Smalltalk › Squeak › Squeak VM

Appveyor red again

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

5 messages Options

Nicolas Cellier

Appveyor red again

For a few builds, appveyor is failing for squeak.cog.v3 at final link stage:

i686-w64-mingw32-gcc -mwindows -m32 -mthreads -Wl,--large-address-aware,--export-all-symbols -o build/vm/SqueakUnstripped.exe \

...snip...

collect2: fatal error: ld terminated with signal 11 [Segmentation fault], core dumped

https://ci.appveyor.com/project/OpenSmalltalk/vm/history shows it happened since
https://ci.appveyor.com/project/OpenSmalltalk/vm/build/1.0.710

The same build pass on linux and osx.

The tools shall never crash, bu worse problem I have is that I can't reproduce the segfault in my windows VM, even if updating properly cygwin...

Any idea of what's going on?

K K Subbu

Re: Appveyor red again

On Thursday 15 June 2017 03:48 AM, Nicolas Cellier wrote:
>
> i686-w64-mingw32-gcc -mwindows -m32 -mthreads
> -Wl,--large-address-aware,--export-all-symbols -o
> build/vm/SqueakUnstripped.exe \
> ...snip...
> collect2: fatal error: ld terminated with signal 11 [Segmentation
> fault], core dumped

Looks like a toolchain bug to me. Collect2 is a utility that creates the
init glue code before call to main(). ld tools work with the object
files created by previous passes. Failing with a seg fault is a symptom
of toolchain defect, possibly triggered by size or number of files being
processed.

You may want to try a different (older?) version of gcc toolchain.

More info on collect2 at https://gcc.gnu.org/onlinedocs/gccint/Collect2.html

Regards .. Subbu

Nicolas Cellier

Re: Appveyor red again

2017-06-15 8:35 GMT+02:00 K K Subbu <[hidden email]>:

On Thursday 15 June 2017 03:48 AM, Nicolas Cellier wrote:

i686-w64-mingw32-gcc -mwindows -m32 -mthreads
-Wl,--large-address-aware,--export-all-symbols -o
build/vm/SqueakUnstripped.exe \
...snip...
collect2: fatal error: ld terminated with signal 11 [Segmentation
fault], core dumped

Looks like a toolchain bug to me. Collect2 is a utility that creates the init glue code before call to main(). ld tools work with the object files created by previous passes. Failing with a seg fault is a symptom of toolchain defect, possibly triggered by size or number of files being processed.

You may want to try a different (older?) version of gcc toolchain.

More info on collect2 at https://gcc.gnu.org/onlinedocs/gccint/Collect2.html

Regards .. Subbu

Yes,
what is troubling is that previous build did succeed with no less .o and .lib files, and that other flavours using the exact same toolchain are succeeding...

It has to be something specific to v3, but might as well be triggered by a cygwin upgrade...

That's a problem with cygwin, because we can't version the tools easily.

If we respect coding standards, the tools should not matter that much in theory, but that's the subtle difference with practice ;)

Holger Freyther

Re: Appveyor red again

In reply to this post by Nicolas Cellier

> On 15. Jun 2017, at 06:18, Nicolas Cellier <[hidden email]> wrote:
>

Hi!

> The same build pass on linux and osx.
> The tools shall never crash, bu worse problem I have is that I can't reproduce the segfault in my windows VM, even if updating properly cygwin...
> Any idea of what's going on?

no specific idea here but I once debugged a pharo-vm issue that seems to exist because of mixing cygwin and mingw. What I found helpful was to use the RDP support of Appveyor to access the build VM. In the end I just installed the version of gcc that didn't crash when passed --version.

Nicolas Cellier

Re: Appveyor red again

2017-06-16 9:00 GMT+02:00 Holger Freyther <[hidden email]>:

> On 15. Jun 2017, at 06:18, Nicolas Cellier <[hidden email]> wrote:
>

Hi!

> The same build pass on linux and osx.
> The tools shall never crash, bu worse problem I have is that I can't reproduce the segfault in my windows VM, even if updating properly cygwin...
> Any idea of what's going on?

no specific idea here but I once debugged a pharo-vm issue that seems to exist because of mixing cygwin and mingw. What I found helpful was to use the RDP support of Appveyor to access the build VM. In the end I just installed the version of gcc that didn't crash when passed --version.

Thanks Holger

That's something to try.