Smalltalk › Squeak › Squeak VM

Why spaceOkay test needed?

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

5 messages Options

Adrian Lienhard

Why spaceOkay test needed?

Out of curiosity, does anybody know why in...

primitiveNew
"Allocate a new fixed-size instance. Fail if the allocation would
leave less than lowSpaceThreshold bytes free. May cause a GC"
| class spaceOkay |
class := self stackTop.
"The following may cause GC!"
spaceOkay := self sufficientSpaceToInstantiate: class indexableSize: 0.
self success: spaceOkay.
successFlag ifTrue: [ self push: (self instantiateClass: self
popStack indexableSize: 0) ]

... there is the call to #sufficientSpaceToInstantiate:indexableSize:
although #instantiateClass:indexableSize: also indirectly calls
#sufficientSpaceToAllocate:?
Both, the primitive fallback code and the vm code signal the low
space sem in case of failure.

Adrian

Andreas.Raab

Re: Why spaceOkay test needed?

Adrian Lienhard wrote:

>
> Out of curiosity, does anybody know why in...
>
> primitiveNew
> "Allocate a new fixed-size instance. Fail if the allocation would
> leave less than lowSpaceThreshold bytes free. May cause a GC"
> | class spaceOkay |
> class := self stackTop.
> "The following may cause GC!"
> spaceOkay := self sufficientSpaceToInstantiate: class indexableSize: 0.
> self success: spaceOkay.
> successFlag ifTrue: [ self push: (self instantiateClass: self
> popStack indexableSize: 0) ]
>
> ... there is the call to #sufficientSpaceToInstantiate:indexableSize:
> although #instantiateClass:indexableSize: also indirectly calls
> #sufficientSpaceToAllocate:?

Where does it do that?

> Both, the primitive fallback code and the vm code signal the low space
> sem in case of failure.

The fallback code in the image is broken IMO. It needs to signal
OutOfMemory instead of the low space semaphore. We had our servers go
belly-up because of malformed requests that requested a 1 gigabyte or
more memory.

Cheers,
- Andreas

Adrian Lienhard

Re: Why spaceOkay test needed?

On Aug 20, 2007, at 18:37 , Andreas Raab wrote:

instantiateClass:indexableSize: ->
allocate:headerSize:h1:h2:h3:doFill:with: -> allocateChunk: ->
sufficientSpaceToAllocate:

Adrian

Mathieu SUEN

Re: Why spaceOkay test needed?

In reply to this post by Andreas.Raab

So #sufficientSpaceToInstantiate:indexableSize: call can be removed?

Mth

On Aug 20, 2007, at 6:37 PM, Andreas Raab wrote:

> Adrian Lienhard wrote:
>> Out of curiosity, does anybody know why in...
>> primitiveNew
>> "Allocate a new fixed-size instance. Fail if the allocation
>> would leave less than lowSpaceThreshold bytes free. May cause a GC"
>> | class spaceOkay |
>> class := self stackTop.
>> "The following may cause GC!"
>> spaceOkay := self sufficientSpaceToInstantiate: class
>> indexableSize: 0.
>> self success: spaceOkay.
>> successFlag ifTrue: [ self push: (self instantiateClass: self
>> popStack indexableSize: 0) ]
>> ... there is the call to
>> #sufficientSpaceToInstantiate:indexableSize: although
>> #instantiateClass:indexableSize: also indirectly calls
>> #sufficientSpaceToAllocate:?
>
> Where does it do that?
>
>> Both, the primitive fallback code and the vm code signal the low
>> space sem in case of failure.
>
> The fallback code in the image is broken IMO. It needs to signal
> OutOfMemory instead of the low space semaphore. We had our servers
> go belly-up because of malformed requests that requested a 1
> gigabyte or more memory.
>
> Cheers,
> - Andreas

timrowledge

Re: Why spaceOkay test needed?

On 23-Aug-07, at 3:52 PM, Mathieu Suen wrote:

> So #sufficientSpaceToInstantiate:indexableSize: call can be removed?

From where?

You might *possibly* be able to make a good argument for removing it
from within the instantiateClass:indexableSize: depending on its
usage. I wouldn't recommend removing it from the primitiveNew code
though; how else are you going to be able to fail the prim cleanly?
Sure, you could make the instantiateClass:indexableSize: code return
nil if the instantiation cannot be performed. Then you have to modify
all senders to handle that eventuality. You'd have to not push: the
nil though.

tim
--
tim Rowledge; [hidden email]; http://www.rowledge.org/tim
A conscience is what hurts when all your other parts feel so good.