1270832 - turn on debug mode for libstdc++ headers

Status: REOPENED

(Firefox Build System :: General, task)

Description

[Nathan Froyd [:froydnj]]

STL usages have crept into the tree and will continue to creep in, especially
as we move to a C++11 STL implementation everywhere.  Let's at least enable a
bit of checking for those data structures so we're not completely shooting
ourselves in the foot.

Comment 1

[Nathan Froyd [:froydnj]]

Attachment: turn on debug mode for libstdc++ headers

This patch seems to work; the FIXME appears to be no longer applicable with
newer GCC:

https://treeherder.mozilla.org/#/jobs?repo=try&revision=7d9eb75e3d99

Comment 2

[Mike Hommey [:glandium]]

Comment on attachment 8749640 [details] [diff] [review]
turn on debug mode for libstdc++ headers

Review of attachment 8749640 [details] [diff] [review]:
-----------------------------------------------------------------

::: config/gcc-stl-wrapper.template.h
@@ -45,5 @@
>  
>  #if defined(DEBUG) && !defined(_GLIBCXX_DEBUG)
>  // Enable checked iterators and other goodies
> -//
> -// FIXME/bug 551254: gcc's debug STL implementation requires -frtti.

Is this comment outdated?

Comment 3

[Nathan Froyd [:froydnj]]

(In reply to Mike Hommey [:glandium] from comment #2)
> Comment on attachment 8749640 [details] [diff] [review]
> turn on debug mode for libstdc++ headers
> 
> Review of attachment 8749640 [details] [diff] [review]:
> -----------------------------------------------------------------
> 
> ::: config/gcc-stl-wrapper.template.h
> @@ -45,5 @@
> >  
> >  #if defined(DEBUG) && !defined(_GLIBCXX_DEBUG)
> >  // Enable checked iterators and other goodies
> > -//
> > -// FIXME/bug 551254: gcc's debug STL implementation requires -frtti.
> 
> Is this comment outdated?

This comment appears to stem from bug 551254 comment 30, but the libstdc++ debug mode documentation makes no mention of the -frtti constraint:

https://gcc.gnu.org/onlinedocs/libstdc++/manual/debug_mode_using.html#debug_mode.using.mode

The only instances of |typeid| I can find in the debug headers are all guarded by preprocessor checks that RTTI is enabled.  Spelunking in libstdc++ leads me to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=40160 , which would have been fixed somewhere in the GCC 4.5 timeframe.  So yes, I think this comment is outdated.  OK to commit, then?

Comment 4

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/fe3f7921e2dc

Comment 5

[Pulsebot]

Backout:
https://hg.mozilla.org/integration/mozilla-inbound/rev/1d918e13194c

Comment 6

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/ccff1c4580ab

Comment 7

[Nathan Froyd [:froydnj]]

(In reply to Pulsebot from comment #6)
> https://hg.mozilla.org/integration/mozilla-inbound/rev/ccff1c4580ab

This revision landed with a change to ignore debug headers for static analysis.  We're not running the static analysis build, so I figure that not having the debug-ness turned on here is not a bad thing.

Comment 8

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/ccff1c4580ab

Comment 9

[Mike Hommey [:glandium]]

Backed out in
https://hg.mozilla.org/integration/mozilla-inbound/rev/a640e6fa8ab9

Comment 10

[Nathan Froyd [:froydnj]]

The fuzzing team was interested in this; they may also be interested in comment 9, and the commit message for the backout.

Comment 11

[Release mgmt bot [:suhaib / :marco/ :calixte]]

The bug assignee didn't login in Bugzilla in the last 7 months, so the assignee is being reset.

Comment 12

[Tom Ritter [:tjr]]

This was talked about recently in the Google-organized Memory Saftey Summit. They enabled these on Chromium and saw over 800 unique issues in 3 months so it definitely seems worth running down.

Comment 13

[Christian Holler (:decoder)]

For easier reading, here is the backout reason from comment 9:

It turns out that, since we're including <new> before setting
_GLIBCXX_DEBUG, the debug parts of c++config.h are not activated, and
that has an impact of how much of the debug features of the STL are
activated.

In contrast, the upcoming changes to the STL wrappers are avoiding the
include of <new> before _GLIBCXX_DEBUG is set, which in turn breaks the
build because, as we link things that use STL wrappers with things that
don't, they end up with a different state of STL debugging, and have
mismatching symbols.

Mike, do you know if this is still an issue? Are there any other blockers for pursuing this?

Comment 14

[Mike Hommey [:glandium]]

I think it still is, short of building some things twice (e.g. some parts of the updater). Best way to tell for sure would be to do a try push.