Gergich is a command-line tool (and ruby lib) for easily posting comments on a Gerrit review from a CI environment. It can be wired up to linters (rubocop, eslint, etc.) so that you can get nice inline comments right on the Gerrit review. That way developers don't have to go digging through CI logs to see why their builds failed.
Gergich maintains a little sqlite db of any draft comments/labels/etc. for the current patchset (defined by revision+ChangeId). This way different processes can all contribute to the review. For example, various linters add inline comments, and when the CI build finishes, Gergich publishes the review to Gerrit.
Because everything is synchronized/stored in a local sqlite db, you
should only call Gergich from a single box/build per patchset unless you
have a unique GERGICH_COMMENT_PREFIX
set for each box/build per patchset.
Gergich does a check when publishing to ensure he hasn't already posted on
this patchset before (w/ the same GERGICH_COMMENT_PREFIX
); if he has,
publish will be a no-op. This protects against reposts (say, on a retrigger),
but it does mean that you shouldn't have completely different builds posting
Gergich comments on the same revision, unless you set up different
credentials for each.
Add the following to your Gemfile (perhaps in your :test
group?):
gem "gergich"
To use Gergich, you'll need a Gerrit user whose credentials it'll use
(ideally not your own). With your shiny new username and password in hand,
set GERGICH_USER
and GERGICH_KEY
accordingly in your CI environment.
Additionally, Gergich needs to know where your Gerrit installation
lives, so be sure to set GERRIT_BASE_URL
(e.g.
https://gerrit.example.com
) or GERRIT_HOST
(e.g. gerrit.example.com
).
Lastly, if you have no .git directory in CI land (say if you are building
in docker and want to keep your images small), you also need to set
GERRIT_CHANGE_ID
and GERRIT_PATCHSET_REVISION
. If you use Jenkins and
the gerrit-trigger plugin, typcially all GERRIT_*
vars will already be
set, it's just a matter of plumbing them down to docker.
Run gergich help
for detailed information about all supported commands.
In your build scripts, you'll typically be using gergich comment
,
gergich capture
and gergich publish
. Comments are stored locally in a
sqlite database until you publish. This way you can queue up comments from
many disparate processes. Comments are published to HEAD
's corresponding
patchset in Gerrit (based on Change-Id + <sha>
)
<comment_data>
is a JSON object (or array of objects). Each comment
object should have the following properties:
- path - the relative file path, e.g. "app/models/user.rb"
- position - either a number (line) or an object (range). If an object,
must have the following numeric properties:
- start_line
- start_character
- end_line
- end_character
- message - the text of the comment
- severity -
"info"|"warn"|"error"
- this will automatically prefix the comment (e.g."[ERROR] message here"
), and the most severe comment will be used to determine the overallCode-Review
score (0, -1, or -2 respectively)
Note that a cover message and Code-Review
score will be inferred from the
most severe comment.
gergich comment '{"path":"foo.rb","position":3,"severity":"error",
"message":"ಠ_ಠ"}'
gergich comment '{"path":"bar.rb","severity":"warn",
"position":{"start_line":3,"start_character":5,...},
"message":"¯\_(ツ)_/¯"}'
gergich comment '[{"path":"baz.rb",...}, {...}, {...}]'
For common linting formats, gergich capture
can be used to automatically
do gergich comment
calls so you don't have to wire it up yourself.
<format>
- One of the following:
brakeman
rubocop
eslint
i18nliner
flake8
stylelint
yamllint
shellcheck
- shellcheck json outputcustom:<path>:<class_name>
- file path and ruby class_name of a custom formatter.
<command>
- The command to run whose output conforms to <format>
.
Output from the command will still go to STDOUT, and Gergich will
preserve its exit status. If command is "-", Gergich will simply read
from STDIN and the exit status will always be 0.
To create a custom formatter, create a class that implements a run
method that takes a string of command output and returns an array of
comment hashes (see gergich comment
's <comment_data>
format), e.g.
class MyFormatter
def run(output)
output.scan(/^Oh noes! (.+?):(\d+): (.*)$/).map do |file, line, error|
{ path: file, message: error, position: line.to_i, severity: "error" }
end
end
end
gergich capture rubocop "bundle exec rubocop"
gergich capture eslint eslint
gergich capture i18nliner "rake i18nliner:check"
gergich capture shellcheck "shellcheck --format json build.sh"
gergich capture custom:./gergich/xss:Gergich::XSS "node script/xsslint"
docker-compose run --rm web eslint | gergich capture eslint -
# you might be interested in $PIPESTATUS[0]
Publish all draft comments/labels/messages for this patchset. no-op if there are none.
The cover message and Code-Review
label (e.g. -2) are inferred from the
comments, but labels and messages may be manually set (via gergich message
and gergich labels
)
Write tests of course, but also be sure to test it end-to-end via the
CLI... Run gergich
for a list of commands, as well as help for each
command. There's also a citest
thing that we run on our Jenkins that
ensures each CLI command succeeds, but it doesn't test all branches for
each command.
After running a given command, you can run gergich status
to see the
current draft of the review (what will be sent to Gerrit when you do
gergich publish
).
You can even do a test publish
to Gerrit, if you have valid Gerrit
credentials in GERGICH_USER
/ GERGICH_KEY
. It infers the Gerrit patchset
from the working directory, which may or may not correspond to something
actually in Gerrit, so YMMV. That means you can post to a Gergich commit
in Gerrit, or if you run it from another project's directory, you can post
to its Gerrit revision.