Notes ===== ---IMPORTANT CHARACTER SET NOTE--- It is critical that all development work in Java2Script be done in UTF-8. This means: - making sure your Eclipse project is set up for UTF-8 (not the Eclipse default?) - making sure your server can serve up UTF-8 by default for any browser-loaded files - making sure you don't edit a Java2Script class file or one of the site .js files using a non-UTF-8 editor. It may replace non-Latin characters with "?" or garbage. - making sure that your web pages are delivered with proper headings indicating HTML5 and UTF-8 Note that the DOCTYPE tag is critical for some browsers to switch into HTML5 mode. (MSIE?) In particular, the Mandarin character 秘 (mi; "secret") is used extensively throughout the SwingJS class files to distinguish j2s-specific fields and methods that must not ever be shadowed or overridden by subclasses. For example, we see in public static JSThread 秘thisThread; ---------------------------------- updated 12/6/2020 -- note about restrictions on long, including BitSet and Scanner updated 3/21/2020 -- adds note about HashMap, Hashtable, and HashSet iterator ordering updated 3/20/2020 -- adds note about interning, new String("xxx"), and "xxx" updated 2/26/2020 -- adds Graphics.setClip issue updated 12/22/19 -- additional issues updated 11/03/19 -- adds information about File.exists() and points to src/javajs/async updated 10/26/19 -- adds information about File.createTempFile() updated 8/16/19 -- minor typos and added summary paragraph updated 7/19/19 -- clarification that AWT and Swing classes are supported directly updated 5/13/19 -- Mandarin U+79D8 reserved character; Missing Math methods; int and long updated 5/10/19 -- adds a section on static issues in multi-(duplicate)-applet pages updated 1/4/19 -- nio updated 9/15/18 -- adds integer 1/0 == Infinity updated 7/24/18 -- most classes replaced with updated 6/5/17 -- reserved package name "window" updated 3/11/17 -- myClass.getField updated 3/7/17 -- overloading of JSplitPane.setDividerLocation updated 3/2/17 -- more indication of classes not implemented (KeyListener) ============================================================================= SwingJS and OpenJDK 8+ ============================================================================= SwingJS implements a wide range of the Java language in JavaScript. The base version for this implementation is OpenJDK8. some classes are implemented using older source code, and there are some missing methods. For the most part, this is no real problem. You can add or modify any java class just be adding it as source in your project. Or (preferably) you can contact me, and I can get it into the distribution. Or (even more preferably) you can do that via a patch submission. ================= DESIGN PHILOSOPHY ================= The java2script/SwingJS design goal is to recreate a recognizable, easily debuggable equivalent in JavaScript for as much of Java as practical. This means, for example, that one can call in JavaScript new java.util.Hashtable() and for all practical purposes it will appear that Java is running. Method and Field Disambiguation ------------------------------- SwingJS has no problem with the overloading of methods, for example: public void print(int b); public void print(float b); JavaScript does not allow overloading of methods, and the common practice in Java of naming a field the same as a method -- isAllowed and isAllowed() -- is not possible in JavaScript. As a result, SwingJS implements "fully-qualified" method names using "$" parameter type separation. Thus, these methods in SwingJS will be referred to as print$I and print$F. The rules for this encoding are relatively simple: 1. The seven primitive types in Java are encoded $I (int), $L (long), $F (float), $D (double), $B (byte) $Z (boolean), and $H (short). 2. String and Object are encoded as $S and $O, respectively. 3. "java_lang_" is dropped for all other classes in the java.lang package (as in Java). For example: $StringBuffer, not $java_lang_StringBuffer 4. All other classes are encoded as "$" + Class.getName().replace(".","_") For example, in Java we see: public void equals(Object o) {...} Whereas in SwingJS we have: Clazz.newMeth(C$, 'equals$O', function (o) {...} And this.getContentPane().add(bar, "North"); becomes this.getContentPane$().add$java_awt_Component$O(bar, "North"); 5. Arrays are indicated with appended "A" for each level. So setDataVector(Object[][] dataVector, Object[] columnIdentifiers) becomes setDataVector$OAA$OA(dataVector, columnIdentifiers) (It is recognized that this design does introduce a bit of ambiguity, in that in principal there could be user class named XA and X in the same package, and methods a(X[]) and a(XA) in the same class that cannot be distinguished. The benefit of this simple system, however, triumphed over the unlikelyhood of that scenario.) The transpiler could be set to flag this possibility. 6. Constructors are prepended with "c$". So public JLabel(String text) {...} becomes: Clazz.newMeth(C$, 'c$$S', function (text) {...}); Field disambiguation involves prepending. In Java, a class and its subclass can both have the same field name, such as boolean visible; When this happens, it is called "shadowing", and though not recommended, Java allows it. The Java2Script transpiler will prepend such shadowing fields with "$" so that the subclass instance has both "visible" (for use in its methods inherited from its superclass) and "$visible" (for its own methods). Thus, we might see in Java: this.visible = super.visible; while in SwingJS we will see: this.$visible=this.visible; since JavaScript does not have the "super" keyword. Parameterless methods such as toString() are appended with "$" to become toString$(). The one exception to this rule is private methods, which are saved in (truly) private array in the class (and are not accessible by reflection). Private parameterless methods retain their simple Java name, since they cannot conflict with field names. This renaming of methods has a few consequences, which are discussed more fully below. See particularly the section on "qualified field and method names", where it is described how you can use packages or classes or interfaces with ".api.js" in them to represent JavaScript objects for which all method names are to be left unqualified. Note that it is not possible to cherry-pick methods to be unqualified; only full packages, classes or interfaces can hold this status. The swingjs.api.js package in particular contains a number of useful interfaces that you can import into your project for JavaScript-specific capabilities. Applet vs. Application ---------------------- One of the very cool aspects of SwingJS is that it doesn't particularly matter if a browser-based Java app is an "applet" or an "application". We don't need JNLP (Java Network Launch Protocol) because now we can just start up any Java application in a browser just as easily as any applet. The associative array that passes information to the SwingJS applet (information that formerly might have been part of the APPLET tag, such as width, height, and codebase, always referred to in our writing as "the Info array") allows the option to specify the JApplet/Applet "code" class or the application "main" class. Either one will run just fine. Performance ----------- Obviously, there are limitations. One is performance, but we have seen reproducible performance at 1/6 - 1/3 the speed of Java. Achieving this performance may require some refactoring of the Java to make it more efficient in both Java and JavaScript. "for" loops need to be more carefully crafted; use of "new" and "instanceof" need to be minimized in critical areas. Note that method overloading -- that is, the same method name with different parameters, such as read(int) and read(byte) -- is no longer any problem. Threads ------- Although there is only a single thread in JavaScript, meaning Thread.wait(), Thread.sleep(int) and Thread.notify() cannot be reproduced, we have found that this is not a serious limitation. For example, javax.swing.Timer() works perfectly in JavaScript. All it means is that threads that use sleep(int) or notify() must be refactored to allow Timer-like callbacks. That is, they must allow full exit and re-entry of, not the typical while/sleep motif. The key is to create a state-based run() that can be exited and re-entered in JavaScript. Static fields ------------- Final static primitive "constant" fields (String, boolean, int, etc.) such as static final int TEST = 3; static final String MY_STRING = "my " + "string"; are converted to their primitive form automatically by the Eclipse Java compiler and do not appear in the JavaScript by their names. Other static fields are properties of their class and can be used as expected. Note, however, that SwingJS runs all "Java" code on a page in a common "jvm" (like older versions of Java). So, like the older Java schema, the JavaScript equivalents of both applets and applications will share all of their static fields and methods. This includes java.lang.System. Basically, SwingJS implementations of Java run in a browser page-based sandbox instead of an applet-specific one. In general, this is no problem. But if we are to implement pages with multiple applets present, we must be sure to only have static references that are "final" or specifically meant to be shared in a JavaScript environment only (since they will not be shared in Java). A simple solution, if static non-constant references are needed, is to attach the field to Thread.currentThread.threadGroup(), which is an applet-specific reference. Be sure, if you do this, that you use explicit setters and getters: For example, private static String myvar; ... public void setMyVar(String x) { ThreadGroup g = Thread.currentThread().threadGroup(); /** * @j2sNative g._myvar = x; * */ { myvar = x; } } public String getMyVar() { ThreadGroup g = Thread.currentThread().threadGroup(); /** * @j2sNative return g._myvar || null; * */ { return myvar; } } in Java will get and set x the same in JavaScript and in Java. A convenient way to do this in general is to supply a singleton class with explicitly private-only constructors and then refer to it in Java and in JavaScript instead of using static field, referring to myclass.getIntance().xxx instead of in Java (and JavaScript). This was done extensively in the Jalview project. See jalview.bin.Instance. Helper Packages -- swingjs/ and javajs/ --------------------------------------- The SwingJS library is the swingjs/ package. There are interfaces that may be of assistance in swingjs/api, but other than that, it is not recommended that developers access classes in this package. The "public" nature of their methods is really an internal necessity. In addition to swingjs/, though, there are several useful classes in the javajs/ package that could be very useful. This package is a stand-alone package that can be cloned in any Java project that also would be great to have in any JavaScript project -- SwingJS-related or not. Functionality ranges from reading and writing various file formats, including PDF, BMP, PNG, GIF, JPG, JSON, ZIP, and CompoundDocument formats. A variety of highly efficient three- and four-dimensional point, vector, matrix, and quaternion classes are included, as they were developed for JSmol and inherited from that project. Of particular interest should be javajs/async/, which includes javajs.async.Async javajs.async.AsyncColorChooser javajs.async.AsyncDialog javajs.async.AsyncFileChooser See javajs.async.Async JavaDoc comments for a full description of these useful classes. Modal Dialogs ------------- Although true modal dialogs are not possible with only one thread, a functional equivalent -- asynchronous modal dialogs -- is relatively easy to set up. All the JOptionPane dialogs will return PropertyChangeEvents to signal that they have been disposed of and containing the results. See below and classes in the javajs.async package. Native calls ------------ Native calls in Java are calls to operating system methods that are not in Java. JavaScript has no access to these, of course, and they must all be replaced by JavaScript equivalents. Fortunately, they are not common, and those that are present in Java (for example, in calculating checksums in ZIP file creation) are at a low enough level that most developers do not utilize them or do not even have access to them. All native calls in Java classes have been replaced by Java equivalents. Swing GUI Peers and UIClasses ----------------------------- One of the biggest adaptations introduced in SwingJS is in the area of the graphical user interface. The issue here is complex but workable. In Java there are two background concepts -- the Component "peer" (one per "heavy-weight" component, such as a Frame) and the component "uiClass" (one per component, such as JButton or JTextField). Peers are native objects of the operating system. These are the virtual buttons and text areas that the user is interacting with at a very base level. Their events are being passed on to Java or the browser by the operating system. UI classes provide a consistent "look and feel" for these native objects, rendering them onto the native window canvas and handling all user-generated events. They paint the borders, the backgrounds, the highlights, of every control you see in Java. There is one-to-one correspondence of Swing classes and UI classes. Setting the Look and Feel for a project amounts to selecting the directory from which to draw these UI classes. The UI classes can be found in the javax.swing.plaf ("platform look and feel") package. Early on in the development of SwingJS, we decided not to fully reproduce the painfully detailed bit-by-bit painting of controls as is done in Java. Instead, we felt it was wiser to utilize the standard HTML5 UI capabilities as much as possible, using DIV, and INPUT especially, with extensive use of CSS and sometimes jQuery (menus, and sliders, for example). Thus, we have created a new set of UIs -- the "HTML5 Look and Feel". These classes can be found in swingjs.plaf. Besides being more adaptable, this approach allows far more versatility to SwingJS developers, allowing them to modify the GUI to suit their needs if desired. In SwingJS, since we have no access to native peers except through the browser DOM, it seemed logical to merge the peer and UI idea. So instead of having one peer per heavy-weight control and one UI class instance for each control type, we just have one UI class instance per control, and that UI class instance is what is being referred to when a "peer" is notified. In some ways this is a throw back to when all of Swing's components were subclasses of specific AWT components such as Button and List. These "heavy-weight components" all had their own individual native peers and thus automatically took on the look and feel provided by the OS. Later Swing versions implemented full look and feel for all peers, leaving only JDialog, JFrame, and a few other classes to have native peers. But in SwingJS we have again a 1:1 map of component and UI class/peer instance. The origin of most issues (read "bugs") in relation to the GUI will probably be found in the swingjs.plaf code. Swing-only Components -- no longer an issue ------------------------------------------- Swing was introduced into Java well after the Java Abstract Window Toolkit (AWT) was well established. As such, its designers chose to allow AWT controls such as Button and List to be used alongside their Swing counterparts JButton and JList. Reading the code, it is clear that this design choice posed a huge headache for Swing class developers. For SwingJS, we decided from the beginning NOT to allow this mixed-mode programming and instead to require that all components be Swing components. However, this is no longer an issue. All AWT components in SwingJS are now subclasses of javax.swing.JComponent. So far, we have found no problem with this. The a2s Adapter Package ----------------------- Originally, we thought that we would restrict ourselves to JApplets only. That is, only Swing-based applets. But as we worked, we discovered that there are a lot of great applets out there that are pre-Swing pure-AWT java.applet.Applet applets. Our problem was that we also wanted it to be possible to quickly adapt these applets to JavaScript as well. The solution turned out to be simple: Write a package (a2s) that recreates the interface for non-Swing components as subclasses of Swing components. Thus, a2s.Button subclasses javax.swing.JButton but also accepts all of the methods of java.awt.Button. This works amazingly well, with a few special adaptations to the core javax.swing to be "AWT-aware." All AWT components now subclass a2s components, which in turn subclass JComponents. So no changes in code are necessary. We have successfully transpiled over 500 applets using this strategy. (Kind of surprising, actually, that the original Java developers did not see that option. But we have a hindsight advantage here.) Working with Files ================== Simple String file names are not optimal for passing information about read files within SwingJS applications. All work with files should either use Path or File objects exclusively. These objects, after a file is read or checked for existence, will already contain the file byte[] data. Doing something like this: File f = File("./test.dat"); boolean isOK = f.exists(); will load f with its byte[] data, if the file exists. But if after that, we use: File f2 = new File(f.getAbsolutePath()); f2 will not contain that data. Such copying should be done as: File f2 = new File(f); in which case, the byte[] data will be transferred. SwingJS uses the following criteria to determine if File.exists() returns true: (1) if this File object has been used directly to read data, or (2) if reading data using this File object is successful. Note that you cannot check to see if a file exists before input or if it was actually written or if it already exists prior to writing in SwingJS. Thus, you should check each use of file.exists() carefully, and if necessary, provide a J2sNative block that gives an appropriate "OK" message, for example: (/** @j2sNative 1 ? false : */ outputfile.exits()) or (/** @j2sNative 1 ? true : */ inputfile.exits()) Temporary files can be created in SwingJS. SwingJS will maintain a pseudo-filesystem for files created with File.createTempFile(). This is useful in that closure of writing to a temporary file does not generate a pseudo-download to the user's machine. UNIMPLEMENTED CLASSES BY DESIGN =============================== The SwingJS implementation of the following classes are present in a way that gracefully bypasses their functionality: accessibility security serialization TODO LIST FOR UNIMPLEMENTED CLASSES =================================== JEditorPane (minimal implementation) - DONE 12/2018; some issues still JSplitPane - DONE 8/2018 JTabbedPane - DONE 10/2018 JTree - done 12/2019 MINOR ISSUES--required some rewriting/refactoring by Bob and Udo ================================================================ Thread.currentThread() == dispatchThread MINOR ISSUES--requiring some rewriting/refactoring outside of SwingJS ===================================================================== See below for a full discussion. Restrictions on long Restriction on BitSet and Scanner HashMap, Hashtable, and HashSet iterator ordering interning, new String("xxx") vs "xxx" Names with "$" and "_" positive integers do not add to give negative numbers ArrayIndexOutOfBounds java.awt.Color native methods javax.swing.JFileDialog key focus LookAndFeel and UI Classes System.exit(0) does not stop all processes list cell renderers must be JComponents myClass.getField not implemented "window" and other reserved JavaScript names reserved field and method names qualified field and method names missing Math methods Component.getGraphics(), Graphics.dispose() Graphics.setClip() MAJOR ISSUES--for Bob and Udo within SwingJS ============================================ fonts OS-dependent classes AWT component peers some aspects of reflection MAJOR ISSUES--to be resolved by implementers ============================================ fonts threads modal dialogs image loading BigDecimal not fully implemented no format internationalization no winding rules text-related field implementation Formatter/Regex limitations integer 1/0 == Infinity ======================================================================== DISCUSS ======= Table row/col sorter needs checking after removal of java.text.Collator references I had to move all of SunHints class to RenderingHints, or the two classes could not be loaded. Shouldn't be a problem, I think. The sun classes are not accessible to developers in Java anyway, since they are generally package private. ========================================================================== ////////////////////////////////////////////////////////////////////////////// UNIMPLEMENTED CLASSES ===================== accessibility ------------- All Accessibility handling has been commented out to save the download footprint. This removes the need for sun.misc.SharedSecrets as well. Nothing says we could not implement accessibility. We just didn't. security -------- All JavaScript security is handled by the browser natively. Thus, Java security checking is no longer necessary, and has been simplified to work without native security checking. Note that private methods in a class are REALLY private. serialization ------------- All serialization has been removed. It was never very useful for Swing anyway, because one needs exactly the same Java version to save and restore serialized objects. keyboard accelerators and mnemonics ----------------------------------- This work was completed in the spring of 2019. Note that in a browser, some key strokes, particularly CTRL-keys, are not available. Bummer. MINOR ISSUES--required some rewriting/refactoring by Bob and Udo ================================================================ Thread.currentThread() == dispatchThread ---------------------------------------- changed to JSToolkit.isDispatchThread() MINOR ISSUES--requiring some rewriting/refactoring outside of SwingJS ===================================================================== restrictions on long -------------------- Java's 64-bit long type is not supported in JavaScript. There is no Int64Array in JavaScript, and 0x20000000000000 + 1 evaluates to 0x20000000000000, not 0x20000000000001. (Likewise, -0x20000000000000 - 1 is left unchanged.) The largest "integer" value in JavaScript is 9007199254740991 (9.007199254740991E13, or 0x1FFFFFFFFFFFFFF). Effectively, you get to use only 53 bits of the long, not 64. Trying to set a long larger than 0x1FFFFFFFFFFFFFF or smaller than -0x1FFFFFFFFFFFFFF will result in a NumberFormatException. The transpiler handles conversion to long the same as Java for all cases other than from double. For small double values, there is no problem, and, in fact, this is a known trick used to round doubles and floats toward zero: double d; d = (long) 3.8; assert(d == 3); d = (long) -3.8; assert(d == -3); SwingJS will evaluate (long) d as 0 for d > 9007199254740991 or d < -9007199254740991, same as Java returns for Double.NaN. So, in Java we have: assert(((long) Double.NaN) == 0); assert(((int) Double.NaN) == 0); assert(((long) Float.NaN) == 0); assert(((int) Float.NaN) == 0); and also, in JavaScript only, we also have: double d = 0x2000000000000L; assert(((long) d) == 0); restrictions on BitSet and Scanner ---------------------------------- Because of the issue of long being only 53 bits, any time a method returns a long value, considerations must be made as to whether this will work in JavaScript. In particular, BitSet and Scanner have issues. In SwingJS, java.util.BitSet has been implemented as a 32-bit integer-based bitset. This was no problem in Java 6, but starting with Java 7, a method was added to BitSet that allows for the extraction of the underlying long[] word data. This is not work in JavaScript. Instead, SwingJS java.util.Bitset.toLongArray() will deliver 32-bit int[] data. SwingJS Scanner has hasNextLong() and nextLong(), and although it will scan through long numbers, Scanner will choke on long numbers greater than the JavaScript 53-bit limit. hasNextLong() will return false, and nextLong() will throw an InputMismatchException triggered by the NumberFormatException thrown by Long.parseLong(). HashMap, Hashtable, and HashSet iterator ordering ------------------------------------------------- In Java, iterators for HashMap, Hashtable, and HashSet do not guarantee any particular order. From the HashMap documentation for Java 8: This class makes no guarantees as to the order of the map; in particular, it does not guarantee that the order will remain constant over time. Likewise, for HashSet (because it is simply a convenience method for HashMap: [HashSet] makes no guarantees as to the iteration order of the set. JavaScript's Map object is different. It is basically a LinkedHashMap, so it guarantees iteration in order of object addition. Starting with java2script 3.2.9.v1, these classes use the JavaScript Map object rather than hash codes whenever all keys are strictly of JavaScript typeof "string". If any key is introduced that is not a string, the implementation falls back to using hash codes, the same as Java. Note strings created using new String("xxxx") are NOT typeof "string"; they are typeof "object". The result is significantly faster performance (3-12 x faster) than originally, and up to 3 x faster performance in JavaScript than in Java itself. Right. Faster than Java. The JavaScript Map implementation is implemented UNLESS the constructor used is the one that specifies both initial capacity and load factor in their constructor. Thus, new Hashtable() new HashMap() new HashMap(16) new HashSet() all use the JavaScript Map. But new Hashtable(11, 0.75f) new HashMap(16, 0.75f) new HashSet(16, 0.75f) do not. This design allows for opting out of the JavaScript Map use in order to retain the exact behavior of iterators in JavaScript as in Java. interning, new String("xxx") vs "xxx" ------------------------------------- Note that the following are true in JavaScript: typeof new String("xxxx") == "object" typeof "xxxx" == "string" var s = "x";typeof ("xxx" + s) == "string" There is no equivalence to this behavior in Java, where a String is a String is a String. Be aware that SwingJS does not always create a JavaScript String object using JavaScript's new String(...) constructor. It only does this for Java new String("xxxx") or new String(new String()). In all other cases, new String(...) (in Java) results in a simple "xxxx" string in JavaScript. That is, it will be JavaScript typeof "string", not typeof "object". The reason for this design is that several classes in the Java core use toString() methods that return new String(), and those classes that do that would cause a JavaScript error if implicitly stringified if new String() returned a JavaScript String object. This is fine in JavaScript test1 = function() { return { toString:function(){ return "OK" } } } "testing" + new test1() >> "testingOK" But for whatever reason in JavaScript: test2 = function() { return { toString:function(){ return new String("OK") } } } "testing" + new test2() >> Uncaught TypeError: Cannot convert object to primitive value The lesson here is never to use return new String("..."); in a Java toString() method. In Java it will be fine; in JavaScript it will also be fine as long as that method is never called in JavaScript implicitly in the context of string concatenation. A note about interning. Consider the following six Java constructions, where we have a == "x"; "xxx" "xx" + "x" new String("xxx").intern() new String("xxx") "xx" + a.toString() "xx" + a All six of these will return java.lang.String for .getClass().getName(). However, the first three are String literals, while the last three are String objects. Thus: "xxx" == "xxx" "xxx" == "xx" + "x" "xxx" == new String("xxx").intern() but none of the other three are equivalent to "xxx" or each other: "xxx" != new String("xxx") "xxx" != "xx" + a.toString() "xxx" != "xx" + a new String("xxx") != new String("xxx") "xx" + a != new String("xxx") etc. As in Java, in SwingJS, all of the following Java assertions pass as true: assert("xxx" == "xx" + "x"); assert("xxx" == ("xx" + a).intern()); assert("xxx" === new String("xxx").intern()); and both of these do as well: assert(new String("xxx") != "xxx"); assert(new String("xxx") != new String("xxx")); But the following two fail to assert true: assert("xxx" != "xx" + a); assert("xxx" != "xx" + a.toString()); because in JavaScript, both of these right-side expressions evaluate to a simple "interned" string. In Java, however, these assertions are true because Java implicitly "boxes" String concatentaion as a String object, not a literal. Most of us know not to generally use == with Strings unless they are explicitly interned. Where this problem may arise, though, is in IdentityHashMap, which compares objects using System.identityHashCode(), which is not the same for different objects or their string literal equivalents. My recommendation, if you need to use IdentityHashMap with strings is to always use an explicit String.intern() for any keys -- unless you really want to keep every string as separate keys even if they are the same sequence, in which case, use new String(). This will work in Java and in JavaScript. Be aware when working with strings that come from SwingJS and are being used by other JavaScript modules that those that are String objects will return "object" for the JavaScript typeof operator, not "string". The easy way to ensure this is no problem is to concatenate strings with "" to force immediate interning: var x = aJavaObject.getString() + ""; unless you are certain that the string is being returned is a raw JavaScript string. Names with "$" and "_" ---------------------- For the most part, this should be no problem. Note that the use of $ and _ in Java field names has always been discouraged: [] You may find some situations where auto-generated names will contain the dollar sign, but your variable names should always avoid using it. A similar convention exists for the underscore character; while it's technically legal to begin your variable's name with "_", this practice is discouraged. Some impacts of transpiling method names with full qualification: 1) SwingJS will introduce fields that start with $ or _. These will not conflict if the above convention is followed. 2) Fields that have the same Java name as a method are not an issue. 3) Fields that have a Java name with $ that matches a transpiled method name, such as toString$, will need to be refactored in Java to not have that name collision. 4) Fields in a subclass that have the same name as private fields in a superclass represent a name collision, because the superclass method needs to call its private field even if invoked from a subclass. The solution was to modify the subclass field name using one or more prepended $. 5) Use of Class.getDeclaredMethods() reflection will return Method objects having the transpiled name, not the Java name. This could require some j2sNative adjustment to strip the $... parameters from the name if that is needed. 6) Use of Method.getParameterTypes() should work fine, provided class names do not contain "_". This is because the transpiler converts "." to "_" when creating the fully qualified JavaScript name. positive integers do not add to give negative numbers ----------------------------------------------------- In Java, the following is true: 2000000000 + 2000000000 == -294967296 But in SwingJS, that will be 4000000000. So, for example, the following strategy will fail in SwingJS: int newLength = lineBuf.length * 2; if (newLength < 0) { newLength = Integer.MAX_VALUE; } "-1" in JavaScript is not 0xFFFFFFFF. And one must take care to not compare a negative number with a 32-bit mask. So (b & 0xFF000000) == 0xFF000000 is true in Java for (int) b = -1, but is false in JavaScript, because 0xFF000000 is 4278190080, while (-1 & 0xFF000000) is, strangely enough, -16777216, and, in fact, (0xFF000000 & 0xFF000000) != 0xFF000000 because -16777216 is not 4278190080. The fix is that one must compare similar operations: if ((b & 0xFF000000) == (0xFF000000 & 0xFF000000)) ..... Importantly, the JavaScript Int32Array does behave properly. From the Firefox developer console: >> x = new Int32Array(1) <- Int32Array(1) [ 0 ] >> x[0] = 4000000000 <- 4000000000 >> x[0] <- -294967296 Notice that, perhaps unexpectedly, the following two constructs produce different results in JavaScript: x = new Int32Array(1); b = x[0] = 4000000000; (b will be 4000000000) and x = new Int32Array(1); x[0] = 4000000000; b = x[0]; (b will be -294967296) SwingJS leverages array typing to handle all byte and short arithmetic so as to ensure that any byte or short operation in JavaScript does give the same result in Java. The design decision to not also do this with integer math was a trade-off between performance and handling edge cases. ArrayIndexOutOfBounds --------------------- You cannot implicitly throw an ArrayIndexOutOfBoundsException in JavaScript. JavaScript will simply return "undefined", not throw an Exception. So: boolean notAGoodIdeaIsOutOfBounds(String[] sa, int i) { try { return (sa[i] == sa[i]); } catch (ArrayIndexOutOfBoundsException e) { return false; } } will work in Java but not in JavaScript. Code should not depend upon this sort of trap anyway, if you ask me. Throwable vs Error vs Exception ------------------------------- True JavaScript errors are trapped as Throwable, whereas you can still trap Error and Exception as well. So if you want to be sure to catch any JavaScript error, use try{}catch (Throwable t){}, not try{}catch (Exception e){}. j ava.awt.Color -------------- ColorSpace: only "support" CS_sRGB. TODO -- any volunteers?? javax.swing.JFileDialog ----------------------- HTML5 cannot expose a file reading directory structure. But you certainly can still do file reading and writing. It just works a little differently. It's a simple modification: b = new JButton("FileOpenDialog"); b.addActionListener(new ActionListener() { @Override public void actionPerformed(ActionEvent e) { JFileChooser fc = new JFileChooser(); Test_Dialog.this.onDialogReturn(fc.showOpenDialog(Test_Dialog.this)); // Java will wait until the dialog is closed, then enter the onDialogReturn method. // JavaScript will exit with NaN immediately, and then call back with its actual value // asynchronously. } }); public void onDialogReturn(int value) { if (value != Math.floor(value)) return; // in JavaScript, this will be NaN, indicating the dialog has been opened // If we are here, the dialog has closed, in both Java and JavaScript. System.out.println("int value is " + value); } @Override public void propertyChange(PropertyChangeEvent event) { Object val = event.getNewValue(); String name = event.getPropertyName(); System.out.println(name); switch (event.getSource().getClass().getName()) { case "javax.swing.JOptionPane": switch (name) { case "inputValue": onDialogReturn(val); return; case "value": if (val instanceof Integer) onDialogReturn(((Integer) val).intValue()); else onDialogReturn(val); return; } break; case "javax.swing.ColorChooserDialog": switch (name) { case "SelectedColor": onDialogReturn(val); return; } break; case "javax.swing.JFileChooser": switch (name) { case "SelectedFile": File file = (File) val; byte[] array = (val == null ? null : /** @j2sNative file.秘bytes || */ null); onDialogReturn("fileName is '" + file.getName() + "'\n\n" + new String(array)); return; } break; } System.out.println( event.getSource().getClass().getName() + " " + event.getPropertyName() + ": " + event.getNewValue()); } Developers are encouraged to create a separate class that handles general calls to JFileDialog. An example class can be found in the SwingJS distribution as /sources/ javax.swing.JOptionPane dialogs ------------------------------- For this action to work, the parentComponent must implement propertyChangeListener, and any call to JOptionPanel should allow for an asynchronous response, meaning that there is no actionable code following the call to the dialog opening. In addition, for compatibility with the Java version, implementation should wrap the call to getConfirmDialog or getOptionDialog in a method call to handle the Java: onDialogReturn(JOptionPane.showConfirmDialog(parentFrame, messageOrMessagePanel, "title", JOptionPane.OK_CANCEL_OPTION)); Then parentFrame.propertyChange(event) should also call onDialogReturn. This will then work in both Java and JavaScript. Note that there is an int and an Object version of onDialogReturn(). In JavaScript: The initial return from JOptionPane.showConfirmDialog and showMessageDialog will be (SwingJS) JDialog.ASYNCHRONOUS_INTEGER (NaN), testable as an impossible Java int value using ret != -(-ret) if the parent implements PropertyChangeListener, or -1 (CLOSE_OPTION) if not. For showOptionDialog (which returns Object) or showInputDialog (which returns String), the initial return will be (SwingJS) JDialog.ASYNCHRONOUS_OBJECT, testable as ((Object) ret) instanceof javax.swing.plaf.UIResource if the parent implements PropertyChangeListeneer, or null if not. The second return will be the desired return. In Java: The initial return will be the one and only modal final return. For full compatibility, The calling method must not continue beyond this call. All of the standard Java events associated with Components are also available. Certain fall back mechanisms are possible, where onReturn does not exist, but only for the following cases: For showMessageDialog, for WARNING_MESSAGE and ERROR_MESSAGE, a simple JavaScript alert() is used, returning 0 (OK_OPTION) or -1 (CLOSED_OPTION). For showInputDialog, if the message is a string, a simple JavaScript prompt() with input box is used, returning the entered string or null. For showConfirmDialog, a simple JavaScript confirm() is used, in which case: for YES_NO_OPTION: YES_OPTION or NO_OPTION for YES_NO_CANCEL_OPTION: YES_OPTION or CANCEL_OPTION for OK_CANCEL_OPTION or any other: OK_OPTION or CANCEL_OPTION Note that you should implement a response for CLOSED_OPTION for showConfirmDialog. For other dialogs, a null return indicates the dialog was closed, just as for Java. Developers are encouraged to create a separate class that handles general calls. An example class can be found in the SwingJS distribution as src/javajs/async/ Very simple modifications to the Java allows asynchronous operation using AsyncDialog. Here is a simple "do you want to close this frame" example, where you can see that what we have done is to set the reply into an ActionListener that is defined in the constructor of the AsyncDisplay object: // Original: // // private void promptQuit() { // int sel = JOptionPane.showConfirmDialog(null, PROMPT_EXIT, NAME, JOptionPane.YES_NO_OPTION); // switch (sel) { // case JOptionPane.YES_OPTION: // resultsTab.clean(); // seqs.dispose(); // if (fromMain) { // System.exit(0); // } // break; // } // } private void promptQuitAsync() { new AsyncDialog(new ActionListener() { @Override public void actionPerformed(ActionEvent e) { int sel = ((AsyncDialog)e.getSource()).getOption(); switch (sel) { case JOptionPane.YES_OPTION: resultsTab.clean(); seqs.dispose(); if (fromMain) { System.exit(0); } break; } }}).showConfirmDialog(null, PROMPT_EXIT, NAME, JOptionPane.YES_NO_OPTION); } Very simple! native methods -------------- The J2S compiler ignores all static native method declarations. Anything of this nature needs to be implemented in JavaScript if it is needed, using j2sNative blocks: /** * @j2sNative * * var putYourJavaScriptCodeHere * */ Note that if you follow that directly with a {...} block, then the javadoc code will run in JavaScript, and the {...} code will run in Java. key Focus --------- As of June, 2019, the keyboard focus manager is fully implemented. The one catch is that JTextPane and JTextArea, which already consume VK_TAB in Java, cannot use CTRL-TAB to continue a tabbing cycle around the components in a window. Instead, CTRL-TAB is absorbed by the browser. LookAndFeel and UI Classes -------------------------- SwingJS implements the native browser look and feel as swingjs.plaf.HTML5LookAndFeel. There are small differences between all look and feels -- MacOS, Windows, SwingJS. Expert developers know how to coerce changes in the UI by subclassing the UI for a component. This probably will not work in SwingJS. Note that LookAndFeel in Java usually determines canvas size in a Frame because different operating systems (Mac OS vs Windows vs HTML5) will have different edge sizes on their frames. If you want to ensure a component size, use getContentPane().setPreferredSize(). System.exit(0) does not stop all processes ------------------------------------------ Although System.ext(int) has been implemented in JavaScript, it just closes the frames, stops all pending javax.swing.Timer objects in the queue, and runs any threads added using Runtime.getRuntime().addShutdownHook(Thread). It may not stop all "threads." So don't rely on that. Applications are responsible for shutting down prior to executing System.exit(0). myClass.getField not implemented -------------------------------- java.lang.reflect.Field is implemented minimally. It is not certain that Field.getDeclaringClass() will work. If you just want a value of a field, you can do this: /** *@j2sNative * * return myClass[name] */ But that is not a java.lang.reflection.Field object. "window" and other reserved JavaScript names -------------------------------------------- No reserved top-level JavaScript name is allowed for a package name. So, for example, one must rename packages such as "window" or "document" to names such as "win" or "doc". reserved field and method names ------------------------------- In order to minimize the chance of added SwingJS field and method names colliding with ones developers might use in subclassing Java classes, we have added U+79D8 (first character of Mandarin "secret") to the characters already disrecommended by Java documentation ("$" and "_"). The only problem would be if you use that character followed by certain English words in certain classes. For example \u79D8canvas for JComponents (in java.awt.JSComponent) and \u79D8byte (in qualified field and method names -------------------------------- Method names in SwingJS are fully qualified, meaning two methods with the same Java name but different parameters, such as write(int) and write(double), must not have the same name in JavaScript. (In this case, we will have write$I and write$D.) However, in certain cases it may be desirable to leave the method names unqualified. In particular, when an interface actually represents a JavaScript object, the transpiler can leave a method name unqualified. The default situation for this is a class name includes ".api.js" (case-sensitive). This means that any method in any class in a package js within a package api, or any private interface js that has an outer interface api, will have all-unqualified methods. An example of this is swingjs.plaf.JSComboPopupList, which needs to communicate with a jQuery object directly using the following interface: private interface api { interface js extends JQueryObject { abstract js j2sCB(Object options); abstract Object[] j2sCB(String method); abstract Object[] j2sCB(String method, Object o); abstract Object[] j2sCB(String method, int i); abstract int j2sCB(String OPTION, String name); } } Notice that all these variants of j2sCB() will call the same method in JavaScript by design. missing Math methods -------------------- java.lang.Math is worked out, but some methods are missing, either because they involve long integer value that are inaccessible in JavaScript, or because I just didn't implement them. This is a result of continued Java development. It is easy enough to add these methods if you have the source. They go into j2sClazz.js, which is combined with other initial libraries into swingjs2.js by build_site.xml Component.getGraphics(), Graphics.dispose() ------------------------------------------- Use of component.getGraphics() is discouraged in Java and in SwingJS. Specifically in SwingJS, any call to component.getGraphics() or BufferedImage.createGraphics() or Graphics.create(...) should be matched with graphics.dispose(), particularly when it is called outside the context of a paint(Graphics) call from the system. If you see your graphics scrolling down the page with each repaint, look for where you have used Component.getGraphics() and not Graphics.dispose(). For example, this will definitely NOT work in SwingJS: this.paint(getGraphics()) and really should not work in Java, either, as it is technically a resource memory leak. Instead, if you really do not want to use repaint(), use this: Graphics g = getGraphics(); paint(g); g.dispose(); Graphics.setClip() ------------------ The HTML5 canvas.clip() method is permanent. You can only reset the clip using save/restore. This is different from Java, where you can temporarily change it using Shape oldClip = Graphics.getClip(); Graphics.setClip(newClip); ... Graphics.setClip(oldClip); If you need to do something like this, you must schedule the paints to not have overlapping clip needs. MAJOR ISSUES--for Bob and Udo within SwingJS ============================================ fonts ----- Fonts and FontMetrics will all be handled in JavaScript. Font matching will not be exact, and composite (drawn) fonts will not be supported. SwingJS handles calls such as font.getFontMetrics(g).stringWidth("xxx") by creating a
containing that text, placing it in an obscure location on the page, and reading div.getBoundingClientRect(). This is a VERY precise value, but can be a pixel or two off from what Java reports for the same font. OS-dependent classes -------------------- Static classes such as: java.awt.Toolkit java.awt.GraphicsEnvironment which are created using Class.forName are implemented using classes in the swingjs package. AWTAccessor is not implemented. AWT component peers and component "ui" user interfaces ------------------------------------------------------ ComponentPeer is a class that represents a native AWT component. Components with such peers are called "heavy-weight" components. They are expected to do the dirty work of graphics drawing. Java Swing implements peers only for JApplet, JDialog, JFrame, and JWindow. References to such objects have been removed, but clearly there must be some connection to similar DOM objects, even for "light-weight" components. MAJOR ISSUES--to be resolved by implementers ============================================ fonts ----- Glyph/composite/outline fonts are not supported. threads ------- Thread locking and synchronization are not relevant to JavaScript. Thus, anything requiring "notify.." or "waitFor.." could be a serious issue. All threading must be "faked" in JavaScript. Specifically not available is: Thread.sleep() javax.swing.AbstractButton#doClick(pressTime) will not work, as it requires Thread.sleep(); However, java.lang.Thread itself is implemented and used extensively. Methods thread.start() and both work fine. For simple applications that use Thread.sleep() just to have a delay, as in a frame rate, for example, one can use javax.swing.Timer instead. That is fully implemented. Likewise, java.util.Timer can be replaced with no loss of performance with javax.Swing.Timer. Note that java.util.TimerTask is implemented, but it can also be replaced by an implementation of Runnable. task = new TimerTask(){....}; t = new java.util.Timer(); t.schedule(task, 0, 1); becomes task = new TimerTask(){....}; // or task = new Runnable() {...} t = new javax.swing.Timer(1, new ActionListener() { @Override public void actionPerformed(ActionEvent e) {; } }; t.setInitialDelay(0); // not particularly necessary t.start(); In addition, SwingJS provides swingjs.JSThread, which can be subclassed if desired. This class allows simple while(!interrupted()){ wait() ... } action through an asynchronous function run1(mode). For example: protected void run1(int mode) { try { while (true) switch (mode) { case INIT: // once-through stuff here mode = LOOP; break; case LOOP: if (!doDispatch || isInterrupted()) { mode = DONE; } else { Runnable r = new Runnable() { public void run() { // put the loop code here } }; dispatchAndReturn(r); if (isJS) return; } break; // add more cases as needed case DONE: // finish up here if (isInterrupted()) return; // or here break; } } finally { // stuff here to be executed after each loop in JS or at the end in Java } } image loading ------------- - All image loading in SwingJS is synchronous. A MediaTracker call will immediately return "complete". However, it still may take one system clock tick to fully load images. Thus, it is recommended that images be preloaded in the static block of the applet if it is necessary that they be available in init(). This is only an issue if you are trying to access the pixel buffer of the image in JavaScript. - Applet.getImage(path, name) will return null if the image does not exist. - BufferedImage: only "support" imageType RGB and ARGB -BH: This is a temporary edit, just to get us started. Certainly GRAY will be needed BigInteger and BigDecimal ------------------------- java.math.BigInteger is fully supported; java.math.BigDecimal is roughed in and not fully tested (07/2019). Both classes present significant issues for JavaScript, as they are based in Java's 64-bit long for all their operations. Here is the JavaDoc note I added to BigInteger: * SwingJS note: Because of the limitations of JavaScript with regard * to long-integer bit storage as a double, this implementation drops * the integer storage bit length to 24, giving 48 for long and leaving * the last 16 bits clear for the exponent of the double number. This should * not affect performance significantly. It does increase the storage * size by about 33%. By bringing an "int" to 3 bytes, we can easily construct * and use byte[] data intended for the original BitSet. "Easily" may be a bit strong there. This was a serious challenge. BigDecimal seems to run normally, but in order to do that, my hack involves reducing the size of an integer that is allowed to be stored as such and not in byte[] as a BigInteger. I'm sure there is a performance hit, but it does work. no format internationalization ------------------------------ For now, just en for number and date formatters no winding rules ---------------- When filling a graphic, only nonzero winding rule is implemented in HTML5 Canvas2D. text-related field implementation --------------------------------- Text fields are: JTextField (JavaScript ) JTextArea (JavaScript