Smalltalk › Squeak › Squeak - Dev

compiler bug with 4.5?

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

4 messages Options

Eliot Miranda-2

compiler bug with 4.5?

Hi All,

just found a worrying bug in 4.5. that may be a bytecode compiler bug or a Monticello bug. It's completely repeatable.

The bug is that after building the VMMaker image that compiles the VMMaker package, Spur32BitCoMemoryManager>>#getScavengeThreshold (the 42st inst var) is compiled with a header that says return the 7th inst var.

To repeat, check-out http://www.squeakvm.org/svn/squeak/branches/Cog/image, run buildsqueak45image.sh, then start-up CogVMMaker.image. Navigate to Spur32BitCoMemoryManager>>#getScavengeThreshold
-- and explore the method (there's a menu pick in the browser).

best,

Eliot

Eliot Miranda-2

Re: compiler bug with 4.5?

On reflection this is unlikely to be anything to do with the bytecode compiler. It is most likely an issue with the binary package and the bogus methods (for, dear reader, there are at least a few more) could have lain there-in undetected for some time. I'll look more temperately soon.

Eliot (phone)

David T. Lewis

Re: compiler bug with 4.5?

In reply to this post by Eliot Miranda-2

On Thu, May 01, 2014 at 04:47:09PM -0700, Eliot Miranda wrote:

> Hi All,
>
> just found a worrying bug in 4.5. that may be a bytecode compiler bug
> or a Monticello bug. It's completely repeatable.
>
> The bug is that after building the VMMaker image that compiles the VMMaker
> package, Spur32BitCoMemoryManager>>#getScavengeThreshold (the 42st inst
> var) is compiled with a header that says return the 7th inst var.
>
> To repeat, check-out http://www.squeakvm.org/svn/squeak/branches/Cog/image,
> run buildsqueak45image.sh, then start-up CogVMMaker.image. Navigate to
> Spur32BitCoMemoryManager>>#getScavengeThreshold
> -- and explore the method (there's a menu pick in the browser).
> best,
> Eliot

Something strange is going on for sure. I have tried running the
"$VM CogVMMaker.image BuildSqueak45Image.st" step a few times with both
a Cog VM and an interpreter VM, and in all cases the image and VM are just
exiting somewhere in the middle of the process. No error messages, no core
dumps, just an exit. I am not seeing any consistent pattern as to when it is
exiting although I think it is happening during Monticello processing. I am
on a 64 bit linux box (BTW the "md5" command used in the shell scripts is
not present on my machine).

Dave

tty

Re: compiler bug with 4.5?

I can confirm Dave's random exits.

I am porting the buildsqueak*.sh scripts into a "cmakify.sh" script for my CMake work

and have experienced random "poof! its gone" moments too.

I had to ditch the md5 checks as well.

That said, the thing usually runs and runs to completion.

cordially,

tty