707648 - make check fails on 10.6 with clang (Assertion failure: OnLionOrLater() || gCriticalAddress.mAddr != NULL)

Status: RESOLVED FIXED

(Firefox Build System :: General, defect)

Description

[Mark Banner (:standard8)]

I'm trying to run make check in debug mode on Mac 10.6 with clang, I see this:

Assertion failure: OnLionOrLater() || gCriticalAddress.mAddr != NULL

If I run it in compiled with gcc I get no assertions. Commenting out the line on clang runs successfully.

Comment 1

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

This blocks a switch to clang, not enabling leak tests on gcc built binaries.

I will debug this.

Comment 2

[Mark Banner (:standard8)]

Attachment: Mozconfig

This was the mozconfig I used for compiling TB.

Comment 3

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

I have not been able to reproduce this on m-c. What repository are you using? what revision?

Comment 4

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

Sorry, I am having a hard time reproducing this. I have built the same version of clang on 10.6 and tried you exact mozconfig on comm-central (cd77f678aa1c).

The only failures I got from make check were from bug529130.js complaining about too much recursion, which I believe is because clang -O0 produces binaries that use a lot of stack space.

Comment 5

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

Attachment: Only try to init the critical range if we are able to walk the stack

This fixes the case of a 64 bit OS X without unwind.h.

Comment 6

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

https://tbpl.mozilla.org/?tree=Try&rev=8fd905218656

Comment 7

[(no longer active)]

Comment on attachment 579197 [details] [diff] [review]
Only try to init the critical range if we are able to walk the stack

This patch does not fix the crash for me.  The real issue, as far as I can tell, is that the set of conditions for this code to be compiled is different with the set of conditions for a useful implementation of NS_StackWalk to be compiled.

Comment 8

[(no longer active)]

Attachment: My mozilla-config.h

Comment 9

[(no longer active)]

Attachment: My nsStackWalk.i

Comment 10

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

Note that  OnLionOrLater is not on the .ii file. If you are still seeing an assert with it, we are missing a dependency somewhere.

Comment 11

[(no longer active)]

Comment on attachment 579197 [details] [diff] [review]
Only try to init the critical range if we are able to walk the stack

I don't know what was happening last time that I tried this, but this is definitely fixing the crashes.  So, r=me.

Comment 12

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

https://tbpl.mozilla.org/?tree=Mozilla-Inbound&rev=d980f5b9f338

Comment 13

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/d980f5b9f338

Comment 14

[Mark Banner (:standard8)]

Comment on attachment 579197 [details] [diff] [review]
Only try to init the critical range if we are able to walk the stack

This is now biting me on the mozilla-esr10 branch, can we backport it there as well?

The fix has been shipped in 11 already.

Comment 15

[Lukas Blakk [:lsblakk] use ?needinfo]

Comment on attachment 579197 [details] [diff] [review]
Only try to init the critical range if we are able to walk the stack

[triage comment]
esr branch is strictly production and afaik no one should be doing clang builds here, please re-nom if there's a case for how this affects the user or production builds.