37262 - captureEvents method causes handler to trigger twice

Status: RESOLVED WONTFIX

(Core :: XUL, defect, P3)

Description

[prass]

On Linux only, CR on alert() method opens another alert message. Using the mouse 
click closes the alert as expected. Only Enter key causes this problem.

Windows and Macintosh behave as expected.

Build:20000426
Platform: Linux Red hat 6.0

Comment 1

[prass]

Attachment: Created test case for alert problem on linux

Comment 2

[Jesse Ruderman]

*** This bug has been marked as a duplicate of 29165 ***

Comment 3

[prass]

verifying as dupe. Tracking 29165 as the depends bug.

Comment 4

[John Morrison]

Reopening this bug (37262) since the dup is fixed, but this is not:

On *both* windows and linux builds for today, using *either* the mouse or CR to 
dismiss the dialog, I get a second instance of the alert dialog thrown up when
using the testcase above.

Adding dogfood, as prass had requested that review on bug #29162.

Comment 5

[John Morrison]

changing product/component

Comment 6

[leger]

Putting on [dogfood+] radar.

Comment 7

[prass]

Actually with the fix of 29165, the problem for which this bug was opened is 
also fixed. In AIM, enter key works fine. Marking works for me.

Changing QA contact to prass, since I can verify this on linux AIM.

Thanks!

Comment 8

[John Morrison]

Well, with a fresh build on linux, as well as with the a.m. verification builds
for mac and windows, this demo is still doing two odd things:
  
  1) every key event is processed twice on all three platforms
 
  2) characters in the printable ascii range (e.g. [a-zA-z0-9 etc.]) have
     a keycode of 0 (zero). Characters like DEL, ESC, TAB have their correct
     keycodes (although 4.x traps TAB and ESC before the app gets them [win32])

NOTE: this "double event" happens whether the alert is dismissed by CR, or by 
a mouse click [or if the alert is replaced by a dump()]

Removing dogfood(+) designations, since prass was willing to close for his 
requirements. Prashant -- do you have open DOM 0 bugs for the problems noted 
here?

Comment 9

[Dan M]

Key event processing. Tempted to dump on saari, but this might also be related to 
35921. Keeping for now and putting on same milestone.

Comment 10

[John Morrison]

bug 35921 -- 'event notification outlives its event' happens only on linux, 
though, but this bug is XP. 

By the way, see also bug 38889 -- 'click on label sends two onclicks to parent'
(although that may be specific to the way <label> relates to form controls).

Comment 11

[John Morrison]

Resummarizing from "Enter key on alerts opens multiple alert instead on 
closing it." 

Comment 12

[Dan M]

Note to self: read more carefully. John seems to have clinched it as a problem 
where event notifications are being fired twice. To saari, with compassion.

Comment 13

[saari (gone)]

I believe both behaviors are correct. The attached example sets up both a 
document.onkeypress and a document.capture. First the document.capture fires 
*but does not cancel the event* and then the document sees the event as normal 
and brings up the alert the second time. Prove this to yourself by removing the 
document.capture and the alert only comes up once.

ask for the keycode, I believe that is correct behavior, but I'll double check

Comment 14

[saari (gone)]

Okay, well, this apparently works in 4.x. Tom, can you comment on how we should 
fix this?

Comment 15

[Heikki Toivonen (remove -bugzilla when emailing directly)]

When working on firing the select event, I found that if I have the caret in the 
beginning of the INPUT text field and I press shift+end, I get two keypress 
events. Also if I have the caret at the end and press shift+home I get two 
keypress events. Because of this we get two select events.

Comment 16

[saari (gone)]

Target Mozilla 1.0

Comment 17

[joki (gone)]

This seems to be a bug with the captureEvents method, a deprecated 4.x method. 
In order to most closely model 4.x behavior, the captureEvents method should
translate the event listener registered on the document into a capture-only
listener so that it is not triggered a second time in the bubbling phase.  This
doesn't seem to be working and the listener is triggered twice.

As for the keycodes I believe they have all been fixed.  The last comment on the
bug about them is pretty old and they've worked for a while now.

Changing owner, summary and milestone.

Comment 18

[piers]

Is it this bug which is causing accesskeys to seemingly cause a function to
occur twice. This is a fairly major accessability bug. If this is a different
bug, i'll file a new report:

To reproduce:
1. Open Edit -> Preferences -> Navigator -> Languages
2. Add a few languages (say 5 to make sure the list is long enough)
3. Select a language and press Alt+U and Alt+D (to move the item up and down)

Actual results: the first time after adding, the accesskey works as expected,
subsequently, the accesskey causes the item to move two places up or down.

Expected: just move one place up or down.

(Note, this seems to occur with _every_ accesskey)

Comment 19

[Heikki Toivonen (remove -bugzilla when emailing directly)]

Aaron, this seems to affect accessibility (see comment above).

Comment 20

[Neil Deakin]

captureEvents no longer does anything