- 1 Tutorial: your first patch
- 2 Overview of the JS engine
- 3 Collaboration and teamwork
- 4 Code considerations
Tutorial: your first patch
The first step to getting involved with SpiderMonkey is to make your first patch. This guides you through it, and at the end you should have learnt a lot of the procedures and formalisms involved in getting things done here.
We'll assume you're on a Unix-y platform, and that you know what you're doing. We'll ignore nearly all details.
Get the code
Spidermonkey development happens in the "mozilla-central" mercurial repository:
hg clone http://hg.mozilla.org/mozilla-central spidermonkey
Build the js shell
cd spidermonkey/js/src cp configure.in configure && chmod +x configure # or autoconf2.13 or autoconf-2.13 mkdir build_DBG.OBJ cd build_DBG.OBJ ../configure --enable-debug --disable-optimize make # or make -j8 cd ..
If you're having trouble or are missing dependencies, refer to Building SpiderMonkey Tip.
At this point, you're ready to make your first fix.
TODO: something useless. Maybe adding a datemicrosecond function.
Building your changes
Having made the change, build the shell again.
make -C build_DBG.OBJ
Testing your changes
It builds. Hurray. Time to run the tests to check you haven't broken anything.
(The location of the shell has changed, it used to be build_DBG.OBJ/js)
The jit-tests are pretty quick, and should give you an idea if you've gotten something wrong. Assuming nothing goes wrong, try the ref-tests:
Note that some of these tests fail depending on the timezone and locale they're run in, so if you get any errors that don't seem related to your changes, re-run the tests with a shell build without your changes applied and compare the results.
Benchmark your changes
Tests all pass. Congrats, you didn't break anything. Now, did you make it faster or slower? We benchmark using the v8 and SunSpider benchmarks. Get the benchmarks:
And now run them:
cd SunSpider ./sunspider --shell=../build_DBG.OBJ/dist/bin/js --run=30 --suite=sunspider-0.9.1 cd ..
Whoops, we benchmarked the debug version. Let's make an optimized build to test instead.
mkdir build_OPT.OBJ cd build_OPT.OBJ ../configure --disable-debug --enable-optimize make cd ..
Repeat the sunspider steps above, and take note of the line that looks like:
Results are located at sunspider-0.9.1-results/sunspider-results-2010-08-07-17.56.48.js
You'll need that later.
Baseline version / Mercurial Queues
We need to time our optimized version against the baseline version. This calls for a brief introduction to mercurial queues, which used to be the way most people managed their SpiderMonkey workflow (though these days many people use native hg, native hg + evolution, or git):
hg qinit hg qnew my_first_patch -f hg qrefresh
This puts your current work into a patch, managed by Mercurial, symbolically called my_first_patch. To pop the patch off:
And we're back to our pristine version.
Build again and rerun SunSpider again. You should now have two files like:
Compare them using the compare script:
cd SunSpider ./sunspider-compare-results --shell=../build_DBG.OBJ/js --suite=sunspider-0.9.1 FILE1-withoutPatch FILE2-withPatch
Get a real bug
At this point, you've seen nearly everything you need to do hack on SpiderMonkey. So it's time to get a real bug to work on. You can get a bug on Bugs Ahoy.
Fix the bug, updating the bug report with your progress, and asking questions as you go (either in the bug comments, or in #jsapi). When it's done, repeat all the steps above. Then it's time to get your patch into the tree.
Submit a patch
To get the patch from mercurial, use:
hg qdiff # if you're using queues or hg diff # if you're not
Add it to the bug as an attachment.
Get a review
Nothing gets into the tree without a review, so you'll need one. The SpiderMonkey hackers list is a good place to start: if your patch changes something listed as an area of expertise for someone there, that's a good person to ask for a review.
hg blame on the files you've changed, and check who has been changing related code recently. They're likely to be good candidates. An irc bot can automate this process for a single file if you enter "/msg mrgiggles who can review something.cpp?" into your irc client.
The review will consist of comments on your changes, suggesting or requesting alternative ways to do something and asking you to make changes where needed. They might also request additional changes, for example tests. Fix what they ask, resubmit the patch to bugzilla, and ask for another review. After you repeat this step a few times, they will mark the patch as "
r+" meaning it's now good to commit.
You can't commit to mozilla-central / mozilla-inbound until you have "level 3" access, so you'll need someone to do this for you. Try asking in #jsapi, or add the
checkin-needed keyword to the bug. After you have been contributing for a while, you can get level 3 access by applying for it.
After committing, a large series of tests will be run to make sure you didn't break anything. You will need to hang around to make sure you didn't break something. It is difficult to determine what failures are real and what are what we call "intermittent oranges" ("orange" because that is the color used on our continuous integration dashboard for test failures). If you do break something, a sheriff will probably let you know via IRC and will probably back your patch out. You can always ask for help determining what is going on. (Over time you'll get a feel for figuring out when breakage is real on your own.)
Overview of the JS engine
The JS engine is a swiftly moving target. The most detailed information is available at https://developer.mozilla.org/en/SpiderMonkey. Here are some particularly interesting, mostly up-to-date resources:
High level overviews
Medium level documentation
Frequently used coding recipes and mappings from JS idioms to SpiderMonkey code: https://developer.mozilla.org/En/SpiderMonkey/JSAPI_Cookbook
Function reference: https://developer.mozilla.org/en/SpiderMonkey/JSAPI_Reference
Collaboration and teamwork
Communication (in descending order of information content)
Nearly all communication is handled through Bugzilla. All bugs, feature requests, issues, enhancements, etc, are all referred to as bugs. No code may enter the Mozilla repository without being a patch attached to a bugzilla bug first. To follow all SpiderMonkey related bugs:
SpiderMonkey contributors generally hang out in the very active #jsapi IRC channel.
The js-internals mailing list is used for communicating with other SpiderMonkey hackers. (Internals as in discussing the internals of the SpiderMonkey engine. All are welcome on the list.).
Reading Planet Mozilla is the best way to keep up with the Mozilla project, which includes some SpiderMonkey related blogs:
The js-engine mailing list is generally used for communicating with people who embed SpiderMonkey, such as asking questions and announcing API changes. No actual development happens on the list.
Most active work on SpiderMonkey is done in the mozilla-central branch of the mozilla repository.
For many years, SpiderMonkey was written in C, and is gradually moving to C++. We still avoid many features such as run-time type information and exceptions, and generally avoid virtual functions for core data, but have come around to the glory of templates and namespaces relatively recently. Read the Coding Style. Do NOT read the portability guidelines, which I will not link to, since they are very out of date.
Everything goes through bugzilla. We find a bug or have an idea, then submit it to bugzilla, then file patches to solve it. When it is solved, the patch is reviewed by team members, and is then committed to the mozilla-inbound repository. A link to the commit is then automatically added as a comment to the bug. mozilla-inbound is periodically merged to mozilla-central by a Sheriff. This exempts us from watching treeherder to check for failures caused by our patches, and commits are automatically backed out if errors are found.
The following docs are for the Mozilla project, and differ ever so slightly from what you should do for SpiderMonkey. For example, you should find reviewers in #jsapi rather than #developers.
This .hgrc file contains a lot of the wisdom distilled through the wiki:
[extensions] mq = [ui] username = First Last <firstname.lastname@example.org> [alias] qfulldiff = diff --rev qparent:. [defaults] diff = -p -U 8 qdiff = -p -U 8 qnew = -U commit = -v [diff] git = true showfunc = true unified = 8 [paths] try = ssh://email@example.com@hg.mozilla.org/try/
Often, you may be a little wary of breaking things. Mozilla has a Try Server where you can send a patch, and it'll be built and tested on tons of machines, which report to TreeHerder. Although you don't have access to TryServer just yet, you can get someone else to push it there for you. Just attach a patch to your bug, and ask in a comment, or on #jsapi. To be able to push to try-server yourself, you will need "level 1" access, which you can request once you've contributed a patch or two.
Since most of our lives revolve around patches, and we use Mercurial, nearly everybody uses Mercurial queues.
Queues are based on the idea of patch management. Each queue consists of a series of patches, applied sequentially. The aim of the game is to split potential commits into simple, bite-sized chunks which are easy to review. This also makes it simple to experiment without polluting your existing work on a bug, to spin parts off into new bugs, and to rapidly apply and unapply patches (as demonstrated in the tutorial above).
Add the following snippet to your .hgrc file to enable MQ
[extensions] mq =
Example MQ workflow
Clone the repository to deal with a bug:
Initialize your queue:
hg qinit -c
Create a new patch, with some name:
hg qnew first_attempt
Work on the patch, try to fix the bug, test, compile, etc.
Refresh (save your work into the patch):
Repeat a few times.
Rename the patch (because your first name wasn't appliable):
hg qrename refactor_the_whatsit
Create a new patch to try a logically separate part of the same bug:
hg qnew rip_out_the_old_thing
Repeat the process a few times, until you have solved the problem. During this time, it can often be useful to go back and forth between patches:
hg qpop # unapply a patch hg qpush # reapply a patch hg qpop -a # unapply all patches hg qpush -a # reapply all patches
Combine all the patches into a single patch, which you submit to bugzilla:
hg qpush -a hg qdiff --rev qparent:. > my_wonderful_patch.patch
Commit the patches to your local patch repository, in case you make a mistake (You might do this periodically):
hg qcommit -m "Some message. Doesn't have to be good, this won't be committed to the repository, it's just for you"
Go back to the old patches and fiddle with them based on feedback:
hg qpop hg qrefresh etc
There are some more complex techniques that we won't go into in detail. You can enable only some patches using qguard and qselect, and you can reorder the patches by manually editing the .hg/patches/series file. But be careful!