Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

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

Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

Clément Béra
Hi All,

It seems Pharo is lacking the slotAt: / slotAt:put: primitive and that the instVarAt: / instVarAt:put: primitive is not doing what it says it does. When I look at Object>>#instVarAt: I see:

"[...]Answer a fixed variable in an object. The numbering of the variables
corresponds to the named instance variables, followed by the indexed instance
variables. Fail if the index is not an Integer or is not the index of a fixed variable[...]"

Now in Pharo 7 when one does: 

#(iv1) instVarAt: 1, one gets #iv1.

So the comment is wrong.

The reason for this is that instVarAt: / instVarAt:put: should be primitive 73/74, while they are in Pharo 173/174 (primitive slotAt:/slotAt:put:). The slot primitive access pointer slots of objects, whichever inst var or variable fields. A good example would be:

| t |
t := #temp.
[ t ] instVarAt: 4, one gets #temp

So I have two solutions for this problem:
1) We change the instVarAt: / instVarAt:put: primitives comments to match the slotAt: / slotAt:put: behavior, and in Pharo instVarAt: / instVarAt:put: means in factslotAt: / slotAt:put: .
2) We change the instVarAt: / instVarAt:put: primitives to primitive 73 and 74, we introduce slotAt: / slotAt:put: with primitive 173/174 and we ask everyone to fix their code relying on instVarAt: / instVarAt:put: acting as slotAt: / slotAt:put:

What do you think ?

Best,


--
Clément Béra
Pharo consortium engineer
Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq
Reply | Threaded
Open this post in threaded view
|

Re: Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

Denis Kudriashov
Hi.

Just to notice: it is here long time (I checked Pharo 1 and current Squeak).


2017-11-22 14:35 GMT+01:00 Clément Bera <[hidden email]>:
Hi All,

It seems Pharo is lacking the slotAt: / slotAt:put: primitive and that the instVarAt: / instVarAt:put: primitive is not doing what it says it does. When I look at Object>>#instVarAt: I see:

"[...]Answer a fixed variable in an object. The numbering of the variables
corresponds to the named instance variables, followed by the indexed instance
variables. Fail if the index is not an Integer or is not the index of a fixed variable[...]"

Now in Pharo 7 when one does: 

#(iv1) instVarAt: 1, one gets #iv1.

So the comment is wrong.

The reason for this is that instVarAt: / instVarAt:put: should be primitive 73/74, while they are in Pharo 173/174 (primitive slotAt:/slotAt:put:). The slot primitive access pointer slots of objects, whichever inst var or variable fields. A good example would be:

| t |
t := #temp.
[ t ] instVarAt: 4, one gets #temp

So I have two solutions for this problem:
1) We change the instVarAt: / instVarAt:put: primitives comments to match the slotAt: / slotAt:put: behavior, and in Pharo instVarAt: / instVarAt:put: means in factslotAt: / slotAt:put: .
2) We change the instVarAt: / instVarAt:put: primitives to primitive 73 and 74, we introduce slotAt: / slotAt:put: with primitive 173/174 and we ask everyone to fix their code relying on instVarAt: / instVarAt:put: acting as slotAt: / slotAt:put:

I prefer explicit behaviour. These methods have clear intention in names: inst var accessing. It should be error when given index is out of #instSize range. 

For new methods I would not use slot based naming because slots in Pharo is already something more then just object fields. 
Maybe #fieldAt:, #fieldAt:put: would be good? And also #fieldSize for consistency 
 

What do you think ?

Best,


--
Clément Béra
Pharo consortium engineer

Reply | Threaded
Open this post in threaded view
|

Re: Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

Clément Béra


On Wed, Nov 22, 2017 at 3:34 PM, Denis Kudriashov <[hidden email]> wrote:
Hi.

Just to notice: it is here long time (I checked Pharo 1 and current Squeak).

Probably some Spur changes to integrate in the image then.
 

2017-11-22 14:35 GMT+01:00 Clément Bera <[hidden email]>:
Hi All,

It seems Pharo is lacking the slotAt: / slotAt:put: primitive and that the instVarAt: / instVarAt:put: primitive is not doing what it says it does. When I look at Object>>#instVarAt: I see:

"[...]Answer a fixed variable in an object. The numbering of the variables
corresponds to the named instance variables, followed by the indexed instance
variables. Fail if the index is not an Integer or is not the index of a fixed variable[...]"

Now in Pharo 7 when one does: 

#(iv1) instVarAt: 1, one gets #iv1.

So the comment is wrong.

The reason for this is that instVarAt: / instVarAt:put: should be primitive 73/74, while they are in Pharo 173/174 (primitive slotAt:/slotAt:put:). The slot primitive access pointer slots of objects, whichever inst var or variable fields. A good example would be:

| t |
t := #temp.
[ t ] instVarAt: 4, one gets #temp

So I have two solutions for this problem:
1) We change the instVarAt: / instVarAt:put: primitives comments to match the slotAt: / slotAt:put: behavior, and in Pharo instVarAt: / instVarAt:put: means in factslotAt: / slotAt:put: .
2) We change the instVarAt: / instVarAt:put: primitives to primitive 73 and 74, we introduce slotAt: / slotAt:put: with primitive 173/174 and we ask everyone to fix their code relying on instVarAt: / instVarAt:put: acting as slotAt: / slotAt:put:

I prefer explicit behaviour. These methods have clear intention in names: inst var accessing. It should be error when given index is out of #instSize range. 

For new methods I would not use slot based naming because slots in Pharo is already something more then just object fields. 
Maybe #fieldAt:, #fieldAt:put: would be good? And also #fieldSize for consistency 

fieldSize is not a primitive in this case,

Object>>#fieldSize
     ^ self class instSize + self basicSize

 

What do you think ?

Best,


--
Clément Béra
Pharo consortium engineer




--
Clément Béra
Pharo consortium engineer
Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq
Reply | Threaded
Open this post in threaded view
|

Re: Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

Stephane Ducasse-3
In reply to this post by Clément Béra
Hi clement

you should discuss with marcus. I had the impression that he was
thinking that slot should disappear.

Stef

On Wed, Nov 22, 2017 at 2:35 PM, Clément Bera <[hidden email]> wrote:

> Hi All,
>
> It seems Pharo is lacking the slotAt: / slotAt:put: primitive and that the
> instVarAt: / instVarAt:put: primitive is not doing what it says it does.
> When I look at Object>>#instVarAt: I see:
>
> "[...]Answer a fixed variable in an object. The numbering of the variables
> corresponds to the named instance variables, followed by the indexed
> instance
> variables. Fail if the index is not an Integer or is not the index of a
> fixed variable[...]"
>
> Now in Pharo 7 when one does:
>
> #(iv1) instVarAt: 1, one gets #iv1.
>
> So the comment is wrong.
>
> The reason for this is that instVarAt: / instVarAt:put: should be primitive
> 73/74, while they are in Pharo 173/174 (primitive slotAt:/slotAt:put:). The
> slot primitive access pointer slots of objects, whichever inst var or
> variable fields. A good example would be:
>
> | t |
> t := #temp.
> [ t ] instVarAt: 4, one gets #temp
>
> So I have two solutions for this problem:
> 1) We change the instVarAt: / instVarAt:put: primitives comments to match
> the slotAt: / slotAt:put: behavior, and in Pharo instVarAt: / instVarAt:put:
> means in factslotAt: / slotAt:put: .
> 2) We change the instVarAt: / instVarAt:put: primitives to primitive 73 and
> 74, we introduce slotAt: / slotAt:put: with primitive 173/174 and we ask
> everyone to fix their code relying on instVarAt: / instVarAt:put: acting as
> slotAt: / slotAt:put:
>
> What do you think ?
>
> Best,
>
>
> --
> Clément Béra
> Pharo consortium engineer
> https://clementbera.wordpress.com/
> Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq

Reply | Threaded
Open this post in threaded view
|

Re: Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

Clément Béra


On Wed, Nov 22, 2017 at 9:05 PM, Stephane Ducasse <[hidden email]> wrote:
Hi clement

you should discuss with marcus. I had the impression that he was
thinking that slot should disappear.

Then let's use fieldAt:/fieldAt:put:
 

Stef

On Wed, Nov 22, 2017 at 2:35 PM, Clément Bera <[hidden email]> wrote:
> Hi All,
>
> It seems Pharo is lacking the slotAt: / slotAt:put: primitive and that the
> instVarAt: / instVarAt:put: primitive is not doing what it says it does.
> When I look at Object>>#instVarAt: I see:
>
> "[...]Answer a fixed variable in an object. The numbering of the variables
> corresponds to the named instance variables, followed by the indexed
> instance
> variables. Fail if the index is not an Integer or is not the index of a
> fixed variable[...]"
>
> Now in Pharo 7 when one does:
>
> #(iv1) instVarAt: 1, one gets #iv1.
>
> So the comment is wrong.
>
> The reason for this is that instVarAt: / instVarAt:put: should be primitive
> 73/74, while they are in Pharo 173/174 (primitive slotAt:/slotAt:put:). The
> slot primitive access pointer slots of objects, whichever inst var or
> variable fields. A good example would be:
>
> | t |
> t := #temp.
> [ t ] instVarAt: 4, one gets #temp
>
> So I have two solutions for this problem:
> 1) We change the instVarAt: / instVarAt:put: primitives comments to match
> the slotAt: / slotAt:put: behavior, and in Pharo instVarAt: / instVarAt:put:
> means in factslotAt: / slotAt:put: .
> 2) We change the instVarAt: / instVarAt:put: primitives to primitive 73 and
> 74, we introduce slotAt: / slotAt:put: with primitive 173/174 and we ask
> everyone to fix their code relying on instVarAt: / instVarAt:put: acting as
> slotAt: / slotAt:put:
>
> What do you think ?
>
> Best,
>
>
> --
> Clément Béra
> Pharo consortium engineer
> https://clementbera.wordpress.com/
> Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq




--
Clément Béra
Pharo consortium engineer
Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq
Reply | Threaded
Open this post in threaded view
|

Re: Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

Stephane Ducasse-3
Did you talk with marcus?


On Wed, Nov 22, 2017 at 10:12 PM, Clément Bera <[hidden email]> wrote:

>
>
> On Wed, Nov 22, 2017 at 9:05 PM, Stephane Ducasse <[hidden email]>
> wrote:
>>
>> Hi clement
>>
>> you should discuss with marcus. I had the impression that he was
>> thinking that slot should disappear.
>
>
> Then let's use fieldAt:/fieldAt:put:
>
>>
>>
>> Stef
>>
>> On Wed, Nov 22, 2017 at 2:35 PM, Clément Bera <[hidden email]>
>> wrote:
>> > Hi All,
>> >
>> > It seems Pharo is lacking the slotAt: / slotAt:put: primitive and that
>> > the
>> > instVarAt: / instVarAt:put: primitive is not doing what it says it does.
>> > When I look at Object>>#instVarAt: I see:
>> >
>> > "[...]Answer a fixed variable in an object. The numbering of the
>> > variables
>> > corresponds to the named instance variables, followed by the indexed
>> > instance
>> > variables. Fail if the index is not an Integer or is not the index of a
>> > fixed variable[...]"
>> >
>> > Now in Pharo 7 when one does:
>> >
>> > #(iv1) instVarAt: 1, one gets #iv1.
>> >
>> > So the comment is wrong.
>> >
>> > The reason for this is that instVarAt: / instVarAt:put: should be
>> > primitive
>> > 73/74, while they are in Pharo 173/174 (primitive slotAt:/slotAt:put:).
>> > The
>> > slot primitive access pointer slots of objects, whichever inst var or
>> > variable fields. A good example would be:
>> >
>> > | t |
>> > t := #temp.
>> > [ t ] instVarAt: 4, one gets #temp
>> >
>> > So I have two solutions for this problem:
>> > 1) We change the instVarAt: / instVarAt:put: primitives comments to
>> > match
>> > the slotAt: / slotAt:put: behavior, and in Pharo instVarAt: /
>> > instVarAt:put:
>> > means in factslotAt: / slotAt:put: .
>> > 2) We change the instVarAt: / instVarAt:put: primitives to primitive 73
>> > and
>> > 74, we introduce slotAt: / slotAt:put: with primitive 173/174 and we ask
>> > everyone to fix their code relying on instVarAt: / instVarAt:put: acting
>> > as
>> > slotAt: / slotAt:put:
>> >
>> > What do you think ?
>> >
>> > Best,
>> >
>> >
>> > --
>> > Clément Béra
>> > Pharo consortium engineer
>> > https://clementbera.wordpress.com/
>> > Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq
>>
>
>
>
> --
> Clément Béra
> Pharo consortium engineer
> https://clementbera.wordpress.com/
> Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq

Reply | Threaded
Open this post in threaded view
|

Re: Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

Clément Béra


On Fri, Nov 24, 2017 at 12:00 AM, Stephane Ducasse <[hidden email]> wrote:
Did you talk with marcus?

I don't understand the connection between slots and this problem with primitives. 


On Wed, Nov 22, 2017 at 10:12 PM, Clément Bera <[hidden email]> wrote:
>
>
> On Wed, Nov 22, 2017 at 9:05 PM, Stephane Ducasse <[hidden email]>
> wrote:
>>
>> Hi clement
>>
>> you should discuss with marcus. I had the impression that he was
>> thinking that slot should disappear.
>
>
> Then let's use fieldAt:/fieldAt:put:
>
>>
>>
>> Stef
>>
>> On Wed, Nov 22, 2017 at 2:35 PM, Clément Bera <[hidden email]>
>> wrote:
>> > Hi All,
>> >
>> > It seems Pharo is lacking the slotAt: / slotAt:put: primitive and that
>> > the
>> > instVarAt: / instVarAt:put: primitive is not doing what it says it does.
>> > When I look at Object>>#instVarAt: I see:
>> >
>> > "[...]Answer a fixed variable in an object. The numbering of the
>> > variables
>> > corresponds to the named instance variables, followed by the indexed
>> > instance
>> > variables. Fail if the index is not an Integer or is not the index of a
>> > fixed variable[...]"
>> >
>> > Now in Pharo 7 when one does:
>> >
>> > #(iv1) instVarAt: 1, one gets #iv1.
>> >
>> > So the comment is wrong.
>> >
>> > The reason for this is that instVarAt: / instVarAt:put: should be
>> > primitive
>> > 73/74, while they are in Pharo 173/174 (primitive slotAt:/slotAt:put:).
>> > The
>> > slot primitive access pointer slots of objects, whichever inst var or
>> > variable fields. A good example would be:
>> >
>> > | t |
>> > t := #temp.
>> > [ t ] instVarAt: 4, one gets #temp
>> >
>> > So I have two solutions for this problem:
>> > 1) We change the instVarAt: / instVarAt:put: primitives comments to
>> > match
>> > the slotAt: / slotAt:put: behavior, and in Pharo instVarAt: /
>> > instVarAt:put:
>> > means in factslotAt: / slotAt:put: .
>> > 2) We change the instVarAt: / instVarAt:put: primitives to primitive 73
>> > and
>> > 74, we introduce slotAt: / slotAt:put: with primitive 173/174 and we ask
>> > everyone to fix their code relying on instVarAt: / instVarAt:put: acting
>> > as
>> > slotAt: / slotAt:put:
>> >
>> > What do you think ?
>> >
>> > Best,
>> >
>> >
>> > --
>> > Clément Béra
>> > Pharo consortium engineer
>> > https://clementbera.wordpress.com/
>> > Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq
>>
>
>
>
> --
> Clément Béra
> Pharo consortium engineer
> https://clementbera.wordpress.com/
> Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq




--
Clément Béra
Pharo consortium engineer
Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq
Reply | Threaded
Open this post in threaded view
|

Re: Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

Marcus Denker-4


On 24 Nov 2017, at 00:21, Clément Bera <[hidden email]> wrote:



On Fri, Nov 24, 2017 at 12:00 AM, Stephane Ducasse <[hidden email]> wrote:
Did you talk with marcus?

I don't understand the connection between slots and this problem with primitives. 


There is no connection, just the same name.

For Slots aka First Class Instance variables:

I sometimes think that Slot for the first class instance variables might not be a good name, but then,
the only alternative would be “instance variable”, but that is not that nice either, as these Slots include
virtual variables (that are computed) or variables that are combine stored in a hidden base slot (e.g.
used for BooleanSlot or PropertySlot). 

So for now I will keep the Slot term…


Marcus
Reply | Threaded
Open this post in threaded view
|

Re: Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

Clément Béra
In the VM primitiveSlotAt: / at:Put: mean access to any field (pointer or non pointer, sorry for the confusion Marcus earlier on) of an object. The primitive in the image can be named fieldAt:/fieldAt:put: if it makes more sense for Pharo.

#[0] slotAt: 1 => 0
#(#a) slotAt: 1 => #a
0@2 slotAt: 1 => 0
With Pharo 6 blocks:
| t | t := #tmp. {[ t ] slotAt: 3 . [ t ] slotAt: 4} => { 0 "numArgs". #tmp } 

Right now you can do that with instVarAt: on the contrary to what instVarAt: comment states, and this is wrong. 

On Fri, Nov 24, 2017 at 10:32 AM, Marcus Denker <[hidden email]> wrote:


On 24 Nov 2017, at 00:21, Clément Bera <[hidden email]> wrote:



On Fri, Nov 24, 2017 at 12:00 AM, Stephane Ducasse <[hidden email]> wrote:
Did you talk with marcus?

I don't understand the connection between slots and this problem with primitives. 


There is no connection, just the same name.

For Slots aka First Class Instance variables:

I sometimes think that Slot for the first class instance variables might not be a good name, but then,
the only alternative would be “instance variable”, but that is not that nice either, as these Slots include
virtual variables (that are computed) or variables that are combine stored in a hidden base slot (e.g.
used for BooleanSlot or PropertySlot). 

So for now I will keep the Slot term…


Marcus



--
Clément Béra
Pharo consortium engineer
Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq
Reply | Threaded
Open this post in threaded view
|

Re: Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

Stephane Ducasse-3
Hi clement

I was confused. I did not understand the problem. Not enough
concentration right now.

Stef

On Fri, Nov 24, 2017 at 10:44 AM, Clément Bera <[hidden email]> wrote:

> In the VM primitiveSlotAt: / at:Put: mean access to any field (pointer or
> non pointer, sorry for the confusion Marcus earlier on) of an object. The
> primitive in the image can be named fieldAt:/fieldAt:put: if it makes more
> sense for Pharo.
>
> #[0] slotAt: 1 => 0
> #(#a) slotAt: 1 => #a
> 0@2 slotAt: 1 => 0
> With Pharo 6 blocks:
> | t | t := #tmp. {[ t ] slotAt: 3 . [ t ] slotAt: 4} => { 0 "numArgs". #tmp
> }
>
> Right now you can do that with instVarAt: on the contrary to what instVarAt:
> comment states, and this is wrong.
>
> On Fri, Nov 24, 2017 at 10:32 AM, Marcus Denker <[hidden email]>
> wrote:
>>
>>
>>
>> On 24 Nov 2017, at 00:21, Clément Bera <[hidden email]> wrote:
>>
>>
>>
>> On Fri, Nov 24, 2017 at 12:00 AM, Stephane Ducasse
>> <[hidden email]> wrote:
>>>
>>> Did you talk with marcus?
>>
>>
>> I don't understand the connection between slots and this problem with
>> primitives.
>>>
>>>
>>
>> There is no connection, just the same name.
>>
>> For Slots aka First Class Instance variables:
>>
>> I sometimes think that Slot for the first class instance variables might
>> not be a good name, but then,
>> the only alternative would be “instance variable”, but that is not that
>> nice either, as these Slots include
>> virtual variables (that are computed) or variables that are combine stored
>> in a hidden base slot (e.g.
>> used for BooleanSlot or PropertySlot).
>>
>> So for now I will keep the Slot term…
>>
>>
>> Marcus
>
>
>
>
> --
> Clément Béra
> Pharo consortium engineer
> https://clementbera.wordpress.com/
> Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq

Reply | Threaded
Open this post in threaded view
|

Re: Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

Eliot Miranda-2
In reply to this post by Clément Béra
Hi Clément,

On Wed, Nov 22, 2017 at 5:35 AM, Clément Bera <[hidden email]> wrote:
Hi All,

It seems Pharo is lacking the slotAt: / slotAt:put: primitive and that the instVarAt: / instVarAt:put: primitive is not doing what it says it does. When I look at Object>>#instVarAt: I see:

"[...]Answer a fixed variable in an object. The numbering of the variables
corresponds to the named instance variables, followed by the indexed instance
variables. Fail if the index is not an Integer or is not the index of a fixed variable[...]"

Now in Pharo 7 when one does: 

#(iv1) instVarAt: 1, one gets #iv1.

But the failure code should indeed answer #iv1.
 
So the comment is wrong.

The reason for this is that instVarAt: / instVarAt:put: should be primitive 73/74, while they are in Pharo 173/174 (primitive slotAt:/slotAt:put:).

Why do you say "should"?  The 173/174 primitives work well and avoid unnecessary primitive failures (which is why they're there).
 
The slot primitive access pointer slots of objects, whichever inst var or variable fields. A good example would be:

| t |
t := #temp.
[ t ] instVarAt: 4, one gets #temp

So I have two solutions for this problem:
1) We change the instVarAt: / instVarAt:put: primitives comments to match the slotAt: / slotAt:put: behavior, and in Pharo instVarAt: / instVarAt:put: means in factslotAt: / slotAt:put: .

This is the performant solution, because...
 
2) We change the instVarAt: / instVarAt:put: primitives to primitive 73 and 74, we introduce slotAt: / slotAt:put: with primitive 173/174 and we ask everyone to fix their code relying on instVarAt: / instVarAt:put: acting as slotAt: / slotAt:put:

Early in Squeak Spur testing Chris Muller noticed a slow down in his code   He was accessing an Array using instVarAt:, which was failing because he was using it to access an indexed field in an Array.  Now in Spur the VM must scan the object graph of a primitive's  receiver and arguments to some depth on primitive failure to follow forwarders.  Because a) at that time the VM computed a depth > 0 for instVarAt:[put:] it scanned the whole array on each access, and b) the Array was large, this meant that the failures were very expensive.  So I added the slot primitives, since the old 73/74 primitive pair were only used in the instVarAt:[put:] methods that accessed indexable slots on failure.  So the new 173/174 pair subsumed the 73/74 pair.

I eventually added an accessorDepth: pragma rot specify the accessor depth manually when the automatically computed depth is too deep.  It is for at:[put:] because the String version of the at: 60 primitive appears to access the innards of non-immediate Characters (it doesn't do this in Spur, but the depth inferencer isn't clever enough to notice).  So primitiveAt looks like:

primitiveAt
<accessorDepth: 0>
self commonAt: false

So we could go back to using 73/74 for instVarAt:[put:] provided that their accessorDepth ends up being 0, which likely means they will require the pragma.  But I think the simplest thing is to keep using 173/174.  Here's the Squeak code:

Object>>instVarAt: index
"Primitive. Answer a fixed variable in an object. The numbering of the variables
corresponds to the named instance variables, followed by the indexed instance
variables. Fail if the index is not an Integer or is not the index of a fixed or
indexed variable. Essential. See Object documentation whatIsAPrimitive."

<primitive: 173 error: ec>
self primitiveFailed

Object>>instVarAt: index put: anObject
"Primitive. Store a value into a fixed variable in an object. The numbering of the
variables corresponds to the named instance variables, followed by the indexed
instance variables. Fail if the index is not an Integer or is not the index of a fixed
or indexed variable. Essential. See Object documentation whatIsAPrimitive."

<primitive: 174 error: ec>
self primitiveFailed

I've attached these methods.


What do you think ?

I think for Spur it makes sense to use 173/174.
 

Best,


--
Clément Béra
Pharo consortium engineer



--
_,,,^..^,,,_
best, Eliot

Object-instVarAtput.st (808 bytes) Download Attachment
Object-instVarAt.st (770 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

Pharo Smalltalk Developers mailing list
In reply to this post by Clément Béra

Marcus,

I agree with keeping it as Slot. The existing literature (non-Smalltalk) uses Slot in a fairly consistent manner so what would the advantage be of changing it?

Reg

Sent from my iPhone

On Nov 24, 2017, at 4:44 AM, Clément Bera <[hidden email]> wrote:

In the VM primitiveSlotAt: / at:Put: mean access to any field (pointer or non pointer, sorry for the confusion Marcus earlier on) of an object. The primitive in the image can be named fieldAt:/fieldAt:put: if it makes more sense for Pharo..

#[0] slotAt: 1 => 0
#(#a) slotAt: 1 => #a
0@2 slotAt: 1 => 0
With Pharo 6 blocks:
| t | t := #tmp. {[ t ] slotAt: 3 . [ t ] slotAt: 4} => { 0 "numArgs". #tmp } 

Right now you can do that with instVarAt: on the contrary to what instVarAt: comment states, and this is wrong. 

On Fri, Nov 24, 2017 at 10:32 AM, Marcus Denker <[hidden email]> wrote:


On 24 Nov 2017, at 00:21, Clément Bera <[hidden email]> wrote:



On Fri, Nov 24, 2017 at 12:00 AM, Stephane Ducasse <[hidden email]> wrote:
Did you talk with marcus?

I don't understand the connection between slots and this problem with primitives. 


There is no connection, just the same name.

For Slots aka First Class Instance variables:

I sometimes think that Slot for the first class instance variables might not be a good name, but then,
the only alternative would be “instance variable”, but that is not that nice either, as these Slots include
virtual variables (that are computed) or variables that are combine stored in a hidden base slot (e.g..
used for BooleanSlot or PropertySlot). 

So for now I will keep the Slot term…


Marcus



--
Clément Béra
Pharo consortium engineer
Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq
Reply | Threaded
Open this post in threaded view
|

Re: Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

David T. Lewis
In reply to this post by Eliot Miranda-2
On Mon, Nov 27, 2017 at 12:11:13PM -0800, Eliot Miranda wrote:

> Hi Cl??ment,
>
> On Wed, Nov 22, 2017 at 5:35 AM, Cl??ment Bera <[hidden email]>
> wrote:
>
> > Hi All,
> >
> > It seems Pharo is lacking the slotAt: / slotAt:put: primitive and that the
> > instVarAt: / instVarAt:put: primitive is not doing what it says it does.
> > When I look at Object>>#instVarAt: I see:
> >
> > "[...]Answer a fixed variable in an object. The numbering of the variables
> > corresponds to the named instance variables, followed by the indexed
> > instance
> > variables. Fail if the index is not an Integer or is not the index of a
> > fixed variable[...]"
> >
> > Now in Pharo 7 when one does:
> >
> > #(iv1) instVarAt: 1, one gets #iv1.
> >
>
> But the failure code should indeed answer #iv1.
>
>
> > So the comment is wrong.
> >
> > The reason for this is that instVarAt: / instVarAt:put: should be
> > primitive 73/74, while they are in Pharo 173/174 (primitive
> > slotAt:/slotAt:put:).
> >
>
> Why do you say "should"?  The 173/174 primitives work well and avoid
> unnecessary primitive failures (which is why they're there).
>
>
> > The slot primitive access pointer slots of objects, whichever inst var or
> > variable fields. A good example would be:
> >
> > | t |
> > t := #temp.
> > [ t ] instVarAt: 4, one gets #temp
> >
> > So I have two solutions for this problem:
> > 1) We change the instVarAt: / instVarAt:put: primitives comments to match
> > the slotAt: / slotAt:put: behavior, and in Pharo instVarAt: /
> > instVarAt:put: means in factslotAt: / slotAt:put: .
> >
>
> This is the performant solution, because...
>
>
> > 2) We change the instVarAt: / instVarAt:put: primitives to primitive 73
> > and 74, we introduce slotAt: / slotAt:put: with primitive 173/174 and we
> > ask everyone to fix their code relying on instVarAt: / instVarAt:put:
> > acting as slotAt: / slotAt:put:
> >
>
> Early in Squeak Spur testing Chris Muller noticed a slow down in his code
> He was accessing an Array using instVarAt:, which was failing because he
> was using it to access an indexed field in an Array.  Now in Spur the VM
> must scan the object graph of a primitive's  receiver and arguments to some
> depth on primitive failure to follow forwarders.  Because a) at that time
> the VM computed a depth > 0 for instVarAt:[put:] it scanned the whole array
> on each access, and b) the Array was large, this meant that the failures
> were very expensive.  So I added the slot primitives, since the old 73/74
> primitive pair were only used in the instVarAt:[put:] methods that accessed
> indexable slots on failure.  So the new 173/174 pair subsumed the 73/74
> pair.
>
> I eventually added an accessorDepth: pragma rot specify the accessor depth
> manually when the automatically computed depth is too deep.  It is for
> at:[put:] because the String version of the at: 60 primitive appears to
> access the innards of non-immediate Characters (it doesn't do this in Spur,
> but the depth inferencer isn't clever enough to notice).  So primitiveAt
> looks like:
>
> primitiveAt
> <accessorDepth: 0>
> self commonAt: false
>
> So we could go back to using 73/74 for instVarAt:[put:] provided that their
> accessorDepth ends up being 0, which likely means they will require the
> pragma.  But I think the simplest thing is to keep using 173/174.  Here's
> the Squeak code:
>
> Object>>instVarAt: index
> "Primitive. Answer a fixed variable in an object. The numbering of the
> variables
> corresponds to the named instance variables, followed by the indexed
> instance
> variables. Fail if the index is not an Integer or is not the index of a
> fixed or
> indexed variable. Essential. See Object documentation whatIsAPrimitive."
>
> <primitive: 173 error: ec>
> self primitiveFailed
>
> Object>>instVarAt: index put: anObject
> "Primitive. Store a value into a fixed variable in an object. The numbering
> of the
> variables corresponds to the named instance variables, followed by the
> indexed
> instance variables. Fail if the index is not an Integer or is not the index
> of a fixed
> or indexed variable. Essential. See Object documentation whatIsAPrimitive."
>
> <primitive: 174 error: ec>
> self primitiveFailed
>
> I've attached these methods.
>
>
> What do you think ?
> >
>
> I think for Spur it makes sense to use 173/174.
>

+1

The 73/74 primitives (primitiveInstVarAt and primitiveInstVarAtPut) are semantically
different from the 173/174 primitives (primitiveSlotAt and primitiveSlotAtPut).

The 173/174 primitives are preferred for Spur images, and the 73/74 primitives are
used for V3 images. The VM can support both, and the image should call the primitive
that it needs or wants to use. In any case, the primitive table numbering should not
be confused between them.

Squeak has the correct implementation for calling the primitives from a Spur image.
See Cuis has an implementation of the image calling the appropriate primitive for
whatever image format it happens to be using.

Dave


Reply | Threaded
Open this post in threaded view
|

Re: Problem with primitive instVarAt: / instVarAt:put:, what should we do ?

Clément Béra
Hi,

So you're advocating for solution 1:
1) We change the instVarAt: / instVarAt:put: primitives comments in Pharo to match
 the slotAt: / slotAt:put: primitives behavior
 
I marked in bold the changes in the Pharo comment (so the comment now reflects the actual behavior):

instVarAt: index
"Primitive. Answer a fixed variable in an object. The numbering of the variables
corresponds to the named instance variables, followed by the indexed instance
variables. Fail if the index is not an Integer or is not the index of a fixed variable 
         or indexed variable. Essential. See Object documentation whatIsAPrimitive."

<primitive: 173 error: ec>
self primitiveFailed

I'll make a patch later on.

On Tue, Nov 28, 2017 at 2:55 AM, David T. Lewis <[hidden email]> wrote:
On Mon, Nov 27, 2017 at 12:11:13PM -0800, Eliot Miranda wrote:
> Hi Cl??ment,
>
> On Wed, Nov 22, 2017 at 5:35 AM, Cl??ment Bera <[hidden email]>
> wrote:
>
> > Hi All,
> >
> > It seems Pharo is lacking the slotAt: / slotAt:put: primitive and that the
> > instVarAt: / instVarAt:put: primitive is not doing what it says it does.
> > When I look at Object>>#instVarAt: I see:
> >
> > "[...]Answer a fixed variable in an object. The numbering of the variables
> > corresponds to the named instance variables, followed by the indexed
> > instance
> > variables. Fail if the index is not an Integer or is not the index of a
> > fixed variable[...]"
> >
> > Now in Pharo 7 when one does:
> >
> > #(iv1) instVarAt: 1, one gets #iv1.
> >
>
> But the failure code should indeed answer #iv1.
>
>
> > So the comment is wrong.
> >
> > The reason for this is that instVarAt: / instVarAt:put: should be
> > primitive 73/74, while they are in Pharo 173/174 (primitive
> > slotAt:/slotAt:put:).
> >
>
> Why do you say "should"?  The 173/174 primitives work well and avoid
> unnecessary primitive failures (which is why they're there).
>
>
> > The slot primitive access pointer slots of objects, whichever inst var or
> > variable fields. A good example would be:
> >
> > | t |
> > t := #temp.
> > [ t ] instVarAt: 4, one gets #temp
> >
> > So I have two solutions for this problem:
> > 1) We change the instVarAt: / instVarAt:put: primitives comments to match
> > the slotAt: / slotAt:put: behavior, and in Pharo instVarAt: /
> > instVarAt:put: means in factslotAt: / slotAt:put: .
> >
>
> This is the performant solution, because...
>
>
> > 2) We change the instVarAt: / instVarAt:put: primitives to primitive 73
> > and 74, we introduce slotAt: / slotAt:put: with primitive 173/174 and we
> > ask everyone to fix their code relying on instVarAt: / instVarAt:put:
> > acting as slotAt: / slotAt:put:
> >
>
> Early in Squeak Spur testing Chris Muller noticed a slow down in his code
> He was accessing an Array using instVarAt:, which was failing because he
> was using it to access an indexed field in an Array.  Now in Spur the VM
> must scan the object graph of a primitive's  receiver and arguments to some
> depth on primitive failure to follow forwarders.  Because a) at that time
> the VM computed a depth > 0 for instVarAt:[put:] it scanned the whole array
> on each access, and b) the Array was large, this meant that the failures
> were very expensive.  So I added the slot primitives, since the old 73/74
> primitive pair were only used in the instVarAt:[put:] methods that accessed
> indexable slots on failure.  So the new 173/174 pair subsumed the 73/74
> pair.
>
> I eventually added an accessorDepth: pragma rot specify the accessor depth
> manually when the automatically computed depth is too deep.  It is for
> at:[put:] because the String version of the at: 60 primitive appears to
> access the innards of non-immediate Characters (it doesn't do this in Spur,
> but the depth inferencer isn't clever enough to notice).  So primitiveAt
> looks like:
>
> primitiveAt
> <accessorDepth: 0>
> self commonAt: false
>
> So we could go back to using 73/74 for instVarAt:[put:] provided that their
> accessorDepth ends up being 0, which likely means they will require the
> pragma.  But I think the simplest thing is to keep using 173/174.  Here's
> the Squeak code:
>
> Object>>instVarAt: index
> "Primitive. Answer a fixed variable in an object. The numbering of the
> variables
> corresponds to the named instance variables, followed by the indexed
> instance
> variables. Fail if the index is not an Integer or is not the index of a
> fixed or
> indexed variable. Essential. See Object documentation whatIsAPrimitive."
>
> <primitive: 173 error: ec>
> self primitiveFailed
>
> Object>>instVarAt: index put: anObject
> "Primitive. Store a value into a fixed variable in an object. The numbering
> of the
> variables corresponds to the named instance variables, followed by the
> indexed
> instance variables. Fail if the index is not an Integer or is not the index
> of a fixed
> or indexed variable. Essential. See Object documentation whatIsAPrimitive."
>
> <primitive: 174 error: ec>
> self primitiveFailed
>
> I've attached these methods.
>
>
> What do you think ?
> >
>
> I think for Spur it makes sense to use 173/174.
>

+1

The 73/74 primitives (primitiveInstVarAt and primitiveInstVarAtPut) are semantically
different from the 173/174 primitives (primitiveSlotAt and primitiveSlotAtPut).

The 173/174 primitives are preferred for Spur images, and the 73/74 primitives are
used for V3 images. The VM can support both, and the image should call the primitive
that it needs or wants to use. In any case, the primitive table numbering should not
be confused between them.

Squeak has the correct implementation for calling the primitives from a Spur image.
See Cuis has an implementation of the image calling the appropriate primitive for
whatever image format it happens to be using.

Dave





--
Clément Béra
Pharo consortium engineer
Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq