[Inspector][EFL] Can't resume a special break point on EFL inspector

[Gyuyoung Kim]

As mentioned in Bug 129247, the test page has some problems. One is a crash when enabling JIT, the other is that a break point is not resumed if a break point is set inside "addEventListener" when disabling JIT.

Reproduced steps :

1. Turn JIT off on EFL port. (Source/cmake/WebKitFeatures.cmake)
1. Tools/Script/build-webkit --efl --cmakeargs="-DSHARED_CORE=ON" --debug
2. WebKitBuild/Debug/bin/MiniBrowser ./inspector.html
3. Run inspector by using context menu after clicking mouse right button.
4. Set a break point in below line of inspector.js

    var textbox = document.querySelector('.contents');
    textbox.addEventListener("click", function(){
=>      box = document.querySelector('#textbox');
        box.innerHTML = box.innerHTML == "Basic" ? "Sample" : "Basic";
    });

5. Click "Basic" test on MiniBrowser.
6. The break point is cached.
7. Press "resume" button on inspector.
8. Looks look up.


According to my investigation until now, inspector can't receive "resume" event from Inspector frontend(UIProcess).

[Radar WebKit Bug Importer]

<rdar://problem/16158460>

[Gyuyoung Kim]

Created attachment 225131 [details]
Inspector.html

[Gyuyoung Kim]

Created attachment 225132 [details]
Inspector.js

[Csaba Osztrogonác]

After r164651, it can be reproducible with enabled/disabled JIT
in release/debug mode too.

[Gyuyoung Kim]

(In reply to comment #4)
> After r164651, it can be reproducible with enabled/disabled JIT
> in release/debug mode too.

According to current our investigation, this problem occurs when events should be processed as a nested loop in ecore_main_loop. Because the ecore_main_loop doesn't support a nested loop basically.

So, ecore_main_loop_iterate() can't process "continue" inspector button as soon as a js click event occurs.


*ecore_main_loop_iterate()*

"Runs a single iteration of the main loop to process everything on the queue.

It does everything that is already done inside an Ecore main loop, like checking for expired timers, idlers, etc. But it will do it only once and return, instead of keep watching for new events.

DO NOT use this function unless you are the person God comes to ask for advice when He has trouble managing the Universe."

http://docs.enlightenment.org/auto/emotion/group__Ecore__Main__Loop__Group.html#ga7f5463c1d4f3f020968ed06d6e5816cb

[Csaba Osztrogonác]

Apple ran into similar bug - https://bugs.webkit.org/show_bug.cgi?id=130032

[Timothy Hatcher]

That one was very specific to JSC debugging. We didn't have a similar issues in WebCore debugging.

[SangYong Park]

Created attachment 227838 [details]
Patch

[WebKit Commit Bot]

Attachment 227838 [details] did not pass style-queue:


ERROR: Source/WTF/wtf/efl/RunLoopEfl.cpp:112:  This { should be at the end of the previous line  [whitespace/braces] [4]
Total errors found: 1 in 6 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Jinwoo Song]

Comment on attachment 227838 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=227838&action=review

> Source/WTF/ChangeLog:7
> +

You'd better to add description for this patch.

[Gyuyoung Kim]

CC'ing Byungwoo.

[SangYong Park]

Created attachment 229114 [details]
Patch

[Ryuan Choi]

(In reply to comment #12)
> Created an attachment (id=229114) [details]
> Patch

How was it going?

I merged patch into EFL for this.
(https://phab.enlightenment.org/rEFL203006799186d47851509514786d05b4ae7fb644)

So, another option is adding patch at the ecore package of jhbuild.

[Gyuyoung Kim]

(In reply to comment #13)
> (In reply to comment #12)
> > Created an attachment (id=229114) [details] [details]
> > Patch
> 
> How was it going?
> 
> I merged patch into EFL for this.
> (https://phab.enlightenment.org/rEFL203006799186d47851509514786d05b4ae7fb644)
> 
> So, another option is adding patch at the ecore package of jhbuild.

If we can solve this problem by using ecore's patch, I prefer to use it.

[Tibor Mészáros]

Created attachment 233233 [details]
Patch

It's a patch, that based on SangYong Park's work.

[Tibor Mészáros]

Created attachment 233234 [details]
Patch

The previously patch has been mistakenly uploaded, sorry about it. This will be the good one.

[Gyuyoung Kim]

(In reply to comment #13)
> (In reply to comment #12)
> > Created an attachment (id=229114) [details] [details]
> > Patch
> 
> How was it going?
> 
> I merged patch into EFL for this.
> (https://phab.enlightenment.org/rEFL203006799186d47851509514786d05b4ae7fb644)
> 
> So, another option is adding patch at the ecore package of jhbuild.

Ryuan, I wonder if the EFL patch is already merged into efl 1.9. Could you check it ?

[Ryuan Choi]

(In reply to comment #17)
> (In reply to comment #13)
> > (In reply to comment #12)
> > > Created an attachment (id=229114) [details] [details] [details]
> > > Patch
> > 
> > How was it going?
> > 
> > I merged patch into EFL for this.
> > (https://phab.enlightenment.org/rEFL203006799186d47851509514786d05b4ae7fb644)
> > 
> > So, another option is adding patch at the ecore package of jhbuild.
> 
> Ryuan, I wonder if the EFL patch is already merged into efl 1.9. Could you check it ?

No, it's not merged at 1.9 branch and 1.9 will not be released anymore.
It will be available from the 1.10.
FYI, latest version for 1.9.X is 1.9.5

So, what do you think about adding a patch in the jhbuild ?

[Tibor Mészáros]

Created attachment 233427 [details]
Patch v3

I agree with Gyuyoung Kim, we can solve this problem by using ecore's patch. I had tested it, and it's working like a charm.

[Gyuyoung Kim]

Comment on attachment 233427 [details]
Patch v3

LGTM. I prefer to use this approach to fix this problem.

[Csaba Osztrogonác]

Comment on attachment 233427 [details]
Patch v3

Clearing flags on attachment: 233427

Committed r170236: <http://trac.webkit.org/changeset/170236>

[Csaba Osztrogonác]

Comment on attachment 229114 [details]
Patch

Fixing EFL was preferred to workarounding WebKit.