The completion offers nothing if invoked for the second or next time after the file modification. The first completion invocation after the file modification works well. This is observable in JSP files, works well with .html files.

Correction: The completion works only during the short interval between modification of the file and parsing it, e.g. works only if the latest parser result is dirty.

There some of the observations I consider as strange (all tested on a JSP file w/ js and html code inside):

1) SourceCache.refresh() is triggered only by a caret move in the opened editor window. Isn't it weird since no file content has been modified?

2) After a modification of the file the calls to EPs are following:
a) JspEmbeddingPprovider called from SourceCache.refresh()
b) JsEmbeddingProvider (note the prefix is Js, not Jsp!!!) called for the first time from GsfHintsProvider by ParserManager.parse() 
c) JsEmbeddingProvider called for second time from SourceCache.refresh(). 

Since the embeddings have been already refreshed by the first call to the JsEP and the file hasn't been modified meanwhile there isn't (or is????) a reason why create the js embeddings again.

3) If I try following scenario: Type to the javascript section in the file and immediately invoke completion, then it behaves as follows: 
a) First JsEmbeddingProvider.getEmbeddings() called from HtmlCompletionQuery by PM.parse()
b) then JspEmbeddingProvider.getEmbeddings() called from the same place
c) then JspEmbeddingProvider.getEmbeddings() called from SourceCache.refresh()
d) finally JsEmbeddingProvider.getEmbeddings() called from SourceCache.refresh()

The #2 and #3 scenarios causes the resultIterator.getParserResult (caretOffset) at the GsfCompletionProvider:326 to return HtmlParserResult for the case #2 and JsParserResult for the scenario #3. But in both cases the JsParserResult should be returned. That is the reason why the javascript completion doesn't work normally and does work only in the scenario #3.

I must admit I am not an expert on the parsing.api so I'm passing the issue to parsing.api and kindly asking Tomas to take a look at it.

Some more info:
The snapshots given to the JsEmbeddingProvider are in all the scenarios and cases the same - the whole JSP file content and the resulting JS embedding are also always the same - correct.

There's also a nasty workaround for not creating the JS embedding from HTML virtual source created from JSP content. Look at JsEmbeddingProvider$Factory.create() for more info.

The current embeddings for JSP files are following AFAIK:

JSP -> HTML
JSP -> HTML -> CSS
JSP -> JS (the transitional JSP->HTML-JS is disabled by the workaround I mentioned)

Marku, how can I reproduce it?
Can you attach the file (project) which causes such a behavior and describe what should I do to reproduce it?
Dik

Created attachment 106979 [details]
A file to reproduce the issue

To reproduce the issue, just place caret somewhere in a white area in the <script> tag content and then follow the steps from comment #3. Briefly - modify the file (hit a spacebar for example), wait a second or two and invoke cc => nothing. Then modify the file and immediately invoke cc => you'll got the proper javascript results.

Thanks, Marku

I've looked at it and it doesn't seem as parsing api problem.
It seems rather as JSP ep problem.
I've opened the provided file and called cc as described above.

The HtmlCompletionQuery is trying to get result for offset 411 which is OK in the document (the place where the cursor is placed) but it's not OK in the provided snapshot which comes from the JSP in the snapshot the position points somewhere into body element.

Just to make it clear: We've clarified the problem mentioned in comment #10 offline and the conclusion is it is not an issue and the code is correct.

The problem is in wrong embedding layout in the JSP.
The jsp has following embeddings for the test document
jsp
  -javascript
  -java
  -html

while it should be


jsp
  -java
  -html
      -javascript

The javascript and html both are having the same positions(javascript is contained in html), so the first (which is random) wins.
As Marek explained that the wrong layout is needed and it's impossible to change it I am proposing
a fix to parsing api which will preserve the order of embedding providers. The order will ensure that javascript will win over the html.

Fixed jet-main a5396573a821

Thanks Tome for the fix. Just a few notes: I do not think the current embeddings layout is wrong. There's nothing forbidding such layout generation in the parsing api javadoc. I remember when I discussed the problem with Hanz some time ago and he said it is simply not defined how it should behave :-). Note #2 - I didn't say it is impossible to change the layout, but the way how it's done now has some advantages and redoing this now would be possibly more risky than the ordering fix you did. Thanks for your work and patience once again!

>it is simply not defined how it should behave :-)
This is exactly what this bug is about. It's not defined, sometimes the JS wins and CC works sometimes HTML wins and it does not as the both embedding have overlapping range. The resultIt.getResult(offset) does not know which one to choose as both are OK.
I've did a change which orders them according to task priority.

Integrated into 'main-golden', will be available in build *201103160400* on http://bits.netbeans.org/dev/nightly/ (upload may still be in progress)
Changeset: http://hg.netbeans.org/main/rev/a5396573a821
User: Tomas Zezula <tzezula@netbeans.org>
Log: #196429:Javascript code completion not working in JSP files

I'll verify it as soon as the build will be available

For now this is fixed in trunk, right? Adding 70_HR_FIX_CANDIDATE.

verified

The patch seems fine to me.

Integrated into NB 7.0 http://hg.netbeans.org/releases/rev/b5a6cd1e4f6c

I'm afraid I have to reopen this issue. In jsp files in common Java Web - Web Application it works fine, but in Java Web - Web Free form application, the completion doesn't work properly. 

I tried sample project from some TS http://wiki.netbeans.org/TS_61_WebFreeform (project archive is at the bottom) and opened time.jsp.

In time.jsp I simply added js code with function definition and tried to invoke code completion. The completion offers js items (decodeURI(), keywords...) but it doesn't offer anything that I have defined, no functions, no variables... Reopening/saving the file doesn't make any difference.

Product Version: NetBeans IDE 7.0 (Build 201104080000)
Java: 1.6.0_25
System: Windows XP version 5.1 running on x86; Cp1250; cs_CZ (nb)

The freeform case has nothing to do with this report.
This issues was about embedding ordering which is general not project specific.
Please create a new issue on the JSP or web/freeform.
Thanks

reverting reopening back to verified...