Closed Bug 761861 Opened 11 years ago Closed 11 years ago

Crash with contentEditable, insertParagraph, forwardDelete

Categories

(Core :: DOM: Editor, defect)

Product:
Core
Component:
DOM: Editor
Version:
15 Branch
Type:
defect
Priority:
Not set
Severity:
critical

Tracking

()

Status:
VERIFIED FIXED
Milestone:
mozilla16
Tracking Flags:
Tracking Status
firefox15 + verified
firefox16 + verified

People

(Reporter: jruderman, Assigned: ayg)

References

Details

(Keywords: crash, regression, testcase)

Crash Data

Attachments

(3 files, 2 obsolete files)

testcase (crashes Firefox when loaded)
417 bytes, text/html
Details
stack trace
12.77 KB, text/plain
Details
Patch v3, with better test
2.34 KB, patch
ehsan.akhgari
: review+
akeybl
: approval-mozilla-aurora+
Details | Diff | Splinter Review 
No description provided.
On Windows: bp-8ef6b204-d35a-4653-8332-ddd4e2120606.
Crash Signature: [@ nsHTMLEditor::DeleteSelectionImpl] → [@ nsHTMLEditor::DeleteSelectionImpl] [@ nsINode::GetParent()]
OS: Mac OS X → All
Hardware: x86_64 → All
Attached patch Possible patch, untested (obsolete) — — Splinter Review 
I can see what the problem should be, and this patch should fix it.  But I can't reproduce the crash on a debug build I compile myself at r95861.  I only see it in a nightly.  Can anyone confirm that the problem still exists on latest trunk, or failing that, explain where it does exist?  (We need a fix to backport to aurora, at the very least.)
Okay, I can reproduce in an opt build only.  Not sure why.
Assignee: nobody → ayg
Status: NEW → ASSIGNED
Attached patch Patch v2, with test (obsolete) — — Splinter Review 
Try: https://tbpl.mozilla.org/?tree=Try&rev=d5d4c9ef038a

This also demonstrated something really weird -- insertParagraph doesn't ignore the last argument.  Something is pretty badly wrong with my implementation if that's the case!
Attachment #630480 - Attachment is obsolete: true
Attachment #630496 - Flags: review?(ehsan)
Comment on attachment 630496 [details] [diff] [review]
Patch v2, with test

[Approval Request Comment]
Bug caused by (feature/regressing bug #): Bug 590640, patch part 5.
User impact if declined: Easily-triggered crash, although AFAICT the conditions are relatively unlikely to occur in non-malicious pages (editable root element with non-block children).
Testing completed (on m-c, etc.): None yet; requesting approval in parallel with review request.  See try run link in previous comment.
Risk to taking this patch (and alternatives if risky): Minimal -- simple null check.  Backing out bug 590640 patch 5 would also fix the issue, but would involve much larger changes.  That patch changes over 200 LOC by itself (not counting tests), and backing it out would conflict with a lot of subsequent patches.
String or UUID changes made by this patch: None.
Attachment #630496 - Flags: approval-mozilla-aurora?
This is really caused by bug 590640, not bug 748307.  Bug 590640 part 5 added the problematic code.  I'm pretty sure it's possible to trigger this solely with commands that existed before bug 748307 landed -- construct the DOM yourself instead of using insertParagraph, and use delete instead of forwardDelete.
Blocks: 590640
No longer blocks: 748307
Keywords: regression
Version: Trunk → 15 Branch
[Approval Request Comment] See comment 6.  This is the same patch with a different test.

The new test doesn't rely on the fact that insertParagraph incorrectly works the same as insertText when execCommand() is given a third parameter.  It uses only execCommand("forwarddelete").  That's not needed either; you can do it without execCommand() at all.  Just load this URL, put the cursor before the underscore, and hit the delete key to delete the next character:

data:text/html,<!doctype html>
<script>
function boom() {
  var r = document.documentElement;
  while (r.firstChild) {
    r.removeChild(r.firstChild);
  }

  document.documentElement.contentEditable = "true";
  document.documentElement.appendChild(document.createElement("span"));
  document.documentElement.firstChild.appendChild(document.createTextNode("_"));
}
</script>
<body onload="boom()">

Thus bug 748307 isn't needed for the crash at all (but it makes testing easier).

New try run: https://tbpl.mozilla.org/?tree=Try&rev=1427204ad017

It's worth wondering why forward and backward delete are going down such different codepaths here.  I'd have expected both to trigger the bug.
Attachment #630496 - Attachment is obsolete: true
Attachment #630496 - Flags: review?(ehsan)
Attachment #630496 - Flags: approval-mozilla-aurora?
Attachment #630503 - Flags: review?(ehsan)
Attachment #630503 - Flags: approval-mozilla-aurora?
Attachment #630503 - Flags: review?(ehsan) → review+
https://hg.mozilla.org/mozilla-central/rev/c4da1c7555ee
Status: ASSIGNED → RESOLVED
Closed: 11 years ago
Resolution: --- → FIXED
Flags: in-testsuite+
Target Milestone: --- → mozilla16
Comment on attachment 630503 [details] [diff] [review]
Patch v3, with better test

[Triage Comment]
Simple null check, approved for Aurora 15.
Attachment #630503 - Flags: approval-mozilla-aurora? → approval-mozilla-aurora+
No crash loading the testcase.
Verified fixed on FF 15b3 on Win 7/64, Ubuntu 12.04 and Mac OS X 10.6.
status-firefox15: fixed → verified
Verified fixed on FF 16b1 on Win 7/64, Ubuntu 12.04 and Mac OS X 10.7.4.
Status: RESOLVED → VERIFIED
status-firefox16: fixed → verified
QA Contact: paul.silaghi
You need to log in before you can comment on or make changes to this bug.