The default is 'true'. When set 'false', file: and qrc: URLs are
treated as invalid (unknown scheme) rather than opening local files,
as requested in issue #12752.
In order to test this, I added a mechanism to test/run-tests.py
allowing individual tests to be annotated with command-line
options to pass to phantomjs or the script.
Unfortunately, due to Qt and WebKit limitation, this is technically
clearing all the cache used by every web page instance. It is the
best we can have right now.
https://github.com/ariya/phantomjs/issues/10357
This is available via the new property called `release`. A very simple
example to demonstrate it:
var system = require('system');
console.log('Kernel release', system.os.release);
which will print (on OS X 10.7.5 Lion):
Kernel release 11.4.2
https://github.com/ariya/phantomjs/issues/12587
This means you don't have to wait 30 seconds in between invocations of
run-tests.py.
Also, if the test server fails to bind its port, print the actual
OS-level error message rather than guessing what the problem is.
issue #12439
* correctly fake a package to hold all the response-hook modules
* use StringIO correctly in the response hooks
* prevent .py(c) files in test/www/ from being accessed directly
* prevent test/www/__init__ from being treated as a response hook
* add a test case that makes sure the existing hooks _can_ return 200 OK
(issue #12439; buggy commit 4d60e94)
If there are any positional arguments to test/run-tests.py, they are matched
(verbatim, not as regular expressions) against each test's name. Only those
tests whose names contain at least one of the positional arguments will be
run. Thus, for instance, you can run just the webpage tests with
./test/run-tests.py module/webpage
or one specific test with
./test/run-tests.py exact/name/of/test.js
or the basics and standards tests with
./test/run-tests.py basics standards
If no tests at all match, run-tests.py will print "ALL TESTS SKIPPED" and
exit unsuccessfully.
Ongoing work on issue #12439.
Rather than add all the special URLs we'll ever want to run-tests.py,
the embedded HTTP server now supports scripts in test/www/. If you
try to load http://localhost:9180/path and test/www/path doesn't exist
but test/www/path.py does, then test/www/path.py is loaded as a module.
That module must export one function, handle_request(), which is called
to produce the response.
handle_request() has the same semantics as
SimpleHTTPRequestHandler.send_head(). That is, it takes one argument,
the SimpleHTTPRequestHandler object, conventionally named 'req'.
It should call the send_response(), send_header(), and end_headers()
methods of that object as appropriate. And it should return a readable
filelike whose contents are taken as the body of the response.
If either module import or handle_request() throws any exception, the
HTTP response will be a 500 Internal Server Error with body provided
by cgitb.
Ongoing work on issue #12439.
We already have 'base_path' as a global variable, so use that (or rather,
a new derivative of that) instead of inspect.getfile(inspect.currentframe()).
Rather than splitting the path and then iterating over every component
trying to figure out if it's "special", take advantage of the fact that
posixpath.normpath() will reliably produce a path with all ../ and bare-/
components at the very beginning. Preserve the distinction between paths
with and without a trailing /, as SimpleHTTPRequestHandler expects.
So that paths containing unusual (e.g. non-ASCII, non-UTF-8) segments can
safely be tested, unquote and then requote the entire path. This means
that the *bytes* corresponding to the ASCII characters
0123456789 ABCDEFGHIJKLMNOPQRSTUVWXYZ abcdefghijklmnopqrstuvwxyz -_./
are guaranteed to be unquoted, and all other *bytes* are guaranteed to be
%-quoted. (Note especially that *whether or not* the tests are being run
on Windows, \ in the original URL will be represented as %5C, and will
*not* be treated as a pathname separator.)
So that the tests behave consistently whether or not the OS has a case-
sensitive filesystem, after the above transformation, lowercase the entire
path.
Ongoing work on issue #12439.
This means that there is no longer a need to add every individual test
to the list; only new *directories* have to be mentioned. New tests
can be added just by dropping .js files into appropriate directories.
Tests which are not meant to be run by default now live in tests/manual/.
Currently there is one such test, tests/manual/standards/ecma-test262.js,
which downloads and runs the ECMAScript conformance suite.
Ongoing work on issue #12439.
The HTTP server launched by the test runner has a new endpoint called
/echo. This simply echoes back some information (current on the made
request, in the future it will be expanded to further information)
formatted as JSON. This echo system facilitates easy testing for request
header manipulation feature.
https://github.com/ariya/phantomjs/issues/12439
The server runs on port 9180 and serves the content from test/www
directory. Any tests can use this base URL to load test content and
to exercise certain PhantomJS API.
https://github.com/ariya/phantomjs/issues/12439
The use of phantom.scriptName has been deprecated since System#args was
introduce (version 1.5, March 2012). Any old scripts need to migrate to
System#args.
https://github.com/ariya/phantomjs/issues/12529
The use of phantom.args has been deprecated (see commit 545a3f7)
since a long time ago (version 1.5, March 2012). Any old scripts
need to switch to use System#args instead.
https://github.com/ariya/phantomjs/issues/12527
New option --ssl-ciphers takes a colon-separated list of OpenSSL
cipher names and sets the client cipher list to exactly that list.
The default for this option is arranged to match Chromium 35, which
has had its cipher selection optimized for the modern Web
(see https://briansmith.org/browser-ciphersuites-01.html for
rationales). (Newer versions are the same except that they also add
ChaCha20-based ciphersuites, which OpenSSL 1.0.1 hasn't yet picked up.)
* --ssl-protocol= option now accepts 'tlsv1.2', 'tlsv1.1', 'tlsv1.0'
and 'default' as well as the existing 'tlsv1', 'sslv3', and 'any'.
* The default is now none of the above, but rather QSsl::SecureProtocols,
which means "whatever subset of ANY is still considered secure and also
supported by the OpenSSL library in use". (As of this writing, Qt's idea
of "still considered secure" includes everything from SSLv3 on up, which
is technically wrong -- SSLv3 has known breaks -- but we can live with.
Qt currently doesn't have a way to select "TLSv1.0 and up".)
This mostly matters when system sqlite is selected; that has the side
effect of enabling pkg-config which in turn causes QtWebkit to detect
the presence of system GStreamer and/or QtMultimedia and turn video on.
This way, the build is consistent across the board.
The affected libraries are fontconfig, freetype, harfbuzz, libpng, libjpeg,
sqlite, pcre, and zlib.
* On Darwin, by default, fontconfig is disabled and all of the other
libraries are bundled.
* On non-Darwin, by default, fontconfig is enabled and system-provided
freetype, libpng, libjpeg, and zlib are used. Harfbuzz, pcre, and sqlite
continue to be bundled.
* Individual libraries from the above list may now be enabled or disabled
on the build.sh command line with e.g. --freetype=system or
--freetype=bundled.
* The existing --system-qtdeps, --system-qt, --system-qtwebkit options
have been renamed --qtdeps=system, --qt=system, --qtwebkit=system
respectively. New option --qtdeps=bundled forces all available
bundled libraries to be used.
Also:
* Slight adjustments to organization of preconfig.sh.
* Can now disable PulseAudio and ALSA.
* Directly link OpenSSL instead of loading it at runtime.
* Clarify the pkg-config situation (it must be enabled if and only if
sqlite comes from the system).
Zack Weinberg