[Cog] cmType ~= CMMethod

List overview All Threads
Download

newer

older

[CMake] Can we create...

[Q] Can I convert old image (3.8...

Mariano Martinez Peck

20 Dec 2010 20 Dec '10

2:31 p.m.

Hi Eliot. I am having some asserts failing while running my tests. Most of them during the sweep phase of the GC.

The two places where it fails is in

- NewCoObjectMemory >> lastPointerWhileForwarding: oop

if fails the assert: self assert: (coInterpreter cCoerceSimple: methodHeader to: #'CogMethod *') cmType = CMMethod. In different runs, the cmType answer me 1 (CMFree)

- NewCoObjectMemory >> lastPointerOf: oop recordWeakRoot: recordWeakRoot "<Boolean>"

and it fails in the same assert: self assert: (coInterpreter cCoerceSimple: methodHeader to: #'CogMethod *') cmType = CMMethod.

In different runs, I got values 0 or 6. BTW....seing:

initializeCogMethodConstants CMOpenPIC := 1 + (CMClosedPIC := 1 + (CMBlock := 1 +(CMMethod := 1 + (CMFree := 1))))

I cannot understand how can I get 0 or 6....

Anyway, my question is if you have a clue of what can be happening so that those asserts are failing. I am using Objects as methods, but normal objects (subclasses of ProtoObject).

Thanks in advance,

mariano

Attachments:

attachment.html (text/html — 1.1 KB)

Show replies by date

Mariano Martinez Peck

20 Dec 20 Dec

4:49 p.m.

And in addition to this assert, there is also one more that fail and it is StackInterpreter >> createActualMessageTo: lookupClass the assert: self assert: ((objectMemory isIntegerObject: messageSelector) or: [objectMemory addressCouldBeObj: messageSelector]).

After debugging, I notice that it fails because, it is NOT an integer, and #addressCouldBeObj: all answers true, but "(self headerType: address) ~= HeaderTypeGC"

This means that such object, has the header type HeaderTypeGC.

so...maybe with this and the previous mail it can ring a bell in someone's head.

Thanks a lot in advance for any hint you can point me.

Mariano

On Mon, Dec 20, 2010 at 2:31 PM, Mariano Martinez Peck < marianopeck@gmail.com> wrote:

...

Hi Eliot. I am having some asserts failing while running my tests. Most of them during the sweep phase of the GC.

The two places where it fails is in

NewCoObjectMemory >> lastPointerWhileForwarding: oop

if fails the assert: self assert: (coInterpreter cCoerceSimple: methodHeader to: #'CogMethod *') cmType = CMMethod. In different runs, the cmType answer me 1 (CMFree)

NewCoObjectMemory >> lastPointerOf: oop recordWeakRoot: recordWeakRoot

"<Boolean>"

and it fails in the same assert: self assert: (coInterpreter cCoerceSimple: methodHeader to: #'CogMethod *') cmType = CMMethod.

In different runs, I got values 0 or 6. BTW....seing:

initializeCogMethodConstants CMOpenPIC := 1 + (CMClosedPIC := 1 + (CMBlock := 1 +(CMMethod := 1 + (CMFree := 1))))

I cannot understand how can I get 0 or 6....

Anyway, my question is if you have a clue of what can be happening so that those asserts are failing. I am using Objects as methods, but normal objects (subclasses of ProtoObject).

Thanks in advance,

mariano

Eliot Miranda

7:55 p.m.

Hi Mariano,

this looks like a Cog VM bug. You should try and create a reproducible case and then send it to me.

On Mon, Dec 20, 2010 at 5:31 AM, Mariano Martinez Peck < marianopeck@gmail.com> wrote:

...

Hi Eliot. I am having some asserts failing while running my tests. Most of them during the sweep phase of the GC.

The two places where it fails is in

NewCoObjectMemory >> lastPointerWhileForwarding: oop

if fails the assert: self assert: (coInterpreter cCoerceSimple: methodHeader to: #'CogMethod *') cmType = CMMethod. In different runs, the cmType answer me 1 (CMFree)

NewCoObjectMemory >> lastPointerOf: oop recordWeakRoot: recordWeakRoot

"<Boolean>"

and it fails in the same assert: self assert: (coInterpreter cCoerceSimple: methodHeader to: #'CogMethod *') cmType = CMMethod.

In different runs, I got values 0 or 6. BTW....seing:

initializeCogMethodConstants CMOpenPIC := 1 + (CMClosedPIC := 1 + (CMBlock := 1 +(CMMethod := 1 + (CMFree := 1))))

I cannot understand how can I get 0 or 6....

Anyway, my question is if you have a clue of what can be happening so that those asserts are failing. I am using Objects as methods, but normal objects (subclasses of ProtoObject).

Thanks in advance,

mariano

stephane ducasse

21 Dec 21 Dec

4:54 p.m.

mariano just left and said that he will provide an image read to crash.

Stef

On Dec 20, 2010, at 7:55 PM, Eliot Miranda wrote:

...

Hi Mariano,

this looks like a Cog VM bug. You should try and create a reproducible case and then send it to me.

On Mon, Dec 20, 2010 at 5:31 AM, Mariano Martinez Peck marianopeck@gmail.com wrote:

Hi Eliot. I am having some asserts failing while running my tests. Most of them during the sweep phase of the GC.

The two places where it fails is in

NewCoObjectMemory >> lastPointerWhileForwarding: oop

if fails the assert: self assert: (coInterpreter cCoerceSimple: methodHeader to: #'CogMethod *') cmType = CMMethod. In different runs, the cmType answer me 1 (CMFree)

NewCoObjectMemory >> lastPointerOf: oop recordWeakRoot: recordWeakRoot "<Boolean>"

and it fails in the same assert: self assert: (coInterpreter cCoerceSimple: methodHeader to: #'CogMethod *') cmType = CMMethod.

In different runs, I got values 0 or 6. BTW....seing:

initializeCogMethodConstants CMOpenPIC := 1 + (CMClosedPIC := 1 + (CMBlock := 1 +(CMMethod := 1 + (CMFree := 1))))

I cannot understand how can I get 0 or 6....

Anyway, my question is if you have a clue of what can be happening so that those asserts are failing. I am using Objects as methods, but normal objects (subclasses of ProtoObject).

Thanks in advance,

mariano

Mariano Martinez Peck

7:53 p.m.

On Mon, Dec 20, 2010 at 7:55 PM, Eliot Miranda eliot.miranda@gmail.comwrote:

...

Hi Mariano,

this looks like a Cog VM bug. You should try and create a reproducible case and then send it to me.

Hi Eliot. Define "reproducable case" ;) Really, I don't have a simple TestCase/Doit that reproduce this, but I can be sure that when running all my tests (that use objects as methods) it crashes. If it is not in the first time, for SURE in the second. But as said, it is not always in the same place not the same test. Because the problem seems while running an incremental GC.

What I can provide you is a Pharo image, with all my code loaded, and the test runner open with the test you have to run. And promise: if the first time or second, you will have the crash. I tried to upload the image to the gforge project and we have a problem right now with it. I just send it by private email to you. I hope you don'y bother.

I can tell you all this information:

- My tests are all about using proxies for compiled methods (subclasses of proto object, not SmallInteger). - The crash DOES NOT happens if I compile a StackVM, it ONLY happens in CogVM. - With Cog, I have the crash with both, Development or Production.

Thanks A LOT in advance,

Mariano

...

On Mon, Dec 20, 2010 at 5:31 AM, Mariano Martinez Peck < marianopeck@gmail.com> wrote:

...
Hi Eliot. I am having some asserts failing while running my tests. Most of them during the sweep phase of the GC.

The two places where it fails is in

NewCoObjectMemory >> lastPointerWhileForwarding: oop

if fails the assert: self assert: (coInterpreter cCoerceSimple: methodHeader to: #'CogMethod *') cmType = CMMethod. In different runs, the cmType answer me 1 (CMFree)

NewCoObjectMemory >> lastPointerOf: oop recordWeakRoot: recordWeakRoot

"<Boolean>"

and it fails in the same assert: self assert: (coInterpreter cCoerceSimple: methodHeader to: #'CogMethod *') cmType = CMMethod.

In different runs, I got values 0 or 6. BTW....seing:

initializeCogMethodConstants CMOpenPIC := 1 + (CMClosedPIC := 1 + (CMBlock := 1 +(CMMethod := 1 + (CMFree := 1))))

I cannot understand how can I get 0 or 6....

Anyway, my question is if you have a clue of what can be happening so that those asserts are failing. I am using Objects as methods, but normal objects (subclasses of ProtoObject).

Thanks in advance,

mariano

Mariano Martinez Peck

1 Apr 1 Apr

12:33 p.m.

Hi Eliot. I am still finding this problem with latest CogVM. I cannot say more than what I have already say. I've tried with all your VMs, including the latest VM.r2370. I tried Pharo CI vms, and compiled by my own...all same result so I think it is a problem in the core of Cog, related to the usage of objects as methods.

Just in case I am in Mac OS 10.6.7

I can tell you all this information:

- My tests are all about using proxies for compiled methods (using the #cannotInterpret: and #run:with:in tricks). - The crash DOES NOT happens if I compile a StackVM, it ONLY happens in CogVM. - With Cog, I have the crash with both, Development or Production.

I have prepared an image with a test runner open where you only need to run the test. If it doesn't crash in the first time, in the second it will for sure.

Link to the image prepared: https://gforge.inria.fr/frs/download.php/28432/cogCrash.zip

I also told you that in the gdb console, just before crashing I can see: