The harness (test.py) operates as follows. First it locates executable browsers
(or symlinks or scripts) named "[browser][version]", e.g. "firefox4".
It then launches the located browsers and asks them to load the file
test_slave.html. At the same time, test.py sets up an HTTP server on
localhost:8080 (there's a race condition here currently ;). After
test_slave loads in the browser(s), it fetches the task manifest
(test_manifest.json). The entries in the manifest specify which PDF
to load and how many times to cycle through page rendering. This will
probably evolve over time. test_slave then performs the requested
tasks and POSTs the results back to test.py, which saves them. When
all the results of for a task are in, test.py checks them.
There are three types of tests currently. "==" tests compare the
rendering of a PDF against a master copy. This is not yet implemented
because setting up a master copy is complicated. "fbf" tests render
all a PDF's pages, then go back to page 1 and render all pages a
second time. The renderings from the first round must match the ones
from the second round. "load" tests just check that a PDF's pages
load without errors.
Currently the test harness will only launch a "firefox4" target. This
can be a bash script in your pdf.js checkout, pdf.js/firefox4,
something like the following
#!/bin/bash
dist="/path/to/firefox4/installation"
profile=`mktemp -dt 'pdf.js-test-ff-profile-XXXXXXXXXX'`
$dist/firefox -no-remote -profile $profile $*
rm -rf $profile
(Yes, this script doesn't clean up properly on early termination.)
It's possible to run the tests in a normal browsing session, but that
might be annoying. With that set up, run the harness like so
python test.py
If all goes well, you'll see all "TEST-PASS" messages printed to
stdout. If something goes wrong, you'll see "TEST-UNEXPECTED-FAIL"
printed to stdout.
