This file outlines the policy on reviews for the Mercury system.
All changes to the Mercury repository, including the compiler, documentation, www pages, library predicates, runtime system, and tools need to be reviewed.
git diff
or by creating series of commits and using git format-patch
(see the git documentation for details).
You will need to edit the files that format-patch generates
to modify the subject line (see below),
and to duplicate this line within the e-mail's body.
Use git add -N on any new files
to include them in the diff.
Review the diff and log message yourself. (see below)
E-mail the changes to us at
Mercury Reviews <reviews@lists.mercurylang.org>
The subject should be
"for review: <short description of change>",
the short description is typically the first line of your git log message.
Optionally nominate a reviewer in your email.
Optionally describe which branches your change should be applied to.
(If this change has been reviewed once before,
it might fall into the "commit before review" category --
see the section on exceptions).
If you generated your diffs using git format-patch
then you can use git send-email
to mail them,
or attach them to an e-mail normally.
It is also possible to use github pull-requests, however we prefer e-mail.
Wait for review (see below).
Fix any changes suggested.
Repeat the above steps until approval.
Push the change(s) (see below) if you have repository permissions.
If you do not, the developer who reviewed your change will help you with this.
Log message template
Use this template for each change's log message.
<one-line description (subject)>
<overview or general description of changes>
<directory>/<file>:
<detailed description of changes>
The description should state why the changes were made,
not just what the changes were.
All file modifications related to the same change
should be committed together.
For very small changes,
the <overview or general description> can be omitted,
but the other descriptions should stay.
If adding a new feature, this is a good place
to describe the feature, how it works, how to turn it on and off,
and any present limitations of the feature
(note that all this should also be documented within the change, as well).
If fixing a bug, describe both the bug and the fix.
If the bug is in a bug tracker, also include the bug's number for reference.
However, the bug number should not be
in the one line description at the top.
Self-review
You should also review your own code first, and fix any obvious mistakes.
if there was something you had to understand when making the change,
and it was not documented before,
then document it as part of the change.
This makes it easier to review the change,
and should make future similar changes easier as well.
Review
We post all diffs to reviews@lists.mercurylang.org.
The reasons for posting to reviews are:
-
To increase everyone's awareness of what changes are taking place.
-
Give everyone interested a chance to review your code,
not just a nominated reviewer.
Remember, your changes may impact upon the uncommitted work of others,
so they may want to give input.
-
Allow other people to read the reviewer's comments -
so the same problems don't have to be explained again and again.
-
People can try to see how your changes worked
without having to figure out how to get git to generate the right set of diffs.
-
Important decisions are often made or justified in reviews,
so these should be recorded.
You should try to match the reviewer to the code -
someone familiar with a section of code
can review faster and is more likely to catch errors.
Put a preamble at the start of your diff
to nominate who you would like to review the diff.
Waiting and approval
Waiting for approval need not be wasted time.
This is a good time to start working on something else,
clean up unused workspaces, etc.
In particular, you might want to run long running tests
that have not yet been run on the your change
(different grades, different architectures, different optimisation levels,
and so on).
The reviewer(s) should reply, indicate any problems that need to be corrected,
and whether the change can be committed yet.
Design issues may need to be fully justified before you commit.
You may need to fix the problems,
then go through another iteration of the review process,
or you may be able to just fix a few small problems, then commit.
Pushing changes
Use the log message you prepared for the review
when committing and pushing changes,
updated as necessary to reflect the changes you made as a result of the review.
Exceptions: push before review
The only time changes should be pushed before being reviewed
is when they satisfy all of the following conditions:
-
The change is simple.
-
You are absolutely sure the change will not introduce bugs.
-
You are sure that the change will pass review
with only trivial corrections (spelling errors in comments, etc).
-
Your change contains no new design decisions
or changes to previous design decisions.
The status quo should be the default;
you must convince the reviewer(s) of the validity of your design decisions
before the code is committed.
-
You will be around the next day or two to fix the bugs
that you were sure could never happen.
-
Committing it now will make life significantly easier
for you or someone else in the group.
If the compiler is already broken (i.e. it doesn't pass its nightly tests),
and your change is a bug fix,
then it is not so important to be absolutely sure
that your change won't introduce bugs.
You should still be careful, though.
Make sure you review the diffs yourself.
Similarly, if the code you are modifying is a presently unused part of code -
for example a new feature that nobody else is using yet,
that is switchable and is switched off by default,
or a new tool,
or an `under development' webpage that is not linked to by other webpages yet,
the criteria are a bit looser.
Don't use this one too often - only for small changes.
You don't want to go a long way down the wrong track with your new feature,
before finding that there is a much better way.
If these conditions are satisfied,
then there shouldn't be any problem with mailing the diff,
then committing, then fixing any problems that come up afterwards,
provided you are pretty sure everything will be okay.
This is particularly true if others are waiting for your work.
Usually, a change that has already been reviewed falls into this category,
provided you have addressed the reviewers comments,
and there are no disputes over design decisions.
If the reviewer has specifically asked for another review,
or there were a large number of comments at the review,
you should not commit before a second review.
If you are going to commit before the review, use the subject line:
"for post-commit review: <short description of change>".
Exceptions: no review
The only time changes should be committed without review by a second person
is when they satisfy all of the following conditions:
-
It is a very small diff that is obviously correct, such as
- fix spelling or grammar errors
- fix syntax errors you accidentally introduced
These usually don't need to be reviewed by a second person.
Make sure that you review your own changes, though.
Also make sure your log message is more informative than "fixed a typo";
try "s/foo/bar" or something so that
if you did make a change that people don't approve of,
at least it's seen quickly.
-
It does not change anything publicly visible
such as Web pages, documentation, library, or man pages.
Changes to publicly visible things should always be reviewed.
It is just too easy to make spelling errors,
write incorrect information, etc.
This stuff reflects on the whole group, so it shouldn't be ignored.
If your change falls into this category,
you should still send the diff and log message to the reviews mailing list,
but use the subject line:
"diff: <short description of change>".