Rohan McGovern
and 1 contributors


Gerrit::Client - interact with Gerrit code review tool


  use AnyEvent;
  use Gerrit::Client qw(stream_events);

  # alert me when new patch sets arrive in
  # ssh://
  my $stream = stream_events(
    url => 'ssh://',
    on_event => sub {
      my ($event) = @_;
      if ($event->{type} eq 'patchset-added'
          && $event->{change}{project} eq 'myproject') {
        system("xmessage", "New patch set arrived!");

  AE::cv()->recv(); # must run an event loop for callbacks to be activated

This module provides some utility functions for interacting with the Gerrit code review tool.

This module is an AnyEvent user and may be used with any event loop supported by AnyEvent.


stream_events ssh_url => $gerrit_url, ...

Connect to "gerrit stream-events" on the given gerrit host and register one or more callbacks for events. Returns an opaque handle to the stream-events connection; the connection will be aborted if the handle is destroyed.

$gerrit_url should be a URL with ssh schema referring to a valid Gerrit installation (e.g. "ssh://").

Supported callbacks are documented below. All callbacks receive the stream-events handle as their first argument.

on_event => $cb->($data)

Called when an event has been received. $data is a reference to a hash representing the event.

See the Gerrit documentation for information on the possible events.

on_error => $cb->($error)

Called when an error occurs in the connection. $error is a human-readable string.

Examples of errors include network disruptions between your host and the Gerrit server, or the ssh process being killed unexpectedly. Receiving any kind of error means that some Gerrit events may have been lost.

If this callback returns a true value, stream_events will attempt to reconnect to Gerrit and resume processing; otherwise, the connection is terminated and no more events will occur.

The default error callback will warn and return 1, retrying on all errors.

for_each_patchset(ssh_url => $ssh_url, workdir => $workdir, ...)

Set up a high-level event watcher to invoke a custom callback or command for each existing or incoming patch set on Gerrit. This method is suitable for performing automated testing or sanity checks on incoming patches.

For each patch set, a git repository is set up with the working tree and HEAD set to the patch. The callback is invoked with the current working directory set to the top level of this git repository.

Returns a guard object. Event processing terminates when the object is destroyed.



The Gerrit ssh URL, e.g. ssh:// May also be specified as 'url' for backwards compatibility. Mandatory.

http_auth_cb => $sub->($response_headers, $request_headers)

These arguments have the same meaning as for the review function. Provide them if you want to post reviews via REST.


The top-level working directory under which git repositories and other data should be stored. Mandatory. Will be created if it does not exist.

The working directory is persistent across runs. Removing the directory may cause the processing of patch sets which have already been processed.

on_patchset => $sub->($change, $patchset)
on_patchset_fork => $sub->($change, $patchset)
on_patchset_cmd => $sub->($change, $patchset) | $cmd_ref

Callbacks invoked for each patchset. Only one of the above callback forms may be used.

  • on_patchset invokes a subroutine in the current process. The callback is blocking, which means that only one patch may be processed at a time. This is the simplest form and is suitable when the processing for each patch is expected to be fast or the rate of incoming patches is low.

  • on_patchset_fork invokes a subroutine in a new child process. The child terminates when the callback returns. Multiple patches may be handled in parallel.

    The caveats which apply to AnyEvent::Util::run_cmd also apply here; namely, it is not permitted to run the event loop in the child process.

  • on_patchset_cmd runs a command to handle the patch. Multiple patches may be handled in parallel.

    The argument to on_patchset_cmd may be either a reference to an array holding the command and its arguments, or a reference to a subroutine which generates and returns an array for the command and its arguments.

    Note: since on_patchset_cmd has no way to return a value, it can't be used to generate ReviewInput objects for REST-based reviewing. See below.

All on_patchset callbacks receive change and patchset hashref arguments. Note that a change may hold several patchsets.

Output and the returned value of the callbacks may be used for posting a review back to Gerrit; see documentation of the review argument below.

on_error => $sub->($error)

Callback invoked when an error occurs. $error is a human-readable error string.

All errors are treated as recoverable. To abort on an error, explicitly undefine the loop guard object from within the callback.

By default, a warning message is printed for each error.

review => 0 | 1 | 'code-review' | 'verified' | ...

If false (the default), patch sets are not automatically reviewed (though they may be reviewed explicitly within the on_patchset_... callbacks).

If true, patch sets are automatically reviewed according to the following:

  • If an on_patchset_... callback returned a ReviewInput hashref, a review is done via REST (see the review function). Note that on_patchset_cmd doesn't support returning a value.

  • If an on_patchset_... callback printed text on stdout/stderr, a review is done via SSH using the text as the top-level review message.

  • If both of the above are true, the review is attempted by REST first, and then via SSH if the REST review failed. Therefore, writing a callback to both print out a meaningful message and return a ReviewInput hashref enables a script to be portable across Gerrit versions with and without REST.

See the REST vs SSH discussion in the documentation for the review function for more information on choosing between REST or SSH.

If a string is passed, it is construed as a Gerrit approval category and a review score will be posted in that category. The score comes from the return value of the callback (or exit code in the case of on_patchset_cmd). The returned value must be an integer, so this cannot be combined with reviewing by REST, which requires callbacks to return a hashref.

on_review => $sub->( $change, $patchset, $message, $score, $returnvalue )

Optional callback invoked prior to performing a review (when the `review' option is set to a true value).

The callback should return a true value if the review should be posted, false otherwise. This may be useful for the implementation of a dry-run mode.

The callback may be invoked with an undefined $message and $score, which indicates that a patchset was successfully processed but no message or score was produced.

wanted => $sub->( $change, $patchset )

The optional `wanted' subroutine may be used to limit the patch sets processed.

If given, a patchset will only be processed if this callback returns a true value. This can be used to avoid git clones of unwanted projects.

For example, patchsets for all Gerrit projects under a 'test/' namespace could be excluded from processing by the following:

    wanted => sub { $_[0]->{project} !~ m{^test/} }
git_work_tree => 0 | 1

By default, while processing a patchset, a git work tree is set up with content set to the appropriate revision.

git_work_tree => 0 may be passed to disable the work tree, saving some time and disk space. In this case, a bare clone is used, with HEAD referring to the revision to be processed.

This may be useful when the patch set processing does not require a work tree (e.g. the incoming patch is directly scanned).

Defaults to 1.

query => $query | 0

The Gerrit query used to find the initial set of patches to be processed. The query is executed when the loop begins and whenever the connection to Gerrit is interrupted, to avoid missed patchsets.

Defaults to "status:open", meaning every open patch will be processed.

Note that the query is not applied to incoming patchsets observed via stream-events. The wanted parameter may be used for that case.

If a false value is passed, querying is disabled altogether. This means only patchsets arriving while the loop is running will be processed.


Returns a random Change-Id (the character 'I' followed by 40 hexadecimal digits), suitable for usage as the Change-Id field in a commit to be pushed to gerrit.


Returns the 'next' Change-Id which should be used for a commit created by the current git author/committer (which should be set by git_environment prior to calling this method). The current working directory must be within a git repository.

This method is suitable for usage within a script which periodically creates commits for review, but should have only one outstanding review (per branch) at any given time. The returned Change-Id is (hopefully) unique, and stable; it only changes when a new commit arrives in the git repository from the current script.

For example, consider a script which is run once per day to clone a repository, generate a change and push it for review. If this function is used to generate the Change-Id on the commit, the script will update the same change in gerrit until that change is merged. Once the change is merged, next_change_id returns a different value, resulting in a new change. This ensures the script has a maximum of one pending review any given time.

If any problems occur while determining the next Change-Id, a warning is printed and a random Change-Id is returned.

git_environment(name => $name, email => $email, author_only => [0|1] )

Returns a copy of %ENV modified suitably for the creation of git commits by a script/bot.



The human-readable name used for git commits. Mandatory.


The email address used for git commits. Mandatory.


If 1, the environment is only modified for the git author, and not the git committer. Depending on the gerrit setup, this may be required to avoid complaints about missing "Forge Identity" permissions.

Defaults to 0.

When generating commits for review in gerrit, this method may be used in conjunction with "next_change_id" to ensure this bot has only one outstanding change for review at any time, as in the following example:

    local %ENV = git_environment(
        name => 'Indent Bot',
        email => '',

    # fix up indenting in all the .cpp files
    (system('indent *.cpp') == 0) || die 'indent failed';

    # then commit and push them; commits are authored and committed by
    # 'Indent Bot <>'.  usage of next_change_id()
    # ensures that this bot has a maximum of one outstanding change for
    # review
    my $message = "Fixed indentation\n\nChange-Id: ".next_change_id();
    (system('git add -u *.cpp') == 0)
      || die 'git add failed';
    (system('git', 'commit', '-m', $message) == 0)
      || die 'git commit failed';
    (system('git push gerrit HEAD:refs/for/master') == 0)
      || die 'git push failed';
review $revision, ssh_url => $ssh_url, http_url => $http_url ...

Post a gerrit review, either by the `gerrit review' command using ssh, or via REST.

$revision is mandatory, and should be a git commit in full 40-digit form. (Actually, a few other forms of $revision are accepted, but they are deprecated - unabbreviated revisions are the only input which work for both SSH and REST.)

$ssh_url should be a URL with ssh schema referring to a valid Gerrit installation (e.g. "ssh://"). The URL may optionally contain the relevant gerrit project.

$http_url should be an HTTP or HTTPS URL pointing at the base of a Gerrit installation (e.g. "").

At least one of $ssh_url or $http_url must be provided.

    REST vs SSH

    In Gerrit 2.6, it became possible to post reviews onto patch sets using a new REST API. In earlier versions of gerrit, only the `gerrit review' command may be used.

    Reviewing by REST has a significant advantage over `gerrit review': comments may be posted directly onto the relevant lines of a patch. `gerrit review' in contrast only supports setting a top-level message.

    Gerrit::Client supports posting review comments both by REST and `gerrit review'. Most scripts will most likely prefer to use REST due to the improved review granularity.

    To review by REST:

    • Provide the http_url argument, and most likely http_username and http_password.

    • Provide the project and branch arguments.

    • Provide the reviewInput argument as a hash with the correct structure. (See documentation below.)

    To review by SSH:

    • Provide the ssh_url argument.

    • Provide the message argument, and one or more of the score-related arguments.

    To review by REST where supported, with fallback to SSH if REST is unavailable (e.g on Gerrit < 2.6), provide all of the arguments listed above. message will only be used if reviewing by REST was not available.

Other arguments to this method include:

http_auth_cb => $sub->($response_headers, $request_headers)

A callback invoked when HTTP authentication to Gerrit is required. The callback receives HTTP headers from the 401 Unauthorized response in $response_headers, and should add corresponding request headers (such as Authorization) into $request_headers.

$response_headers also includes the usual AnyEvent::HTTP psuedo-headers, and one additional psuedo-header, Method, which is the HTTP method being used (i.e. "POST").

The callback must return true if it was able to provide the necessary authentication headers. It should return false if authentication failed.

See also http_username, http_password and Gerrit::Client::http_digest_auth, to use the Digest-based authentication used by default in Gerrit.


Passing these values is a shortcut for:

  http_auth_cb => Gerrit::Client::http_digest_auth($username, $password)

... which uses Gerrit's default Digest-based authentication.

reviewInput => $hashref

A ReviewInput object as used by Gerrit's REST API.

This is a simple data structure of the following form:

    message => "Some nits need to be fixed.",
    labels => {
      'Code-Review' => -1
    comments => {
      'gerrit-server/src/main/java/com/google/gerrit/server/project/' => [
          line => 23,
          message => '[nit] trailing whitespace'
          line => 49,
          message => "[nit] s/conrtol/control"

This value is only used if reviewing via REST.

message => $string

Top-level review message.

This value is only used if reviewing via SSH, or if REST reviewing was attempted but failed.

project => $string
branch => $string

Project and branch values for the patch set to be reviewed.

If reviewing via SSH, these are only necessary if the patch set can't be unambiguously located by its git revision alone. If reviewing via REST, these are always necessary.

In any case, it's recommended to always provide these.

on_success => $cb->()
on_error => $cb->( $error )

Callbacks invoked when the operation succeeds or fails.

abandon => 1|0
restore => 1|0
stage => 1|0
submit => 1|0
code_review => $number
sanity_review => $number
verified => $number

Most of these options are passed to the `gerrit review' command. For information on their usage, please see the output of `gerrit review --help' on your gerrit installation, or see the Gerrit documentation.

Note that certain options can be disabled on a per-site basis. `gerrit review --help' will show only those options which are enabled on the given site.

query $query, ssh_url => $gerrit_url, ...

Wrapper for the `gerrit query' command; send a query to gerrit and invoke a callback with the results.

$query is the Gerrit query string, whose format is described in the Gerrit documentation. "status:open age:1w" is an example of a simple Gerrit query.

$gerrit_url is the URL with ssh schema of the Gerrit site to be queried (e.g. "ssh://"). If the URL contains a path (project) component, it is ignored.

All other arguments are optional, and include:

on_success => $cb->( @results )

Callback invoked when the query completes.

Each element of @results is a hashref representing a Gerrit change, parsed from the JSON output of `gerrit query'. The format of Gerrit change objects is described in the Gerrit documentation.

on_error => $cb->( $error )

Callback invoked when the query command fails. $error is a human-readable string describing the error.

all_approvals => 0|1
comments => 0|1
commit_message => 0|1
current_patch_set => 0|1
dependencies => 0|1
files => 0|1
patch_sets => 0|1
submit_records => 0|1

These options are passed to the `gerrit query' command and may be used to increase the level of information returned by the query. For information on their usage, please see the output of `gerrit query --help' on your gerrit installation, or see the Gerrit documentation.

quote $string

Returns a copy of the input string with special characters escaped, suitable for usage with Gerrit CLI commands.

Gerrit commands run via ssh typically need extra quoting because the ssh layer already evaluates the command string prior to passing it to Gerrit. This function understands how to quote arguments for this case.

Note: do not use this function for passing arguments to other Gerrit::Client functions; those perform appropriate quoting internally.

http_digest_auth($username, $password)

Returns a callback to be used with REST-related Gerrit::Client functions. The callback enables Digest-based HTTP authentication with the given credentials.

Note that only the Digest scheme used by Gerrit (as of 2.8) is supported: algorithm = MD5, qop = auth.



The ssh command and initial arguments used when Gerrit::Client invokes ssh.

  # force IPv6 for this connection
  local @Gerrit::Client::SSH = ('ssh', '-oAddressFamily=inet6');
  my $stream = Gerrit::Client::stream_events ...

The default value is ('ssh').


The git command and initial arguments used when Gerrit::Client invokes git.

  # use a local object cache to speed up initial clones
  local @Gerrit::Client::GIT = ('env', "GIT_ALTERNATE_OBJECT_DIRECTORIES=$ENV{HOME}/gitcache", 'git');
  my $guard = Gerrit::Client::for_each_patchset ...

The default value is ('git').


Maximum number of simultaneous git connections Gerrit::Client may make to a single Gerrit server. The amount of parallel git clones and fetches should be throttled, otherwise the Gerrit server may drop incoming connections.

The default value is 2.


Maximum number of processes allowed to run simultaneously for handling of patchsets in for_each_patchset. This limit applies only to local work processes, not git clones or fetches from gerrit.

Note that $AnyEvent::Util::MAX_FORKS may also impact the maximum number of processes. $AnyEvent::Util::MAX_FORKS should be set higher than or equal to $Gerrit::Client::MAX_FORKS.

The default value is 4.


If set to a true value, various debugging messages will be printed to standard error. May be set by the GERRIT_CLIENT_DEBUG environment variable.


Rohan McGovern, <>


Gerrit::Client has been known to work with Gerrit, Gerrit 2.6-rc1, and Gerrit 2.8.5 and hence could reasonably be expected to work with any gerrit version in that range.

Please note that different Gerrit versions may represent objects in slightly incompatible ways (e.g. "CRVW" vs "Code-Review" strings in event objects). Gerrit::Client does not insulate the caller against these changes.


Please use to view or report bugs.

When reporting a reproducible bug, please include the output of your program with the environment variable GERRIT_CLIENT_DEBUG set to 1.


Copyright (C) 2012-2014 Rohan McGovern <>

Copyright (C) 2012 Digia Plc and/or its subsidiary(-ies)

This program is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License version 2.1 as published by the Free Software Foundation.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details.

You should have received a copy of the GNU Lesser General Public License along with this program; if not, write to the Free Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.