a new branch 'test_branch'\n\
 Branch 'test_branch' set up to track remote branch 'maint' from 'origin'."

Add a testtools.matchers.MatchesRegex() to support both styles.

Change-Id: I9f1417c53de2f7d638e845f553df3bd426a4c750

From: https://www.openssh.com/txt/release-7.8 change log:

 * ssh-keygen(1): write OpenSSH format private keys by default
   instead of using OpenSSL's PEM format. The OpenSSH format,
   supported in OpenSSH releases since 2014 and described in the
   PROTOCOL.key file in the source distribution, offers substantially
   better protection against offline password guessing and supports
   key comments in private keys. If necessary, it is possible to write
   old PEM-style keys by adding "-m PEM" to ssh-keygen's arguments
   when generating or updating a key.

This fixes all tests failing with this error:

[2018-12-04 17:46:24,130] WARN  org.apache.sshd.common.keyprovider.FileKeyPairProvider :
   Unable to read key /home/mherber2/pip/src/git-review/.gerrit/site-5123/etc/ssh_host_rsa_key
java.io.IOException: unrecognised object: OPENSSH PRIVATE KEY
	at org.bouncycastle.openssl.PEMParser.readObject(Unknown Source)
	at org.apache.sshd.common.keyprovider.FileKeyPairProvider.doLoadKey(FileKeyPairProvider.java:124)

Change-Id: Id10f9b5be928f2ba57847fa32814e9db979375f5

- Fixes linting error related to undefined unicode on python3.
- Simplifies code

Change-Id: Ib40237e0fb7156b084dfdab47be69ccad09547fe

Python2 has default encoding as ascii which means
that is likely that some print() commands would
fails with UnicodeEncodeError.

This hack changes default encoding in order to
avoid such errors.

Change-Id: I4e21e6e32d4bb815693b7d6ce35efb6a5cca2fc2

Refactoring to improve code style as suggested by pep8.

- "rc" and "status" variables cannot be None beacause the process is
  terminated after the "run_command_status"

- It's better to ignore an empty "username"

Change-Id: I8a34f4ab9e05d91786ce6d62e4db96811787b633

Unnecessary dynamic GitReviewException.EXIT_CODE attribute
lookup refactored into static to benefit static analyzers.

Default process exit code changed from 127 (command not found error)
to 1 (general error)

Change-Id: I1fcb583a740bf32c4427a587e208d099712a7bc4

This is included as a note on the usage page for want of a better place
to put it.

Change-Id: I647cc8239e04e1f064aef3e0942ca94b0d5b4025
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
Story: 2000431
Task: 2663

Change-Id: I8f8880791ad7e46fb9e18623ab8bd295457424b2
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>

The feature to configure branch names based on the presence of "bug",
"lp", "blueprint" or "bp" in the commit message is overly elaborate and
very OpenStack specific. Even with this, it hasn't been updated to keep
up with the times as many projects have migrated to Storyboard, which
isn't handled here.

Given that it frequently does the wrong thing and likely doesn't apply
to anyone outside of the OpenStack ecosystem, the wisest thing it to
simply remove the feature. This is a break in behavior but it seems
better than adding yet another flag for something that many users will
want enabled by default.

Change-Id: I82ecc1719de5c87d59bbfe73a042917e6559da1e
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
Story: 1130330
Task: 566
Story: 2001247
Task: 5777

The sdist target is useless given that releases are automated.
Similarly, the '[testenv] install_command' uses the default value and is
therefore not required. Finally, some other things were slightly off or
otherwise annoying and can be fixed.

Change-Id: Ia029aba9ee5bc79cf39043a5fcaf9463aa1518f7
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>