1137151 - Fix public destructors on reference-counted classes

Status: RESOLVED FIXED

(Firefox Build System :: General, defect)

Description

[:gerard-majax]

Let's track all of the instances of this issue. This seems not to be visible because it needs building with B2G_RIL with a compiler recent enough. That is happening in the context of bug 1038606

Comment 1

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Trying...

  https://treeherder.mozilla.org/#/jobs?repo=try&revision=81af1d16e95e

I build this patch set successfully on nexus-5-l, flatfish, unagi, and pandaboard.

Comment 2

[Thomas Zimmermann [:tzimmermann] [:tdz]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=777aaec90637

Comment 3

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [01] Bug 1137151: Marked destructor of |android::MediaCodecReader| as protected

Comment 4

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [02] Bug 1137151: Remove ref-counting from |OMXVideoEncoder|

Comment 5

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [03] Bug 1137151: Marked destructor of |MuxerOperation| as protected

Comment 6

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [04] Bug 1137151: Marked destructor of |MozNDEFRecord| as protected

Comment 7

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [05] Bug 1137151: Marked destructors of refcounted FM-radio classes as protected

Comment 8

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [06] Bug 1137151: Marked destructors of ref-counted RTSP classes as protected

Comment 9

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [07] Bug 1137151: Mark destructor of |STSThreadPoolListener| as protected

Comment 10

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [08] Bug 1137151: Marked destructors of ref-counted auto-mounter classes as protected

Comment 11

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [09] Bug 1137151: Marked destructors of ref-counted MTP classes as protected

Comment 12

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [10] Bug 1137151: Marked destructors of ref-counted time-zone classes as protected

Comment 13

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [11] Bug 1137151: Marked destructors of ref-counted audio-manager classes as protected

Comment 14

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [12] Bug 1137151: Marked destructors of ref-counted GonkHAL classes as protected

Comment 15

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [13] Bug 1137151: Marked destructor of |MemoryPressureWatcher| as protected

Comment 16

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [14] Bug 1137151: Marked destructor of |GeckoTouchDispatcher| as protected

Comment 17

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [15] Bug 1137151: Marked destructor of |nsClipboard| as protected

Comment 18

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [16] Bug 1137151: Marked destructor of |nsWindow| as protected

Comment 19

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [17] Bug 1137151: Marked destructor of |nsScreenManagerGonk| as protected

Comment 20

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [18] Bug 1137151: Marked destructor of |GrallocReporter| as protected

Comment 21

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [19] Bug 1137151: Marked destructors of ref-counted Bluetooth classes as protected

Comment 22

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [20] Bug 1137151: Enable test for non-public ref-counted destructors on gcc 4.8 and later

Comment 23

[Dave Hylands [:dhylands]]

Comment on attachment 8571316 [details] [diff] [review]
[12] Bug 1137151: Marked destructors of ref-counted GonkHAL classes as protected

Review of attachment 8571316 [details] [diff] [review]:
-----------------------------------------------------------------

Looks good to me.

Comment 24

[Shawn Huang [:shawnjohnjr][:shuang] (as a happy gecko contributor)]

Comment on attachment 8571324 [details] [diff] [review]
[19] Bug 1137151: Marked destructors of ref-counted Bluetooth classes as protected

Review of attachment 8571324 [details] [diff] [review]:
-----------------------------------------------------------------

r=me, thanks for fixing this bug.

Comment 25

[Michael Wu [:mwu]]

Comment on attachment 8571317 [details] [diff] [review]
[13] Bug 1137151: Marked destructor of |MemoryPressureWatcher| as protected

Review of attachment 8571317 [details] [diff] [review]:
-----------------------------------------------------------------

Redirecting review to someone who probably knows this code better than I do. (judging by the commit log)

Comment 26

[Nathan Froyd [:froydnj]]

Comment on attachment 8571413 [details] [diff] [review]
[20] Bug 1137151: Enable test for non-public ref-counted destructors on gcc 4.8 and later

Review of attachment 8571413 [details] [diff] [review]:
-----------------------------------------------------------------

::: xpcom/glue/nsISupportsImpl.h
@@ +49,5 @@
>  #  if MOZ_USING_LIBSTDCXX && MOZ_GCC_VERSION_AT_LEAST(4, 8, 0)
>  #    define MOZ_HAVE_STD_IS_DESTRUCTIBLE
>     // Some GCC versions have an ICE when using destructors in decltype().
> +   // Works on GCC 4.8 at least.
> +#  elif MOZ_GCC_VERSION_AT_LEAST(4, 8, 0)

Out of curiosity, is that GCC 4.8.0 as provided by a distro (presumably with distro-specific patches applied) or GCC 4.8.0 straight from the release tarball?  And have you tested 4.8.1?  I couldn't really find commentary on why this is 4.8.2, but presumably somebody tested it with GCC 4.8.0 and/or 4.8.1 and found problems...

Comment 27

[Thomas Zimmermann [:tzimmermann] [:tdz]]

(In reply to Nathan Froyd [:froydnj] [:nfroyd] from comment #26)
> 
> Out of curiosity, is that GCC 4.8.0 as provided by a distro (presumably with
> distro-specific patches applied) or GCC 4.8.0 straight from the release
> tarball?  And have you tested 4.8.1?  I couldn't really find commentary on
> why this is 4.8.2, but presumably somebody tested it with GCC 4.8.0 and/or
> 4.8.1 and found problems...

Hmm, the original comment sounded as if someone tried with gcc 4.8.2 and found it actually working; but gcc 4.8.1 or 4.8.0 was not available for testing.

The gcc 4.8.0, which I'm using comes prebuilt from the CAF repositories [1] we use for B2G. I haven't tried with gcc 4.8.1 or later. I don't know if there's a prebuilt binary somewhere and building one seems too much work for this simple bug.

[1] https://www.codeaurora.org/cgit/quic/la/platform/prebuilts/gcc/linux-x86/arm/arm-linux-androideabi-4.8/

Comment 28

[Nathan Froyd [:froydnj]]

(In reply to Thomas Zimmermann [:tzimmermann] [:tdz] from comment #27)
> (In reply to Nathan Froyd [:froydnj] [:nfroyd] from comment #26)
> > 
> > Out of curiosity, is that GCC 4.8.0 as provided by a distro (presumably with
> > distro-specific patches applied) or GCC 4.8.0 straight from the release
> > tarball?  And have you tested 4.8.1?  I couldn't really find commentary on
> > why this is 4.8.2, but presumably somebody tested it with GCC 4.8.0 and/or
> > 4.8.1 and found problems...
> 
> Hmm, the original comment sounded as if someone tried with gcc 4.8.2 and
> found it actually working; but gcc 4.8.1 or 4.8.0 was not available for
> testing.

Hm, I assumed that it didn't work on 4.8, given that the 4.8.2 bit was #else'd against a 4.8.0 conditional.  But you have a point, somebody might have just been conservative.

> The gcc 4.8.0, which I'm using comes prebuilt from the CAF repositories [1]
> we use for B2G. I haven't tried with gcc 4.8.1 or later. I don't know if
> there's a prebuilt binary somewhere and building one seems too much work for
> this simple bug.

Sure, toolchain building to resolve something like this would be silly.

Well, somebody will complain if this goes badly wrong, so let's try it.

Comment 29

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [04] Bug 1137151: Marked destructor of |MozNDEFRecord| as protected (v2)

Changes since v1:

  - rebased onto latest m-c

Comment 30

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Attachment: [14] Bug 1137151: Marked destructor of |GeckoTouchDispatcher| as protected (v2)

Changes since v1:

  - rebased onto latest m-c

Comment 31

[Thomas Zimmermann [:tzimmermann] [:tdz]]

Thanks everyone!

https://hg.mozilla.org/integration/b2g-inbound/rev/3d493b734419
https://hg.mozilla.org/integration/b2g-inbound/rev/4e473e4665b3
https://hg.mozilla.org/integration/b2g-inbound/rev/f9692c6fa585
https://hg.mozilla.org/integration/b2g-inbound/rev/5f11e6148aad
https://hg.mozilla.org/integration/b2g-inbound/rev/5bec2dc6ceac
https://hg.mozilla.org/integration/b2g-inbound/rev/e0687d17e8f8
https://hg.mozilla.org/integration/b2g-inbound/rev/6650bbc62372
https://hg.mozilla.org/integration/b2g-inbound/rev/191131b6b165
https://hg.mozilla.org/integration/b2g-inbound/rev/07597ca96575
https://hg.mozilla.org/integration/b2g-inbound/rev/9d6eb9ee78c3
https://hg.mozilla.org/integration/b2g-inbound/rev/34a622f3e1db
https://hg.mozilla.org/integration/b2g-inbound/rev/c95d2c40c94c
https://hg.mozilla.org/integration/b2g-inbound/rev/854abb194184
https://hg.mozilla.org/integration/b2g-inbound/rev/12d1a1d61c52
https://hg.mozilla.org/integration/b2g-inbound/rev/b61c178f6c55
https://hg.mozilla.org/integration/b2g-inbound/rev/ee68d8a9edcb
https://hg.mozilla.org/integration/b2g-inbound/rev/10b2ee690fab
https://hg.mozilla.org/integration/b2g-inbound/rev/da8b000c08c6
https://hg.mozilla.org/integration/b2g-inbound/rev/d645da56de29
https://hg.mozilla.org/integration/b2g-inbound/rev/d5e5fcb29452

https://treeherder.mozilla.org/#/jobs?repo=b2g-inbound&revision=d5e5fcb29452

Comment 32

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/3d493b734419
https://hg.mozilla.org/mozilla-central/rev/4e473e4665b3
https://hg.mozilla.org/mozilla-central/rev/f9692c6fa585
https://hg.mozilla.org/mozilla-central/rev/5f11e6148aad
https://hg.mozilla.org/mozilla-central/rev/5bec2dc6ceac
https://hg.mozilla.org/mozilla-central/rev/e0687d17e8f8
https://hg.mozilla.org/mozilla-central/rev/6650bbc62372
https://hg.mozilla.org/mozilla-central/rev/191131b6b165
https://hg.mozilla.org/mozilla-central/rev/07597ca96575
https://hg.mozilla.org/mozilla-central/rev/9d6eb9ee78c3
https://hg.mozilla.org/mozilla-central/rev/34a622f3e1db
https://hg.mozilla.org/mozilla-central/rev/c95d2c40c94c
https://hg.mozilla.org/mozilla-central/rev/854abb194184
https://hg.mozilla.org/mozilla-central/rev/12d1a1d61c52
https://hg.mozilla.org/mozilla-central/rev/b61c178f6c55
https://hg.mozilla.org/mozilla-central/rev/ee68d8a9edcb
https://hg.mozilla.org/mozilla-central/rev/10b2ee690fab
https://hg.mozilla.org/mozilla-central/rev/da8b000c08c6
https://hg.mozilla.org/mozilla-central/rev/d645da56de29
https://hg.mozilla.org/mozilla-central/rev/d5e5fcb29452

Comment 33

[Naoki Hirata :nhirata (please use needinfo instead of cc)]

This seemed to have broken bluetooth functionality : bug 1142132.

Comment 34

[bhavana bajaj [:bajaj]]

Backed out the bluetooth related cset in https://hg.mozilla.org/mozilla-central/rev/c33922ee3ac3 due to a smoketest bustage : https://bugzilla.mozilla.org/show_bug.cgi?id=1142132#c9

Comment 35

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

Relanded that piece in https://hg.mozilla.org/mozilla-central/rev/e1262bdb5b14 because it broke b2g emulator builds: 

https://treeherder.mozilla.org/logviewer.html#?job_id=1150744&repo=mozilla-central

I guess this re-resolves this bug, though something will need to be done to fix bug 1142132.

Comment 36

[Daniel Holbert [:dholbert]]

If you're backing out any piece of this (which means you're adding back some public destructors), I suspect you *also* need to backout part 20 (which is the part that makes public destructors fatal, on the GCC version that we use on our builders).

Comment 37

[Daniel Holbert [:dholbert]]

So, it's expected that comment 34 would've broken builds -- it needed to also backout part 20 (mozilla-central changeset d5e5fcb29452) in order to still build successfully.

(Having said that -- I don't immediately see how the bluetooth-related cset here (c33922ee3ac3) could actually affect behavior. I did find one refcounting bug in related code, from code-inspection of BluetoothHALInterfaceRunnable0-related-code (the code that's crashing in the regression here), and I filed bug 1142364 on it. It's remotely-possible that that bug's really what's responsible for the crash, and this just exposed it somehow.)

Comment 38

[Thomas Zimmermann [:tzimmermann] [:tdz]]

I take a look at bug 1142132.