Created attachment 608345 [details]
User Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:11.0) Gecko/20100101 Firefox/11.0
Build ID: 20120312181643
Steps to reproduce:
Debugged the attached testcase-1.
1. open Firebug, enable the script tab and load the attached testcase-1
2. debugger stops at breakpoint (A), (reload the site if the debugger don't stops)
3. click step over
4. now the cursor is at line (B), click step over again
5. now, hover the sourcecode in the firebug window at the bottom
Firefox 11.0 + Firebug 1.9.1
Firefox nightly 14.0a1 (2012-03-22) + Firebug 10.0.a5
Crash ID: bp-1de8762a-de1f-4850-976d-afc0e2120322
Luke, this is crashing on CrashIfInvalidSlot called from
any idea offhand what that means?
Ooh yeah, this may be STR for bug 723445! I'll investigate.
Wow, this is a most excellent test case; thanks! I can reproduce locally, I'll look into it hopefully tomorrow.
Created attachment 612785 [details] [diff] [review]
fix forced return
Ah, I see the bug. The problem is that forced_return (jumped to by the debugger) is setting the stack depth to 0, but leaving pc pointing to the JSOP_CALL which the debugger interrupted. Thus, when the ScriptDebugEpilogue runs, js_ReconstructStackDepth(pc) is not equal to the actual stack depth and we go touching invalid memory (the next frame).
Thank you again Attila for taking the time to create a reproducible test-case. It was certainly invaluable! We've been trying to track down reports of this crash for a while and between this and bug 723445, I bet we have it fixed.
I forgot, I'll make a shell test-case tomorrow.
Created attachment 612912 [details] [diff] [review]
fix forced return
This patch adds a shell testcase. In finding a testcase, I can see why this was relatively hard to hit: you have to trap on a 'call' instruction, force an early return, then, in the leave-frame hook for that function, you need to runs some code that inspects the stack via StackIter. Phew.