j2me in your browser
- make it compile
- add softkey support
- add more resolutions
- create helper script that generates url (instead of manually reading META-INF/MANIFEST.MF for classes)
- Spidermonkey (js102 on arch, libmozjs-102-dev on debian)
- Typescript
- Python 2
- Java JDK 8, 11 or 17 (see notice for JDK 17 in Building section)
Arch :
#replace 102 with 91 or 78 if it doesnt work
#get python 2 from AUR
pacman -S js102 npm jdk11-openjdk
ln -s /usr/bin/js102 /usr/local/bin/js
npm install -g typescript
Debian :
#replace 102 with 91 or 78 if it doesnt work
apt install libmozjs-102-dev npm python2 openjdk-11-jdk
ln -s /usr/bin/js102 /usr/local/bin/js
npm install -g typescript
git clone https://github.com/calculatortamer/pluotsorbet
cd pluotsorbet
make
# Notice for JDK 17 users : use this instead
make JAVA_VER=8 TESTS=0
Launch web server
python2 tests/httpServer.py 2>/dev/null &
Get url to copy to browser (note : jar must be inside of the git directory)
./get_url.sh $PATH_TO_JAR
Or directly launch to default browser
./launch_jar.sh $PATH_TO_JAR
- Q, W, E ,R (or A,Z,E,R) = Game A, B, C, D
- Z and X (or W and X) = Left and Right softkeys
- Arrows = Up, Down, Left, Right
- Numpad = Keypad (1,2,3 and 7,8,9 are swapped)
- Numpad Star (*) = Star (*)
- Numpad Minus (-) = Pound (#)
(original readme below)
PluotSorbet is a J2ME-compatible virtual machine written in JavaScript[1]. The goal of PluotSorbet is to run MIDlets in web apps without a native plugin.
The current goals of PluotSorbet are:
- Run MIDlets in a way that emulates the reference implementation of phone ME Feature MR4 (b01)
- Keep PluotSorbet simple and small: Leverage the phoneME JDK/infrastructure and existing Java code as much as we can, and implement as little as possible in JavaScript
PluotSorbet launches MIDlets under two different circumstances.
- Install run only happens once per device. It spends extra time downloading, optimizing, and precompiling so that subsequent runs will be faster.
- Cold runs require starting up both foreground and background MIDlets, if needed
- Hot runs only require switching to the active app
Make sure you have wget and a JRE installed
You need to install the TypeScript compiler, the easiest way is via NPM: npm install -g typescript
.
Get the PluotSorbet repo if you don't have it already
git clone https://github.com/mozilla/pluotsorbet
Build using make:
cd pluotsorbet
make
index.html is a webapp that runs PluotSorbet. The URL parameters you pass to index.html control the specific behavior of PluotSorbet.
You can specify URL parameters to override the configuration. See the full list of parameters at config/urlparams.js.
main
- default iscom/sun/midp/main/MIDletSuiteLoader
midletClassName
- must be set to the main class to run. Only valid when defaultmain
parameter is used. Defaults toRunTestsMIDlet
autosize
- if set to1
, j2me app will fill the page.gamepad
- if set to1
, gamepad will be visible/available.
To run a MIDlet on desktop, you must first start an http server that will host index.html. You can then connect to the http server, passing URL parameters to index.html
python tests/httpServer.py &
http://localhost:8000/index.html?jad=ExampleApp.jad&jars=ExampleApp.jar&midletClassName=com.example.yourClassNameHere
Example - Asteroids
python tests/httpServer.py &
http://localhost:8000/index.html?midletClassName=asteroids.Game&jars=tests/tests.jar&gamepad=1
Some apps require access to APIs that aren't enabled by default on Desktop Firefox and there is no UI built in to Desktop Firefox to enable them. APIs matching this description include:
- mozTCPSocket
- mozContacts
- mozbrowser
- notifications
To enable this type of API for a MIDlet you're running, use Myk's API Enabler Addon
To run a MIDlet on a FirefoxOS device, update the launch_path
property in manifest.webapp. The midletClassName
URL parameter needs to point to an app.
Once you've updated manifest.webapp, connect to the device or emulator as described in the FirefoxOS Developer Phone Guide and select your PluotSorbet directory (the one containing manifest.webapp) when choosing the app to push to device.
Example - Asteroids
"launch_path": "/index.html?midletClassName=asteroids.Game&jars=tests/tests.jar&logConsole=web&autosize=1&gamepad=1"
You can run the test suite with make test
. The main driver for the test suite is tests/automation.js which uses the CasperJS testing framework and SlimerJS (a Gecko backend for CasperJS). This test suite runs on every push (continuous integration) thanks to Travis CI.
make test
downloads SlimerJS for you automatically, but you have to install CasperJS yourself. The easiest way to do that is via NPM: npm install -g casperjs
. On Mac, you may also be able to install it via Brew.
If you want to pass additional casperJS command line options, look at the "test" target in Makefile and place additional command line options before the automation.js filename.
gfx tests use image comparison; a reference image is provided with the test and the output of the test must match the reference image. The output is allowed to differ from the reference image by a number of pixels specified in automation.js.
The main set of unit tests that automation.js runs is the set covered by the RunTests class. The full list of RunTests tests available in the tests/Testlets.java generated file. RunTests runs a number of "Testlets" (Java classes that implement the Testlet
interface). Testlets that require to be executed in a MIDlet environment (classes that implement the MIDletTestlet
interface) are run by RunTestsMIDlet. The full list of these tests is available in the tests/MIDletTestlets.java generated file.
If the test you want to run is a class with a main method, specify a main
URL parameter to index.html, e.g.:
main=gnu/testlet/vm/SystemTest&jars=tests/tests.jar
If the test you want to run is a MIDlet, specify midletClassName
and jad
URL parameters to index.html (main
will default to the MIDletSuiteLoader), e.g.:
midletClassName=tests/alarm/MIDlet1&jad=tests/midlets/alarm/alarm.jad&jars=tests/tests.jar
If the test you want to run is a Testlet , specify an args
URL parameter to index.html. You can specify multiple Testlets separated by commas, and you can use either '.' or '/' in the class name, e.g.:
args=java.lang.TestSystem,javax.crypto.TestRC4,com/nokia/mid/ui/TestVirtualKeyboard
If the testlet uses sockets, you must start 4 servers (instead of just the http server):
python tests/httpServer.py &
python tests/echoServer.py &
cd tests && python httpsServer.py &
cd tests && python sslEchoServer.py &
Frequent causes of failure include:
- timeout: Travis machines are generally slower than dev machines and so tests that pass locally will fail in the continuous integration tests
- Number of differing pixels in a gfx/rendering test exceeds the threshold allowed in automation.js. This will often happen because slimerJS uses a different version of Firefox than the developer. This can also happen because the test renders text, whose font rendering can vary from machine to machine, perhaps even with the same font.
gfx/rendering tests will print a number next to the error message. That number is the number of differing pixels. If it is close to the threshold you can probably just increase the threshold in automation.js with no ill effect.
The test output will include base64 encoded images; copy this into your browser's URL bar as a data URL to see what the actual test output looked like.
When running make test
, verbose test output will be printed to your terminal. Check that for additional info on the failures that occurred.
See logConsole
and logLevel
URL params in libs/console.js
-
Download the SpiderMonkey shell
-
Execute the jsshell.js file as follows:
js jsshell.js package.ClassName
In general, stick with whatever style exists in the file you are modifying.
If you're creating a new file, use 4-space indents for Java and 2-space indents of JS.
Use JavaDoc to document public APIs in Java.
Modeline for Java files:
/* vim: set filetype=java shiftwidth=4 tabstop=4 autoindent cindent expandtab : */
Modelines for JS files:
/* -*- Mode: Java; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set shiftwidth=2 tabstop=2 autoindent cindent expandtab: */
One way to profile PluotSorbet is to use the JS profiler available in Firefox Dev Tools. This will tell us how well the JVM is working and how well natives work. This type of profiling will not measure time that is taken waiting for async callbacks to be called (for example, when using the native JS filesystem API).
The PluotSorbet VM has several profiling tools. The simplest feature is to use counters. runtime.ts
defines several: runtimeCounter
, nativeCounter
, etc ... these are only available in debug builds.
To use them, just add calls to runtimeCounter.count(name, count = 1)
. To view accumulated counts, allow the application to run for some time and then click the Dump Counters
button. If you want, reset the counter count any time by clicking Clear Counters
.
-
Counting events:
function readBytes(fileName, length) { runtimeCounter && runtimeCounter.count("readBytes"); }
-
Counting bucketed events:
function readBytes(fileName, length) { runtimeCounter && runtimeCounter.count("readBytes " + fileName); }
-
Counting events with larger counts:
function readBytes(fileName, length) { runtimeCounter && runtimeCounter.count("readBytes", length); }
-
Counting events with caller context: This is useful to understand which call sites are the most common.
function readBytes(fileName, length) { runtimeCounter && runtimeCounter.count("readBytes " + arguments.callee.caller.name); }
The second, more heavy weight profiling tool is Shumway's timeline profiler. The profiler records enter
/ leave
events in a large circular buffer that can be later displayed visually as a flame chart or saved in a text format. To use it, build PluotSorbet with PROFILE=[1|2|4]
. Then wrap code regions that you're interested in measuring with calls to timeline.enter
/ timeline.leave
.
Java methods are automatically wrapped with calls to methodTimeline.enter
/ methodTimeline.leave
. The resulting timeline is a very detailed trace of the application's execution. Note that this instrumentation has some overhead, and timing information of very short lived events may not be accurate and can lead to the entire application slowing down.
Similar to the way counters work, you can get creative with the timeline profiler. The API looks something like this:
timeline.enter(name: string, details?: Object);
timeline.leave(name?: string, details?: Object);
You must pair the calls to enter
and leave
but you don't necessarily need to specify arguments for name
and details
.
The name
argument can be any string and it specifies a event type. The timeline view will draw different types of events in different colors. It will also give you some statistics about the number of times a certain event type was seen, how long it took, etc..
The details
argument is an object whose properties are shown when you hover over a timeline segment in the profiler view. You can specify this object when you call timeline.enter
or when you call timeline.leave
. Usually, you have more information when you call leave
so that's a more convenient place to put it.
The way in which you come up with event names can produce different results. In the profilingWrapper
function, the key
is used to specify the event type.
You can also create your own timelines. At the moment there are 3:
timeline
: VM Events like loading class files, linking, etc.methodTimeline
: Method execution.threadTimeline
: Thread scheduling.
You may have to change the CSS height style of the profileContainer
if you don't see all timelines.
Top band is an overview of all the timelines. Second band is the timeline
, third is the threadTimeline
and finally the fourth is the methodTimeline
. Use your mouse wheel to zoom in and out, pan and hover.
The tooltip displays:
total
: ms spent in this event including all the child events.self
:total
-total
sum of all child events.count
: number of events seen with this name.all total
andall self
: cumulative total and self times for all events with this name.- the remaining fields show the custom data specified in the
details
object.
If you build with PROFILE=2
, then the timeline will be saved to a text file instead of being shown in the flame chart. On desktop, you will be prompted to save the file. On the phone, the file will automatically be saved to /sdcard/downloads/profile.txt
, which you can later pull with adb pull
. Note that no timeline events under 0.1 ms are written to the file output. You can change this in main.js
if you'd like.
PROFILE=4
works like PROFILE=2
, but allows you to profile a custom "range" of the execution by adding calls to org.mozilla.internal.Sys::startProfile and org.mozilla.internal.Sys::stopProfile.
PROFILE=1
and PROFILE=2
automatically profile (most of) cold startup, from JVM.startIsolate0 to DisplayDevice.gainedForeground0.
The startup benchmark measures from when the benchmark.js file loads to the call of DisplayDevice.gainedForeground0
. It also measures memory usage after startup. Included in a benchmark build are helpers to build baseline scores so that subsequent runs of the benchmark can be compared. A t-test is used in the comparison to see if the changes were significant.
To use:
It is recommended that a dedicated Firefox profile is used with the about:config preference of security.turn_off_all_security_so_that_viruses_can_take_over_this_computer
set to true so garbage collection and cycle collection can be run in between test rounds. To do this on a Firefox OS device, see B2G/QA/Tips And Tricks.
- Check out the version you want to be the baseline (usually mozilla/master).
- Build a benchmark build with
RELEASE=1 BENCHMARK=1 make
.RELEASE=1
is not required, but it is recommended to avoid debug code from changing execution behavior. - Open the midlet you want to test with
&logLevel=log
appended to the url and clickBuild Benchmark Baseline
. - When finished, the message
FINISHED BUILDING BASELINE
will show up in the log. - Apply/check out your changes to the code.
- Rebuild
RELEASE=1 BENCHMARK=1 make
. - Refresh the midlet.
- Click
Run Startup Benchmark
. - Once done, the benchmark will dump results to the log. If it says "BETTER" or "WORSE" the t-test has determined the results were significant. If it says "SAME" the changes were likely not enough to be differentiated from the noise of the test.
midp/fs.js contains native implementations of various midp filesystem APIs.
Those implementations call out to lib/fs.js which is a JS implementation of a filesystem.
Java APIs are sync, so our implementation stores files in memory and makes them available mostly synchronously.
native
keyword tells Java that the function is implemented in native code
e.g.:
public static native long currentTimeMillis();
Java compiler will do nothing to ensure that implementation actually exists. At runtime, implementation better be available or you'll get a runtime exception.
We use Native
object in JS to handle creation and registration of native
functions. See native.js
Native["name/of/function.(parameterTypes)returnType"] = jsFuncToCall;
e.g.:
Native["java/lang/System.arraycopy.(Ljava/lang/Object;ILjava/lang/Object;II)V" = function(src, srcOffset, dst, dstOffset, length) {...};
If raising a Java Exception
, throw new instance of Java Exception
class as defined in vm/runtime.ts, e.g.:
throw $.newNullPointerException("Cannot copy to/from a null array.");
If you need implement a native method with async JS calls, the following steps are required:
- Add the method to the
yieldMap
in jit/analyze.ts - Use
asyncImpl
in native.js to return the asnyc value with aPromise
.
e.g:
Native["java/lang/Thread.sleep.(J)V"] = function(delay) {
asyncImpl("V", new Promise(function(resolve, reject) {
window.setTimeout(resolve, delay.toNumber());
}));
};
The asyncImpl
call is optional if part of the code doesn't make async calls. The method can sometimes return a value synchronously, and the VM will handle it properly. However, if a native ever calls asyncImpl, even if it doesn't always do so, then you need to add the method to yieldMap
.
e.g:
Native["java/lang/Thread.newSleep.(J)Z"] = function(delay) {
if (delay < 0) {
// Return false synchronously. Note: we use 1 and 0 in JavaScript to
// represent true and false in Java.
return 0;
}
// Return true asynchronously with `asyncImpl`.
asyncImpl("Z", new Promise(function(resolve, reject) {
window.setTimeout(resolve.bind(null, 1), delay.toNumber());
}));
};
Remember:
- Return types are automatically converted to Java types, but parameters are not automatically converted from Java types to JS types
this
will be available in any context thatthis
would be available to the Java method. i.e.this
will benull
forstatic
methods.$
is current runtime and$.ctx
current Context- Parameter types are specified in JNI
To override a Java function with a JavaScript function, simply define a Native as
described earlier. Any Java functions can be overridden, not only Java functions
with the native
keyword.
Overriding Java functions only works in debug mode (RELEASE=0).
make app
packages PluotSorbet into an Open Web App in output directory.
It's possible to simply package the entire contents of your working directory,
but these tools will produce a better app.
make aot
compiles some Java code into JavaScript with an ahead-of-time (AOT) compiler.
To use it, first install a recent version of the JavaScript shell.
make closure
compiles some JavaScript code with the Closure compiler.
[1] JavaScript is a trademark or registered trademark of Sun Microsystems, Inc. in the U.S. and other countries, used under license.