Code Coverage Exercise on Firefox Browser

What is the progress so far!!

We have completely instrumented Mozilla Firefox , both C/C++ portions and JavaScript portions, twice and generated code coverage results.

Results would be discussed in a separate section.

Using automated test suites, Mochitest, XPCShell, JSTest, Chrome, Browser-Chrome, RefTest , we got approximately 60% code coverage for C/C++ portion and 55% coverage on the JavaScript portion.

We have generated a lot of trend diagrams to study the interaction between code coverage data for a given file and the number of bug fixes went into the file; code coverage Vs the number of times the file is modified and file churn Vs bug fix rate.

Results would be discussed below.

The plan is to observe the temporal changes in the above trends as we fix more bugs as well as develop new test cases.

This is what is planned for the immediate future

  • We are going to identify top 3 modules that needs further analysis
  • We will identify at least 6 files in each of those modules that have active churn but not good coverage
  • We will work with DEV and QA to come up with scenarios to improve coverage on those files

Results and Findings

Code Coverage report generated using lcov looks like this

For the full report to do your own study, please use the following URL

http://people.mozilla.org/~mnandigama/codecoverage_html/



This is important : We have done a component/module by component/module code coverage Vs Bugs fix analysis.

  • In a happy path case, all files should have 100% code coverage
  • In a normal path case, all files that have high rate of bugs fixed in them, should have HIGH code coverage since there are the most active files in terms of change.
  • Files that have no change since a long time can have either HIGH or LOW coverage in our first iteration since they are not undergoing any change so far.

In a bubble chart format , the same can be visualized as follows

http://people.mozilla.org/~mnandigama/codecov-video2/timechart_for_code_coverage.htm

The bigger the bubble, the higher it should be and all bubbles should ideally bubble up.

How to instrument the Mozilla Firefox Browser

Build config

This is what you do to instrument the Mozilla Firefox browser.

Create a .mozconfig-gold file in your home directory with the following statements


   export CFLAGS="-fprofile-arcs -ftest-coverage"
export CXXFLAGS="-fprofile-arcs -ftest-coverage"
export LDFLAGS="-lgcov -static-libgcc"
export EXTRA_DSO_LDFLAGS="-lgcov -static-libgcc"
mk_add_options MOZ_OBJDIR=@TOPSRCDIR@/../mozcentral-dbg
mk_add_options MOZ_MAKE_FLAGS="-j3"
mk_add_options AUTOCONF=autoconf2.13
ac_add_options --enable-application=browser
ac_add_options --enable-debug
ac_add_options --enable-tests
ac_add_options --enable-mochitest
ac_add_options --disable-optimize
ac_add_options --enable-chrome-format=flat

For more detailed build options and prerequisites, please visit the following URL

https://developer.mozilla.org/en/Configuring_Build_Options

Checkout the source code from the mozilla-central code repository.


     rm -rf /home/user/src
rm -rf /home/user/mozcentral-dbg
hg clone http://hg.mozilla.org/mozilla-central/ src
cd src
cp ../.mozconfig-gold .mozconfig


TIP: Save the above commands into a file named ‘startover’ and make it executable.

TIP:Change the ‘user’ part to your login account name.



The standard build command to build instrumented mozilla firefox is


     make -f client.mk build

Execute some Unit tests

Change Directory to ~/mozcentral-dbg/_tests/testing/mochitest and execute the following command.


 python runtests.py  to execute mochitest suite.
python runtests.py --chrome --browser-chrome to execute chrome tests.
python runtests.py --a11y to run accessibility tests


Please keep the focus on the Firefox browser , when running your tests, at all times or else your tests will fail.


In order to run jstests, please go to ~/src/js/tests and execute the following commands


   ./jsDriver.pl -e smdebug -L lc2 -L lc3 -L spidermonkey-n-1.9.1.tests -s ~/mozcentral-dbg/dist/bin/js
   ./jsDriver.pl -e smdebug -L lc2 -L lc3 -L spidermonkey-n-1.9.1.tests -s ~/mozcentral-dbg/dist/bin/js -o '-j'
Getopt-mixed is a required library by that script it needs to be added.


Gather the data

After you are done with the test runs, you will end up with a bunch of *.gcda and *.gcno files in your build directory.


Now, come back into your home directory and execute the following command


      lcov -c  -d mozcentral-dbg -o app.info


The lcov tool will go through the entire build directory and generates an app.info file in your home directory


TIP: If you don’t know what is an ‘lcov’ tool, please feel free to ‘google’ the term.


Then, execute the following command


       genhtml app.info

If by any chance the above command fails to complete, use the following option.

       genhtml --no-source app.info


The resulting index.html is the landing page for all your code coverage data for the C/C++ code in the Firefox browser.

http://people.mozilla.org/~mnandigama/codecoverage_html/

HINT: Look at the advanced tweaks section on how to modify, geninfo, genhtml scripts in order to generate full code coverage reports.


When we run the instrumented Firefox, there is now a "JSCoverage" menu in the "Tools" menu.


Tools -> JSCoverage -> View Coverage loads the jscoverage.html file


Tools -> JSCoverage -> Store Coverage saves the coverage data into a directory named "./jscoverage-report.

Scripts and Works around required

LCOV

You'll need lcov.

on os X the bin/install.sh needs to be modified on the last line there is a install -D remove the -D switch as it's not supported on OS X.

The following needs to be done :

sudo vi /usr/bin/geninfo
Press the following key sequence:
Esc -> : -> %s/die/print/g -> Enter
sudo vi /usr/bin/genhtml
Press the following key sequence:
Esc -> : -> %s/die/print/g -> Enter



Jscoverage

Code coverage for the Javascript part of the Firefox can be done in multiple ways. But using JSCoverage as the tool seemed to be the best way as JSCoverage is an open source tool.


I wish to fully acknowledge the great help Ed Kelly , who maintains the JSCoverage tool , provided in this instrumentation exercise. Ed was very generous in spending cycles to provide enhancements to the JSCoverage tool to meet Firefox instrumentation requirements.


So ... here is how to instrument the javascript portion of Firefox.


Do not download the binary of the JScoverage from http://siliconforks.com/jscoverage/

You should check out the source code from its public Subversion repository:

     svn co http://svn.siliconforks.com/jscoverage/trunk jscoverage

To compile, run bootstrap.sh (requires Automake and Autoconf):

   ./bootstrap.sh
Then run make as usual.
Finally execute the make install command so that jscoverage is
installed in the /usr/local/bin


Why not use the binary from the site and why to build our own jscoverage !!


The following enhancements are present in our own build with latest revision.

1. Instrumenting symbolic links in the case where the destination is a regular file.

For those who are curious, JSCoverage does not instrument javascript files that are symbolically referenced.

2. It copies file permissions (so that executable files stay executable).

3. It has the --mozilla option to copy the jscoverage.html file directly into Firefox's chrome.

The --mozilla option automatically sets the JavaScript version to 180,so it is not necessary to use both --mozilla and --js-version at the same time.

NOTE: --js-version=180 would allow instrumenting the symbolic links


So, starting from the top of the Firefox build (~/mozcentral-dbg )directory , here are the commands to instrument the whole Firefox directory and run the mochitests with the instrumented version:


       cd /dist
mv bin bin-original
jscoverage \
--mozilla \
--no-instrument=defaults \
--no-instrument=greprefs \
bin-original bin


HINT: If you find a foobar.js file failing in the javascript instrumentation exercise, please add it to the no-instrument list using the following syntax --no-instrument=/path/to/foobar.js


It seems that it is necessary to use --no-instrument on the entire defaults and prefs directories because the preferences .js files should not contain arbitrary JavaScript.

See http://www.xulplanet.com/tutorials/xulqa/q_prefs.html or https://bugzilla.mozilla.org/show_bug.cgi?id=440777 for details.

How to consolidate / merge results and generate graphs

TO BE ADDED 

Comments

Justin Dolske said…
Do you have results from JSCoverage available? I don't see any JS data in the links here.
Shawn Wilsher said…
It doesn't look to me like the results generated here are running [all|any] xpcshell unit tests. I'm seeing 0 test coverage for things that I can verify we have xpcshell test cases for.

Since most of the tests I write for the modules I work in are xpcshell, this analysis doesn't seem so useful...
Justin,
the JSCoverage results are stored with in the chrome directory and I have not yet found a way to display them outside of the actual Firefox browser on which the test are run. I will keep you posted on the progress.
Shawn, as we discussed, you now know the reasons for the low coverage on three files in storage now.Thanks
Anonymous said…
Thanks for this nice health info, it's so useful for me.

regards,
Max
Anonymous said…
topic. If possible, as you gain expertise, would you mind updating your blog with more information? It is extremely helpful and beneficial to your readers.

--
Kerja Keras Adalah Energi Kita

Popular Posts