code review 142470043: cmd/go: strip -fsanitize= flags when building cgo object

cmd/go: strip -fsanitize= flags when building cgo object
Fixes issue 8788.

[dvyukov, 2014-09-23 04:05:25]

Hello iant@golang.org (cc: golang-codereviews@googlegroups.com),

I'd like you to review this change to
https://dvyukov%40google.com@code.google.com/p/go/

[iant, 2014-09-23 13:33:40]

https://codereview.appspot.com/142470043/diff/80001/src/cmd/go/build.go
File src/cmd/go/build.go (right):

https://codereview.appspot.com/142470043/diff/80001/src/cmd/go/build.go#newco...
src/cmd/go/build.go:2340: // Remove any -fsanitize=foo flags.
This isn't what I was suggesting.  I was suggesting doing this in the loop above
that sets bareLDFLAGS.

[dvyukov, 2014-09-24 04:24:24]

Done
PTAL

[iant, 2014-09-24 13:24:37]

LGTM

[dvyukov, 2014-09-24 21:49:08]

*** Submitted as https://code.google.com/p/go/source/detail?r=d981aaf528df ***

cmd/go: strip -fsanitize= flags when building cgo object
Fixes issue 8788.

LGTM=iant
R=iant
CC=golang-codereviews
https://codereview.appspot.com/142470043

[minux, 2014-09-25 20:13:36]

i think this just hide the problem. gcc -Wl,-r will still be broken.

the santizer libraries probably should mark all symbols as weak.

[dvyukov, 2014-09-26 02:15:01]

You are welcome to join the discussion at:
https://code.google.com/p/address-sanitizer/issues/detail?id=344


On Thu, Sep 25, 2014 at 1:13 PM, minux <minux@golang.org> wrote:
> i think this just hide the problem. gcc -Wl,-r will still be broken.
>
> the santizer libraries probably should mark all symbols as weak.

[gobot, 2014-09-26 10:27:34]

This CL appears to have broken the windows-amd64-perf builder.
See http://build.golang.org/log/36d41024d8dcef519ef4c9cd6c07391ddbbedac5

[iant, 2014-09-26 19:58:08]

Wait, I am now puzzled.  In a discussion with Russ he pointed out that along
with -Wl,-r, we also pass -nostdlib.

Does this mean that the clang driver passes the sanitizer libraries to the
linker even if -nostdlib is being used?  Because that is clearly and obviously
wrong.  And if the clang driver did not do that, then we would not need this
code in the go tool.

[dvyukov, 2014-09-28 04:32:25]

On Fri, Sep 26, 2014 at 12:58 PM,  <iant@golang.org> wrote:
> Wait, I am now puzzled.  In a discussion with Russ he pointed out that
> along with -Wl,-r, we also pass -nostdlib.
>
> Does this mean that the clang driver passes the sanitizer libraries to
> the linker even if -nostdlib is being used?  Because that is clearly and
> obviously wrong.  And if the clang driver did not do that, then we would
> not need this code in the go tool.


This was just fixed in clang driver:
https://code.google.com/p/address-sanitizer/issues/detail?id=344#c14
But that will be deployed on majority of user machines in like 5
years, much later than 1.4.

[iant, 2014-09-28 14:59:01]

On Sat, Sep 27, 2014 at 9:32 PM, Dmitry Vyukov <dvyukov@google.com> wrote:
> On Fri, Sep 26, 2014 at 12:58 PM,  <iant@golang.org> wrote:
>> Wait, I am now puzzled.  In a discussion with Russ he pointed out that
>> along with -Wl,-r, we also pass -nostdlib.
>>
>> Does this mean that the clang driver passes the sanitizer libraries to
>> the linker even if -nostdlib is being used?  Because that is clearly and
>> obviously wrong.  And if the clang driver did not do that, then we would
>> not need this code in the go tool.
>
>
> This was just fixed in clang driver:
> https://code.google.com/p/address-sanitizer/issues/detail?id=344#c14
> But that will be deployed on majority of user machines in like 5
> years, much later than 1.4.

Understood.  But now we see that we are simply working around a bug,
not dealing with some complex interaction.

Ian