cute, we crash gcc...

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
13 messages Options
Reply | Threaded
Open this post in threaded view
|

cute, we crash gcc...

Tobias Pape
Reply | Threaded
Open this post in threaded view
|

Re: cute, we crash gcc...

Holger Freyther
 


> On 20. Mar 2018, at 11:54, Tobias Pape <[hidden email]> wrote:
>

Hey!


> https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355802605#L2458

indeed! I think it is this bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59477. I love how GCC developers add a regression test for every issue fixed and they do it for many many years already. And it is a codebase that is 30+ years old and started before SSA was "known".

holger

Reply | Threaded
Open this post in threaded view
|

Re: cute, we crash gcc...

Tobias Pape
 
On 20.03.2018, at 13:58, Holger Freyther <[hidden email]> wrote:

>
>
>
>
>> On 20. Mar 2018, at 11:54, Tobias Pape <[hidden email]> wrote:
>>
>
> Hey!
>
>
>> https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355802605#L2458
>
> indeed! I think it is this bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59477. I love how GCC developers add a regression test for every issue fixed and they do it for many many years already. And it is a codebase that is 30+ years old and started before SSA was "known".

Yes, upgrading GCC fixes this (see https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/builds/355826629 )

But the newspeak-bootstrap now fails.

@Eliot, can you have a look at this?
        https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355826630#L6518
I actually cannot fathom how a gcc upgrade should influence this, but I'll hold whit the Pull Request until you say that its either my fault or nothing to be afraid of.

Best regards
        -Tobias

>
> holger
>

Reply | Threaded
Open this post in threaded view
|

Re: cute, we crash gcc...

Eliot Miranda-2
 
Hi Tobias, Hi Ryan,

On Tue, Mar 20, 2018 at 6:12 AM, Tobias Pape <[hidden email]> wrote:

On 20.03.2018, at 13:58, Holger Freyther <[hidden email]> wrote:
>
>
>
>
>> On 20. Mar 2018, at 11:54, Tobias Pape <[hidden email]> wrote:
>>
>
> Hey!
>
>
>> https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355802605#L2458
>
> indeed! I think it is this bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59477. I love how GCC developers add a regression test for every issue fixed and they do it for many many years already. And it is a codebase that is 30+ years old and started before SSA was "known".

Yes, upgrading GCC fixes this (see https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/builds/355826629 )

But the newspeak-bootstrap now fails.

@Eliot, can you have a look at this?
        https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355826630#L6518
I actually cannot fathom how a gcc upgrade should influence this, but I'll hold whit the Pull Request until you say that its either my fault or nothing to be afraid of.

Well, if it's a result of my enabling read-only objects (-DIMMUTABILITY=1) for all Spur VMs then I'll turn it off for the Newspeak builds, but the key point here is that we should decouple the Newspeak builds from the Pharo and Squeak builds asap.  Newspeak shouldn't block new releases of what should be perfectly good VMs.  This is urgent IMO.

_,,,^..^,,,_
best, Eliot
Reply | Threaded
Open this post in threaded view
|

Re: cute, we crash gcc...

Eliot Miranda-2
In reply to this post by Tobias Pape
 
Hi Tobias,

On Tue, Mar 20, 2018 at 6:12 AM, Tobias Pape <[hidden email]> wrote:

On 20.03.2018, at 13:58, Holger Freyther <[hidden email]> wrote:
>
>
>
>
>> On 20. Mar 2018, at 11:54, Tobias Pape <[hidden email]> wrote:
>>
>
> Hey!
>
>
>> https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355802605#L2458
>
> indeed! I think it is this bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59477. I love how GCC developers add a regression test for every issue fixed and they do it for many many years already. And it is a codebase that is 30+ years old and started before SSA was "known".

Yes, upgrading GCC fixes this (see https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/builds/355826629 )

But the newspeak-bootstrap now fails.

@Eliot, can you have a look at this?
        https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355826630#L6518
I actually cannot fathom how a gcc upgrade should influence this, but I'll hold whit the Pull Request until you say that its either my fault or nothing to be afraid of.

Don't hold up the commit.  I forgot to ask.  Can you repeat, adding -DIMMUTABILITY=0 to the Newspeak VM build options in mvm?  If so, please LMK asap of the results.

_,,,^..^,,,_
best, Eliot
Reply | Threaded
Open this post in threaded view
|

Re: cute, we crash gcc...

Tobias Pape
In reply to this post by Eliot Miranda-2
 
Hi

> On 20.03.2018, at 16:27, Eliot Miranda <[hidden email]> wrote:
>
> Hi Tobias, Hi Ryan,
>
> On Tue, Mar 20, 2018 at 6:12 AM, Tobias Pape <[hidden email]> wrote:
>
> On 20.03.2018, at 13:58, Holger Freyther <[hidden email]> wrote:
>>
>>
>>
>>
>>> On 20. Mar 2018, at 11:54, Tobias Pape <[hidden email]> wrote:
>>>
>>
>> Hey!
>>
>>
>>> https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355802605#L2458
>>
>> indeed! I think it is this bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59477. I love how GCC developers add a regression test for every issue fixed and they do it for many many years already. And it is a codebase that is 30+ years old and started before SSA was "known".
>
> Yes, upgrading GCC fixes this (see https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/builds/355826629 )
>
> But the newspeak-bootstrap now fails.
>
> @Eliot, can you have a look at this?
>        https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355826630#L6518
> I actually cannot fathom how a gcc upgrade should influence this, but I'll hold whit the Pull Request until you say that its either my fault or nothing to be afraid of.
>
> Well, if it's a result of my enabling read-only objects (-DIMMUTABILITY=1) for all Spur VMs then I'll turn it off for the Newspeak builds, but the key point here is that we should decouple the Newspeak builds from the Pharo and Squeak builds asap.  Newspeak shouldn't block new releases of what should be perfectly good VMs.  This is urgent IMO.

Ok I'll see,
but two things:

First, it works for 32bit but not 64bit.

Second, the Newspeak build did not get the added -DIMMUTABILITY=1, so it is probably not that…

Any other suggestions?


Best regards
        -Tobias

>
> _,,,^..^,,,_
> best, Eliot

Reply | Threaded
Open this post in threaded view
|

Re: cute, we crash gcc...

Eliot Miranda-2
 
Hi Tobias,


> On Mar 20, 2018, at 12:33 PM, Tobias Pape <[hidden email]> wrote:
>
>
> Hi
>
>> On 20.03.2018, at 16:27, Eliot Miranda <[hidden email]> wrote:
>>
>> Hi Tobias, Hi Ryan,
>>
>> On Tue, Mar 20, 2018 at 6:12 AM, Tobias Pape <[hidden email]> wrote:
>>
>> On 20.03.2018, at 13:58, Holger Freyther <[hidden email]> wrote:
>>>
>>>
>>>
>>>
>>>> On 20. Mar 2018, at 11:54, Tobias Pape <[hidden email]> wrote:
>>>>
>>>
>>> Hey!
>>>
>>>
>>>> https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355802605#L2458
>>>
>>> indeed! I think it is this bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59477. I love how GCC developers add a regression test for every issue fixed and they do it for many many years already. And it is a codebase that is 30+ years old and started before SSA was "known".
>>
>> Yes, upgrading GCC fixes this (see https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/builds/355826629 )
>>
>> But the newspeak-bootstrap now fails.
>>
>> @Eliot, can you have a look at this?
>>       https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355826630#L6518
>> I actually cannot fathom how a gcc upgrade should influence this, but I'll hold whit the Pull Request until you say that its either my fault or nothing to be afraid of.
>>
>> Well, if it's a result of my enabling read-only objects (-DIMMUTABILITY=1) for all Spur VMs then I'll turn it off for the Newspeak builds, but the key point here is that we should decouple the Newspeak builds from the Pharo and Squeak builds asap.  Newspeak shouldn't block new releases of what should be perfectly good VMs.  This is urgent IMO.
>
> Ok I'll see,
> but two things:
>
> First, it works for 32bit but not 64bit.
>
> Second, the Newspeak build did not get the added -DIMMUTABILITY=1, so it is probably not that…

No, I changed the default from off to on in anticipation of the new Squeak vm release where I want support for read-only objects in the vm so we can have read-only literals in  trunk very soon.  See any *spur*src/vm/interp.h

So I think is very likely that.  If it is I shall make it the default only for Smalltalk, not for Newspeak.

>
> Any other suggestions?
>
>
> Best regards
>    -Tobias
>
>>
>> _,,,^..^,,,_
>> best, Eliot
>
Reply | Threaded
Open this post in threaded view
|

Re: cute, we crash gcc...

Ryan Macnak
In reply to this post by Eliot Miranda-2
 
On Tue, Mar 20, 2018 at 8:27 AM, Eliot Miranda <[hidden email]> wrote:
Hi Tobias, Hi Ryan,

On Tue, Mar 20, 2018 at 6:12 AM, Tobias Pape <[hidden email]> wrote:

On 20.03.2018, at 13:58, Holger Freyther <[hidden email]> wrote:
>
>
>
>
>> On 20. Mar 2018, at 11:54, Tobias Pape <[hidden email]> wrote:
>>
>
> Hey!
>
>
>> https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355802605#L2458
>
> indeed! I think it is this bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59477. I love how GCC developers add a regression test for every issue fixed and they do it for many many years already. And it is a codebase that is 30+ years old and started before SSA was "known".

Yes, upgrading GCC fixes this (see https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/builds/355826629 )

But the newspeak-bootstrap now fails.

@Eliot, can you have a look at this?
        https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355826630#L6518
I actually cannot fathom how a gcc upgrade should influence this, but I'll hold whit the Pull Request until you say that its either my fault or nothing to be afraid of.

Well, if it's a result of my enabling read-only objects (-DIMMUTABILITY=1) for all Spur VMs then I'll turn it off for the Newspeak builds,

Hm, I though object immutability was something implemented years ago? Before Cog even.
 
but the key point here is that we should decouple the Newspeak builds from the Pharo and Squeak builds asap.  Newspeak shouldn't block new releases of what should be perfectly good VMs.  This is urgent IMO.

If any VM configuration breaks because a compiler update, it's very likely because of undefined behavior, and it casts doubt on the correctness of the other configurations sharing code.

Reply | Threaded
Open this post in threaded view
|

Re: cute, we crash gcc...

Tobias Pape
 

> On 22.03.2018, at 03:51, Ryan Macnak <[hidden email]> wrote:
>
> On Tue, Mar 20, 2018 at 8:27 AM, Eliot Miranda <[hidden email]> wrote:
> Hi Tobias, Hi Ryan,
>
> On Tue, Mar 20, 2018 at 6:12 AM, Tobias Pape <[hidden email]> wrote:
>
> On 20.03.2018, at 13:58, Holger Freyther <[hidden email]> wrote:
> >
> >
> >
> >
> >> On 20. Mar 2018, at 11:54, Tobias Pape <[hidden email]> wrote:
> >>
> >
> > Hey!
> >
> >
> >> https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355802605#L2458
> >
> > indeed! I think it is this bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59477. I love how GCC developers add a regression test for every issue fixed and they do it for many many years already. And it is a codebase that is 30+ years old and started before SSA was "known".
>
> Yes, upgrading GCC fixes this (see https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/builds/355826629 )
>
> But the newspeak-bootstrap now fails.
>
> @Eliot, can you have a look at this?
>         https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355826630#L6518
> I actually cannot fathom how a gcc upgrade should influence this, but I'll hold whit the Pull Request until you say that its either my fault or nothing to be afraid of.
>
> Well, if it's a result of my enabling read-only objects (-DIMMUTABILITY=1) for all Spur VMs then I'll turn it off for the Newspeak builds,
>
> Hm, I though object immutability was something implemented years ago? Before Cog even.
>  

Maybe, but that is a different variant then.

> but the key point here is that we should decouple the Newspeak builds from the Pharo and Squeak builds asap.  Newspeak shouldn't block new releases of what should be perfectly good VMs.  This is urgent IMO.
>
> If any VM configuration breaks because a compiler update, it's very likely because of undefined behavior, and it casts doubt on the correctness of the other configurations sharing code.

No, in this case it's just a (known) compiler bug.

Best regards
        -Tobias

Reply | Threaded
Open this post in threaded view
|

Re: cute, we crash gcc...

Eliot Miranda-2
In reply to this post by Ryan Macnak
 
Hi Ryan,


On Mar 21, 2018, at 7:51 PM, Ryan Macnak <[hidden email]> wrote:

On Tue, Mar 20, 2018 at 8:27 AM, Eliot Miranda <[hidden email]> wrote:
Hi Tobias, Hi Ryan,

On Tue, Mar 20, 2018 at 6:12 AM, Tobias Pape <[hidden email]> wrote:

On 20.03.2018, at 13:58, Holger Freyther <[hidden email]> wrote:
>
>
>
>
>> On 20. Mar 2018, at 11:54, Tobias Pape <[hidden email]> wrote:
>>
>
> Hey!
>
>
>> https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355802605#L2458
>
> indeed! I think it is this bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59477. I love how GCC developers add a regression test for every issue fixed and they do it for many many years already. And it is a codebase that is 30+ years old and started before SSA was "known".

Yes, upgrading GCC fixes this (see https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/builds/355826629 )

But the newspeak-bootstrap now fails.

@Eliot, can you have a look at this?
        https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355826630#L6518
I actually cannot fathom how a gcc upgrade should influence this, but I'll hold whit the Pull Request until you say that its either my fault or nothing to be afraid of.

Well, if it's a result of my enabling read-only objects (-DIMMUTABILITY=1) for all Spur VMs then I'll turn it off for the Newspeak builds,

Hm, I though object immutability was something implemented years ago? Before Cog even.

Yes, but it happened in a fork of the Squeak VM we used at Cadence for the original Newspeak platform.  This VM wasn't carried forward and was a pure interpreter, so only some of the work was ported.

 
but the key point here is that we should decouple the Newspeak builds from the Pharo and Squeak builds asap.  Newspeak shouldn't block new releases of what should be perfectly good VMs.  This is urgent IMO.

If any VM configuration breaks because a compiler update, it's very likely because of undefined behavior, and it casts doubt on the correctness of the other configurations sharing code.

In any case, if you have to time, it would be great if you could take a look at the failing bootstrap and diagnose.

regards, Eliot
_,,,^..^,,,_ (phone)
Reply | Threaded
Open this post in threaded view
|

Re: cute, we crash gcc...

Eliot Miranda-2
In reply to this post by Tobias Pape
 
Hi Tobias,


> On Mar 21, 2018, at 10:54 PM, Tobias Pape <[hidden email]> wrote:
>
>
>
>> On 22.03.2018, at 03:51, Ryan Macnak <[hidden email]> wrote:
>>
>> On Tue, Mar 20, 2018 at 8:27 AM, Eliot Miranda <[hidden email]> wrote:
>> Hi Tobias, Hi Ryan,
>>
>> On Tue, Mar 20, 2018 at 6:12 AM, Tobias Pape <[hidden email]> wrote:
>>
>> On 20.03.2018, at 13:58, Holger Freyther <[hidden email]> wrote:
>>>
>>>
>>>
>>>
>>>> On 20. Mar 2018, at 11:54, Tobias Pape <[hidden email]> wrote:
>>>>
>>>
>>> Hey!
>>>
>>>
>>>> https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355802605#L2458
>>>
>>> indeed! I think it is this bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59477. I love how GCC developers add a regression test for every issue fixed and they do it for many many years already. And it is a codebase that is 30+ years old and started before SSA was "known".
>>
>> Yes, upgrading GCC fixes this (see https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/builds/355826629 )
>>
>> But the newspeak-bootstrap now fails.
>>
>> @Eliot, can you have a look at this?
>>        https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355826630#L6518
>> I actually cannot fathom how a gcc upgrade should influence this, but I'll hold whit the Pull Request until you say that its either my fault or nothing to be afraid of.
>>
>> Well, if it's a result of my enabling read-only objects (-DIMMUTABILITY=1) for all Spur VMs then I'll turn it off for the Newspeak builds,
>>
>> Hm, I though object immutability was something implemented years ago? Before Cog even.
>>
>
> Maybe, but that is a different variant then.
>
>> but the key point here is that we should decouple the Newspeak builds from the Pharo and Squeak builds asap.  Newspeak shouldn't block new releases of what should be perfectly good VMs.  This is urgent IMO.
>>
>> If any VM configuration breaks because a compiler update, it's very likely because of undefined behavior, and it casts doubt on the correctness of the other configurations sharing code.
>
> No, in this case it's just a (known) compiler bug.

No, Tobias.  It is urgent we decouple the Newspeak builds from the VM pipeline.  Cadence no longer use Newspeak.  There is very little energy in the Newspeak community to maintain the system.  Therefore when it breaks it can take a long time to fix.  We cannot be dependent on this.  It has nothing to do with specific bugs.  We need to have a build and verification pipeline that tests Squeak and Pharo, not one that first tests Newspeak and stalls, without testing Squeak and Pharo, if there are any problems in Newspeak.

>
> Best regards
>    -Tobias


_,,,^..^,,,_ (phone)
Reply | Threaded
Open this post in threaded view
|

Re: cute, we crash gcc...

Ryan Macnak
In reply to this post by Eliot Miranda-2
 
On Thu, Mar 22, 2018 at 7:22 AM, Eliot Miranda <[hidden email]> wrote:
Hi Ryan,


On Mar 21, 2018, at 7:51 PM, Ryan Macnak <[hidden email]> wrote:

On Tue, Mar 20, 2018 at 8:27 AM, Eliot Miranda <[hidden email]> wrote:
Hi Tobias, Hi Ryan,

On Tue, Mar 20, 2018 at 6:12 AM, Tobias Pape <[hidden email]> wrote:

On 20.03.2018, at 13:58, Holger Freyther <[hidden email]> wrote:
>
>
>
>
>> On 20. Mar 2018, at 11:54, Tobias Pape <[hidden email]> wrote:
>>
>
> Hey!
>
>
>> https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355802605#L2458
>
> indeed! I think it is this bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59477. I love how GCC developers add a regression test for every issue fixed and they do it for many many years already. And it is a codebase that is 30+ years old and started before SSA was "known".

Yes, upgrading GCC fixes this (see https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/builds/355826629 )

But the newspeak-bootstrap now fails.

@Eliot, can you have a look at this?
        https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355826630#L6518
I actually cannot fathom how a gcc upgrade should influence this, but I'll hold whit the Pull Request until you say that its either my fault or nothing to be afraid of.

Well, if it's a result of my enabling read-only objects (-DIMMUTABILITY=1) for all Spur VMs then I'll turn it off for the Newspeak builds,

Hm, I though object immutability was something implemented years ago? Before Cog even.

Yes, but it happened in a fork of the Squeak VM we used at Cadence for the original Newspeak platform.  This VM wasn't carried forward and was a pure interpreter, so only some of the work was ported.

 
but the key point here is that we should decouple the Newspeak builds from the Pharo and Squeak builds asap.  Newspeak shouldn't block new releases of what should be perfectly good VMs.  This is urgent IMO.

If any VM configuration breaks because a compiler update, it's very likely because of undefined behavior, and it casts doubt on the correctness of the other configurations sharing code.

In any case, if you have to time, it would be great if you could take a look at the failing bootstrap and diagnose.

A Strongtalk dictionary is failing to rehash on growth (1229.1) and failing to find an element that was definitely inserted (1230.35). Most likely cause is a unstable hash from one of the Fraction or ByteSymbol elements.
 

regards, Eliot
_,,,^..^,,,_ (phone)

Reply | Threaded
Open this post in threaded view
|

Re: cute, we crash gcc...

Tobias Pape
 
Hi Ryan,

> On 23.03.2018, at 05:24, Ryan Macnak <[hidden email]> wrote:
> […]
>>
>> Yes, upgrading GCC fixes this (see https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/builds/355826629 )
>>
>> But the newspeak-bootstrap now fails.
>>
>> @Eliot, can you have a look at this?
>>         https://travis-ci.org/OpenSmalltalk/opensmalltalk-vm/jobs/355826630#L6518
>> I actually cannot fathom how a gcc upgrade should influence this, but I'll hold whit the Pull Request until you say that its either my fault or nothing to be afraid of.
>>
>> Well, if it's a result of my enabling read-only objects (-DIMMUTABILITY=1) for all Spur VMs then I'll turn it off for the Newspeak builds,
>>
>> Hm, I though object immutability was something implemented years ago? Before Cog even.
>
> Yes, but it happened in a fork of the Squeak VM we used at Cadence for the original Newspeak platform.  This VM wasn't carried forward and was a pure interpreter, so only some of the work was ported.
>
>>  
>> but the key point here is that we should decouple the Newspeak builds from the Pharo and Squeak builds asap.  Newspeak shouldn't block new releases of what should be perfectly good VMs.  This is urgent IMO.
>>
>> If any VM configuration breaks because a compiler update, it's very likely because of undefined behavior, and it casts doubt on the correctness of the other configurations sharing code.
>
> In any case, if you have to time, it would be great if you could take a look at the failing bootstrap and diagnose.
>
> A Strongtalk dictionary is failing to rehash on growth (1229.1) and failing to find an element that was definitely inserted (1230.35). Most likely cause is a unstable hash from one of the Fraction or ByteSymbol elements.

Sounds in line with the stack trace to me. can this even possibly related to immutability? to my ears this sounds pretty far removed from each other…

best regards
        -Tobias