Peer reviewing is an important part of a programming curriculum, so of course RepoBee facilitates this! Like much of the other functionality in RepoBee, the peer review functionality is built around indirect access through teams with limited access privileges. In short, every student repo up for review gets an associated peer review team generated, which has pull
access to the repo. Each student then gets added to 0 < N < num_students
peer review teams, and are to open a peer review issue in the associated repos. This is at least the the default. See review allocation algorithm
for other available review allocation schemes.
The bulk of the work is performed by the reviews assign
. Most of its arguments it has in common with the other commands of RepoBee. The only non-standard arguments are --issue
and --num-reviews
, the former of which we've actually already seen in the issues open
command (see open
). We will assume that both --base-url
and --org-name
are already configured (if you don't know what this means, have a look at config
). Thus, the only things we must specify are --students/--students-file
and --num-reviews
(--issue
is optional, more on that later). Let's make a minimal call with the assign
action, and then inspect the log output to figure out what happened. Recall that students.txt
lists our three favorite students slarse, glassey and glennol (see setup
).
$ repobee reviews assign -a task-1 --sf students.txt --num-reviews 2
# Output nabbed from the log file, this will not appear on stdout
# step 1
[INFO] Created team slarse-task-1-review
[INFO] Created team glennol-task-1-review
[INFO] Created team glassey-task-1-review
# step 2
[INFO] Adding members glennol, glassey to team slarse-task-1-review
[INFO] Adding members glassey, slarse to team glennol-task-1-review
[INFO] Adding members slarse, glennol to team glassey-task-1-review
# steps 3 and 4, interleaved
[INFO] Opened issue glennol-task-1/#1-'Peer review'
[INFO] Adding team glennol-task-1-review to repo glennol-task-1 with 'pull' permission
[INFO] Opened issue glassey-task-1/#2-'Peer review'
[INFO] Adding team glassey-task-1-review to repo glassey-task-1 with 'pull' permission
[INFO] Opened issue slarse-task-1/#2-'Peer review'
[INFO] Adding team slarse-task-1-review to repo slarse-task-1 with 'pull' permission
The following steps were performed:
- One review team per repo was created (
<student>-task-1-review
). - Two students were added to each review team. Note that these allocations are _random. For obvious resons, there can be at most
num_students-1
peer reviews per repo. So, in this case, we are at the maximum. - An issue was opened in each repo with the title
Peer review
, and a body saying something likeYou should peer review this repo.
. The review team students were assigned to the issue as well (although this is not apparent from the logging). - The review teams were added to their corresponding repos with
pull
permission. This permission allows members of the team to view the repo and open issues, but they can't push to (and therefore can't modify) the repo.
That's it for the basic functionality. The intent is that students should open an issue in every repo they are to peer review, with a specific title. The issues can then be searched by title, and the check
action can find which students have opened issues in the repositories they've been assigned to review. Now, let's talk a bit about that --issue
argument.
Important
Assigning peer reviews gives the reviewers read-access to the repos they are to review. This means that if you use issues to communicate grades/feedback to your students, the reviewers will also see this feedback! It is therefore important to remove the peer review teams (see purge peer review
teams
).
The default issue is really meant to be replaced with something more specific to the course and assignment. For example, say that there were five tasks in the task-1
repo, and the students should review tasks 2 and 3 based on some criteria. It would then be beneficial to specify this in the peer review issue, so we'll write up our own little issue to replace the default one. Remember that the first line is taken to be the title, in exactly the same way as issue files are treated in open
.
Review of task-1
Hello! The students assigned to this issue have been tasked to review this
repo. Each of you should open _one_ issue with the title `Peer review` and
the following content:
## Task 2
### Code style
Comments on code style, such as readability and general formatting.
### Time complexity
Is the algorithm O(n)? If not, try to figure out what time complexity it is
and point out what could have been done better.
## Task 3
### Code style
Comments on code style, such as readabilty and general formatting.
Assuming the file was saved as issue.md
, we can now run the command specifying the issue like this:
$ repobee reviews assign -a task-1 --sf students.txt --num-reviews 2 --issue issue.md
This will have the same effect as last time, but with the custom issue being opened instead.
The check
action provides a quick and easy way of checking which students have performed their reviews. You provide it with the same information that you do for assign
, but additionally also provide a regex to match against issue titles. The command then finds all of the associated review teams, and checks which students have opened issues with matching titles in their alloted repositories. Of course, this says nothing about the content of those issues: it only checks that the issues have been opened at all. --num-reviews
is also required here, as it is used as an expected value for how many reviews each student should be assigned to review. It is a simple but fairly effective way of detecting if students have simply left their review teams. Here's an example call:
$ repobee reviews check -a task-1 --sf students.txt --num-reviews 2 --title-regex '\APeer review\Z'
reviewer num done num remaining repos remaining
glennol 0 2 glassey-task-1,slarse-task-1
slarse 2 0
glassey 0 2 glennol-task-1,slarse-task-1
The output is color-coded in the terminal, making it easier to parse. We make use of this when doing peer reviews in a classroom settings, as it allows us to quickly check which students are done without having to ask them out loud every five minutes. The next command lets you clean up review teams and thereby revoke reviewers' read access once reviews are over and done with.
Hint
Use the issues list
command with the --title-regex
(with a regex matching the review issue title) and --show-body
options to actually check the contents of the students' review issues.
The one downside of using teams for access privileges is that we bloat the organization with a ton of teams. Once the deadline has passed and all peer reviews are done, there is little reason to keep them. It can also often be a good idea to revoke the reviewers' access to reviewed repos if you yourself plan to provide feedback on the issue tracker, so as not to let the reviewers see it. Therefore, the end
action can be used to remove all peer review teams for a given set of student repos, both cleaning up the organization and revoking reviewers' read access. Let's say that we're completely done with the peer reviews of task-1
, and want to remove the review teams. It's as simple as:
$ repobee reviews end -a task-1 --sf students.txt
# Progress bars will show how many teams have been deleted thus far
Warning
The end
action deletes review allocations created by assign
. This is an irreversible action. You cannot run check
after running end
for any given set of student repos, and there is no functionality for retrieving deleted review allocations. Only use end
when reviews are truly done, and you have collected what results you need. If being able to backup and restore review allocations is something you need, please open an issue with a feature request on the issue tracker.
And that's it, the review teams are gone. If you also want to close the related issues, you can simply use the issues close
command for that (see close
). The end
action plays one more important role; if you mess something up when assigning the peer reviews. The next section details how you can deal with such a scenario.
Let's say you messed something up with allocating the peer reviews. For example, if you left out a student, there is no easy way to rectify the allocations such that that student is included. Let's say we did just that, and forgot to include the student cabbage
in the reviews for task-1
back at assign reviews
. We then do the following:
- Check if any reviews have already been posted. This can easily be performed with
repobee reviews check -a task-1 --sf students.txt -r '^Peer review$' --num-reviews 2
(assuming the naming conventions were followed!). Take appropriate action if you find any reviews already posted (appropriate being anything you see fit to alleviate the situation of affected students possibly being assigned new repos to review). - Delete the review teams with
repobee reviews end -a task-1 --sf students.txt
- Close all review issues with
repobee issues close -a task-1 --sf students.txt -r '^Review of task-1$'
- Create a new
issue.md
file apologetically explaining that you messed up:
Review of task-1 (for real this time!)
Sorry, I messed up with the allocations previously. Disregard the previous
allocations (repo access has been revoked anyway).
- Assign peer reviews again, with the new issue, with
repobee reviews assign -a task-1 --sf students.txt --num-reviews 2 --issue issue.md
And that's it! Disaster averted.
The default allocation algorithm is as described in peer review
, and is suitable for when reviewers do not need to interact with the students whom they review. This is however not always the case, sometimes it is beneficial for reviewers to to interact with reviewees (is that a word?), especially if the peer review is done in the classroom. Because of this, RepoBee also provides a _pairwise allocation scheme, which allocates reviews such that if student A
reviews student B
, then student B
reviews student A
(except for an A->B->C->A
kind of deal in one group if there are an odd amount of students). This implemented as a plugin, so to run with this scheme, you add -p pairwise
in front of the command.
$ repobee -p pairwise reviews assign -a task-1 --sf students.txt
Note that the pairwise algorithm ignores the --num-reviews
argument, and will issue a warning if this is set (to anything but 1, but you should just not specify it). For more details on plugins in RepoBee, see plugins
.
RepoBee 3.6 adds experimental support for double-blind peer review. The user experience is not finalized, but the functionality is all there. This section provides a walkthrough for how to assign double-blind peer review. It assumes that you've read through the prior sections of the peer review documentation.
The general idea of the double-blind peer review is to assign reviewers to review copies of their peers' repositories. The whole procedure is something like this:
reviews assign
Create copies of all student repositories under review and assign reviewers to them- The commit history is anonymized
- The repository name is anonymized
reviews check
Verify that students have performed their reviewsissues list
Collect reviews from anonymous repos and store them locallyissues open
Distribute anonymously submitted reviews to original repositories- They are opened with your user account, so as long as reviewers haven't put their names in the reviews they will be anonymous!
(Optional)
reviews end
Delete repo copies and associated review teams- Always run
issues list
to collect the reviews before runningreviews end
, or all reviews will be lost!
- Always run
As you may note, this is the same sequence of commands as for no-blind review, except that issues list
and issues open
are sprinkled into the middle. Usage of all commands shown is as usual, with they key exception that you'll be providing them with a secret key for the anonymization.
I order to run reviews assign
in double-blind mode, all you need to do in addition to the no-blind usage is to supply the --double-blind-key
argument.
$ repobee reviews assign -a task-1 --sf students.txt --double-blind-key SUPER_SECRET_KEY
The key is a secret, do not share it with the students. After assigning reviews with a given key, you must also remember or otherwise store that key until those reviews are closed, or you will be unable to interact with the anonymous repos.
Important
The double-blind key is a secret. Given the key, all repositories can be deanonymized.
Important
For each review, you must remember or store the key until reviews are closed. Otherwise, you can't deanonymize the repos, and consequently can't collect and distribute reviews.
Important
If you run double-blind reviews assign
for with --num-reviews
larger than 1
, reviewers reviewing the same repository will be able to see each others' reviews.
Just like with reviews assign
, the only thing you need to add in addition to normal usage is the --double-blind-key
argument.
$ repobee reviews check \
--assignments task-1 \
--sf students.txt \
--num-reviews 1 \
--title-regex '\APeer review\Z' \
--double-blind-key SUPER_SECRET_KEY
The repositories are deanonymized, and the output looks precisely like that of no-blind review. Needless to say, your students should not be shown this output.
SUPER_SECRET_KEY
must match the key you supplied to reviews assign
.
Once you've verified that the students have performed their reviews with reviews check
, you can collect reviews with issues list
. Here, you need to specify two arguments out of the ordinary: --double-blind-key
with your secret key, as well as --hook-results-file
to store the issues locally. To collect only the reviews, with title "Peer review", the command could look like so.
$ repobee issues list \
--assignments task-1 \
--sf students.txt \
--title-regex '\APeer review\Z' \
--hook-results-file results.json \
--double-blind-key SUPER_SECRET_KEY
By specifying the title regex your students use for review, you don't collect the instructions. If you'd like to also collect and distribute the instructions to the original repos, you can either use a carefully crafted regex for it, or simply provide the empty strigle (i.e. --title-regex ""
), which will match any issue.
Note that you can now also browse the reviews before distribution by viewing the results.json
file.
In order for students to actually be able to read the reviews by their peers, the issues need to be distributed to the original repos. To do this, execute issues open
as per usual, but supply --hook-results-file
instead of --issue
.
$ repobee issues open \
--assignments task-1 \
--sf students.txt \
--hook-results-file results.json
Note that you do not need the key here: the issues in the hook results file are already deanonymized.
Important
If using GitHub, your access token must have the delete_repo
scope in order to run this command.
reviews end
is a cleanup command. When doing no-blind peer review, it's often necessary to run it as students otherwise maintain read access to their peers' repositories, and may then be able to view feedback from teachers or TAs. With double-blind reviews, this isn't the case as the reviewers only get access to copies of the reviewed repositories. However, it does leave quite a mess of repositories and review teams with strange names, so cleaning up may be desirable. If you want to do that, simply run reviews end
and supply your key.
$ repobee reviews end \
--assignments task-1 \
--sf students.txt \
--double-blind-key SUPER_SECRET_KEY
Danger
Running reviews end
irrevocably destroys all traces of the reviews, including deleting the anonymous repositories and review teams. Make sure to collect reviews with issues list
before doing this.
And that's pretty much all there is to double-blind review with RepoBee!