Eclipse The Never Ending Source Of Blog Posts

The other month, I tried to filter out classes like "java.*" in the Open Type dialog, with little success; see http://stackoverflow.com/questions/11936524/filter-out-our-own-types-in-eclipse-open-type, for some details. It does not seem to be possible, to any larger extent of usefulness. Now, I forgot about it and moved on with my...so called life. But just now, this episode came back to at me. Trying to auto-assist at "import java.|" (the | being my caret), no java.* type name showed up. Weird! Annoying! Has my RTE library vanished? Did Oracle decide to cancel Java? No.

I had left my filter pattern active (the checkbox checked). It's the correct behavior, of course, we should certainly not expect the implementation to be over-complexified to handle this case, even though it is a bit funny to not get a match on your very explicit request to match "java.|" due to a filtering pattern "java.*". If you have the same sense of humor as me. Let's just be happy computers are still not anywhere near human domination in terms of intelligence. That, or they just don't want to help us.

Undefined Timeout Burns

Doing some pleasurable (not overly) JavaScript development (slow...). I understand the advantages of dynamic languages (no I don't quite understand it, but I am being polite), but sometimes (most of the time) it burns.

var timeout_millis    = outer_params.timeout_millis;
var root_path         = outer_params.root_path;  

function loop_callback() {
  setTimeout(timeout_callback, timeout_millis);
  function timeout_callback() {
    var params = outer_params;
    do_something_and_then_call(params, loop_callback);
  }
}
loop_callback();

Simple enough, right? 

But this code failed miserably. 100% CPU. JS debugger locks up. JS console locks up. Good grief. 

Turns out I didn't set the 'timeout_millis' of 'outer_params'. So 'timeout_millis' will be undefined. And what does setTimeout() do? Does it think this a bit strange? No. It thinks zero is an excellent choice here. Thus your loop will really run wild, with the aforementioned effects. Making it hard to find your mistake, since there is no separation between the debugger and the debuggee code in Chrome, there's no separate debugger thread that one could protect from the debugged thread or process by prioritizing the debugger. The debugger simply locks up, partly because console gets too much ouput. 

Here's an immediate thing to help avoid the same situation in the future: 

if (!timeout_millis) {
  alert("timeout_millis == " + timeout_millis);
}

 And maybe

(function () {
  var original_setTimeout = window.setTimeout;
  window.setTimeout = function (callback, timeout_millis) {
    if (timeout_millis === undefined) {
      alert("timeout_millis == " + timeout_millis);
    }
    original_setTimeout(callback, timeout_millis);
  };
})();

Not well formatted, and untested, but you get the idea. 

Motivation II

One writes about motivation when it's gone missing.

There's this kind of motivation block that seems to be physiological. It's hard to describe. Not quite tired, but it can happen when you've underslept; but I think the most common time is after a night of normal sleep catching up accumulated undersleep. It's a kind of nervousness, isn't it. It can feel rather like excitement, actually. But still blocked, so there's some kind of tension, thus the nervous feel to it.

Eating a little too much can probably also be a contributing factor.

Coffee or caffeine doesn't really help. It rather adds to the nervousness, although I haven't done any study on it.

I don't know what to do about it. Walking doesn't help either. Possibly relaxing, thoroughly, but that's a pretty paradoxical thing to be able to do at work.

Wait it out. How long?

One can do some cleaning. The folders named like "untitled folder 21" in the Desktop folder, for example.

Logging Circular JSON

Something like this:

console.log(FUNCTION_NAME + " " + JSON.stringify(params));

Is often an easy way to get good debug logging. Except when:

Chrome 'Sources' Note

What the -- ?

There is one script that doesn't show up in the Chrome 'Sources' tool. Same page in Safari, the script shows up.

The scripts/sources GUI design seems to have changed for the worse in Chrome, BTW. I have to click to get the list of sources to show up, each time it seems. And the target to click is really small.

But it's easy to fix. Observe screenshot:

See the little two-pane icon under Network? It looks like this ascii graph: "[| ]". Click that, and you get (back to) the favored (by me) GUI: the source list is always showing, clickable.

If you have a small screen, then Maybe the other mode is better. Maybe.

---

This is strange: now my js source that was missing, shows! Is it the same weird Chrome caching behavior that I've experienced many times, and that I imagine I am not just imagining... or am I crazy?

EDIT: no, it's not easy to 'fix': the GUI layout reverts next time you reload. Ditch-worthy. Back to Safari, I guess. EDIT: no, I meant 'when you load a new tab'.

Safe Reformatting of Javadoc Hyperlinks

 I was gonna write, and did write, a little piece on Safe Reformatting of Javadoc Hyperlinks. Ironically, the blogger.com rendered the blog post wrong. I then clicked 'Go back to draft' or whatever it'd be called in English, because I thought I had published the thing. The saved version turned out to be blank after that. That sucked a bit.

I'll have to redo it then, and never mind the Javadoc examples that rendered wrong.

The trick is to use double quotes for your href hyperlink attribute value, not single quotes. The Eclipse Javadoc formatter will never break a doubly-quoted string, it seems. When I used single-quoted href values, the fomatter would often chop off characters from the link text one by one, and put on the next line, and also separate the "
"  would disintegrate too. With doubly quoted string, the '>' and link text will most probably end up on a new line, and be safe from chopping.
A second trick is to put spaces between the angles and the link text, e.g. '[...]> link-text
'. That will probably help with the above shredding too; the formatter will prefer to break on spaces, rather than between '<' and '/a>', which has happened to me a lot.
PS. Or somebody fix the Eclipse formatter... and while you're at it, add a better escape than @code and @literal, which don't work with '@' in the escaped text.

EDIT: Well look at that, the rendering is fuggled up again. Never mind. Here's another tip: never paste html into the blogger editor. And tip 2: edit your posts using not the blogger editor...

UPDATE: Just noticed that leaving a space before the link ('anchor') end tag makes the space be underlined; even if there is another space after the end tag. So leading space may be enough. Or no space at all, the double-quoting may be sufficient.

Eclipse Extract Method Not All Paths End In Return

Here's an almost interesting find. I had code like the following:

[statements]
if (...) {
  return;
}
[statements]
if (...) {
  return;
}
[statements, but not 'return;']

Did 'Extract Method' on the above section. Got an error for "Not All Paths End In Return", even though refactoring should work fine.

No big deal. Actually it makes sense, since refactoring means not to change meaning, and refactored meaning should probably be along this line:

refactoredMethod(...); return; 

And by that reasoning the error is good. But it so happened that in my attempted refactoring, the continuation of the refactored section had no more code, just getting out of the method, so the refactoring in my mind (no return in front) wouldn't have changed the meaning, if you follow me.

Anyhow. I just pushed the refactoring forward, overriding the error.

Stack Overflow - Me Subjective?

I was going to ask another question on the developer's go-to site #2 (Google search is number one, I believe) Stack Overflow. But before I finished the question, I get a 'subjective' warning/hint: 

Sort of funny-weird. Makes me wonder what kind of heuristic they employ for this. The wording so far actually contains nothing at all to go on. Well that's a reason to close the question of course, but subjective? Must be the hardest thing to make a computer decide if something is subjective. Maybe it's some statistical calculation: 'your' and 'own' could be common in asking for subjective opinion, like 'what are your own views on X?'.

And what makes it guess that this is likely to be a finished headline? Now of course, it can't, headlines don't even have to be sentences; and don't have to end with a question mark.

Maybe they do get a lot of subjective questions.

EDIT: Yep, 'your own' seems to trigger the subjectivity. Substituted 'our own'.

Jackson JsonMappingException: Infinite recursion

EDIT2: found a solution...see bottom.

EDIT: Ignore (some of) the stuff below :). Just occurred to me that it's not possible to check for circularity when a getter returns a new instance, and if that's the cause of the infinite recursion. And this seems to be the case for java.awt.Point's getLocation().

---- original text follows ----

What did the Jackson programmer/s do here?

org.codehaus.jackson.map.JsonMappingException: Infinite recursion (StackOverflowError) (through reference chain: java.awt.Point["location"]->java.awt.Point["location"]->java.awt.Point["location"]->java.awt.Point["location"]->java.awt.Point["location"]->java.awt.Point["location"]->java.awt.Point["location"]->java.awt.Point["location"]->java.awt.Point["location"]->java.awt.Point["location"]->java.awt.Point["location"]->jav...

This doesn't help very much.

Wouldn't it be both easier and better feedback to detect circularity directly?

Could be implemented as a [[Identity]Hash]Set, where objects to be serialized are added on entering and removed again on exit (to allow sharing).

Or there might be reasons to use the StackOverflowError; but then the chain should be shown from the root, not from the top, since that would show the path to your problem. If length is a problem, the part just before the cycle is usually the most interesting. Do the hash set thing, and on first hit, you know you have detected the cycle. One might even show both the path leading up to the cycle, and the objects making up the cycle. Very valuable feedback; and almost trivial to implement, if one happens to have the time to do it.

For those of you happen to be coming here in need of a solution: use JsonIgnore, for instance. Like this:

@JsonIgnore
public BufferedImage javaBufferedImage;

If a BufferedImage happens to be the thing that has a java.awt.Point. Or consider a redesign where you don't mix non-serializable types in with serializable stuff.

BTW, the reason Point is not directly Jackson-serializable, is it's getLocation() (also has setLocation()) getter, I think.

EDIT2:

Found a somewhat general way to debug infinite recursion: configure Jackson to output an indented stream. Then check for a reasonable indentation depth, and cut the output at that point, since it's almost guaranteed to be caused by the infinite recursion. This should make it possible to manually inspect the output.

A consideration here is the fact that the Java debug protocol, the Eclipse debugger, and/or the Eclipse console seems fundamentally unable to efficiently handle large amounts of text or lines of text. So you might want to output the text to a file for viewing in your favorite editor. Or suffer.

I don't have the time right now to publish the source code I use; maybe later. Parting gift: a (partial!) java BufferedImage as JSON to admire:

      "bufferedImage" : {
        "type" : 6,
        "data" : {
          "pixelStride" : 4,
          "scanlineStride" : 420,
          "dataOffsets" : [ 3, 2, 1, 0 ],
          "dataStorage" : "/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf/ezMX/3szF/97Mxf
[...]