Smalltalk › Pharo › Pharo Smalltalk Developers

some patterns I would like to kill

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

11 messages Options

stephane ducasse

some patterns I would like to kill

Hi

I was reading Pluggable and friends
I identified some patterns.

The respondsTo plague
Examples:

color := self fillStyle asColor.
(self labelMorph respondsTo: #enabled:)
ifTrue: [self labelMorph enabled: self enabled].
(self labelMorph respondsTo: #interactionState:)
ifTrue: [self labelMorph interactionState: self interactionState]

(self labelMorph respondsTo: #enabled:) ifTrue: [
self labelMorph enabled: aBoolean].

(self enabled not and: [self label isMorph and: [(self label respondsTo: #enabled:) not]])

by construction a labelObject should be a morph and it should answer enabled: and the case where this is not the case should be fixed.

Using submorphs to avoid one single inst var makes code quite ugly to read:

labelMorph
"Answer the actual label morph."

self hasSubmorphs ifFalse: [^nil].
self firstSubmorph hasSubmorphs ifFalse: [^nil].
^self firstSubmorph firstSubmorph

_______________________________________________
Pharo-project mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

Schwab,Wilhelm K

Re: some patterns I would like to kill

Stef,

#respondsTo: - no argument. More defensive programming (aka masked bugs). Tests like this have their place, but are over-used in Squeak.

Using #submorphs *might* be easier to defend. Morphs were designed to be usable in large numbers, and one might argue (playing Devil's Advocate here) that lookups are cheaper than the additional gc load. An instance variable could also cache stale information. Of course, for something that is accessed frequently, an instance variable avoids the lookup and I agree that it is cleaner and easier to read.

Bill

________________________________________
From: [hidden email] [[hidden email]] On Behalf Of stephane ducasse [[hidden email]]
Sent: Thursday, October 21, 2010 4:47 PM
To: Pharo Development
Subject: [Pharo-project] some patterns I would like to **kill**

Hi

I was reading Pluggable and friends
I identified some patterns.

The respondsTo plague
Examples:

color := self fillStyle asColor.
(self labelMorph respondsTo: #enabled:)
ifTrue: [self labelMorph enabled: self enabled].
(self labelMorph respondsTo: #interactionState:)
ifTrue: [self labelMorph interactionState: self interactionState]

(self labelMorph respondsTo: #enabled:) ifTrue: [
self labelMorph enabled: aBoolean].

(self enabled not and: [self label isMorph and: [(self label respondsTo: #enabled:) not]])

by construction a labelObject should be a morph and it should answer enabled: and the case where this is not the case should be fixed.

Using submorphs to avoid one single inst var makes code quite ugly to read:

labelMorph
"Answer the actual label morph."

self hasSubmorphs ifFalse: [^nil].
self firstSubmorph hasSubmorphs ifFalse: [^nil].
^self firstSubmorph firstSubmorph

_______________________________________________
Pharo-project mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

_______________________________________________
Pharo-project mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

Stéphane Ducasse

Re: some patterns I would like to kill

In reply to this post by stephane ducasse

I think that if you have a container that contains morphs probably findA and other queries are useful.
Now for a button that is also at the same place a field is the best.

Stef

On Oct 22, 2010, at 11:14 PM, Schwab,Wilhelm K wrote:

> Stef,
>
> #respondsTo: - no argument. More defensive programming (aka masked bugs). Tests like this have their place, but are over-used in Squeak.
>
> Using #submorphs *might* be easier to defend. Morphs were designed to be usable in large numbers, and one might argue (playing Devil's Advocate here) that lookups are cheaper than the additional gc load. An instance variable could also cache stale information. Of course, for something that is accessed frequently, an instance variable avoids the lookup and I agree that it is cleaner and easier to read.
>
> Bill
>
>
>
> ________________________________________
> From: [hidden email] [[hidden email]] On Behalf Of stephane ducasse [[hidden email]]
> Sent: Thursday, October 21, 2010 4:47 PM
> To: Pharo Development
> Subject: [Pharo-project] some patterns I would like to **kill**
>
> Hi
>
> I was reading Pluggable and friends
> I identified some patterns.
>
>
> The respondsTo plague
> Examples:
>
> color := self fillStyle asColor.
> (self labelMorph respondsTo: #enabled:)
> ifTrue: [self labelMorph enabled: self enabled].
> (self labelMorph respondsTo: #interactionState:)
> ifTrue: [self labelMorph interactionState: self interactionState]
>
> (self labelMorph respondsTo: #enabled:) ifTrue: [
> self labelMorph enabled: aBoolean].
>
> (self enabled not and: [self label isMorph and: [(self label respondsTo: #enabled:) not]])
>
>
> by construction a labelObject should be a morph and it should answer enabled: and the case where this is not the case should be fixed.
>
>
> Using submorphs to avoid one single inst var makes code quite ugly to read:
>
>
> labelMorph
> "Answer the actual label morph."
>
> self hasSubmorphs ifFalse: [^nil].
> self firstSubmorph hasSubmorphs ifFalse: [^nil].
> ^self firstSubmorph firstSubmorph
>
>
>
>
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

_______________________________________________
Pharo-project mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

Igor Stasenko

Re: some patterns I would like to kill

On 23 October 2010 14:33, Stéphane Ducasse <[hidden email]> wrote:
> I think that if you have a container that contains morphs probably findA and other queries are useful.
> Now for a button that is also at the same place a field is the best.
>

<sarcasm>
if morph having a random submorphs, inserted out of nowhere, just for
fun to see what happen. then yes.. findA is userful, as well as
respondsTo:
</sarcasm>

but if we are talking about specific morphs, like Pluggable and
friends, who serving a certain purpose (building UI from spec),
where no place for 'random', then this truly a bad design.

> Stef
>
>
> On Oct 22, 2010, at 11:14 PM, Schwab,Wilhelm K wrote:
>
>> Stef,
>>
>> #respondsTo: - no argument. More defensive programming (aka masked bugs). Tests like this have their place, but are over-used in Squeak.
>>
>> Using #submorphs *might* be easier to defend. Morphs were designed to be usable in large numbers, and one might argue (playing Devil's Advocate here) that lookups are cheaper than the additional gc load. An instance variable could also cache stale information. Of course, for something that is accessed frequently, an instance variable avoids the lookup and I agree that it is cleaner and easier to read.
>>
>> Bill
>>
>>
>>
>> ________________________________________
>> From: [hidden email] [[hidden email]] On Behalf Of stephane ducasse [[hidden email]]
>> Sent: Thursday, October 21, 2010 4:47 PM
>> To: Pharo Development
>> Subject: [Pharo-project] some patterns I would like to **kill**
>>
>> Hi
>>
>> I was reading Pluggable and friends
>> I identified some patterns.
>>
>>
>> The respondsTo plague
>> Examples:
>>
>> color := self fillStyle asColor.
>> (self labelMorph respondsTo: #enabled:)
>> ifTrue: [self labelMorph enabled: self enabled].
>> (self labelMorph respondsTo: #interactionState:)
>> ifTrue: [self labelMorph interactionState: self interactionState]
>>
>> (self labelMorph respondsTo: #enabled:) ifTrue: [
>> self labelMorph enabled: aBoolean].
>>
>> (self enabled not and: [self label isMorph and: [(self label respondsTo: #enabled:) not]])
>>
>>
>> by construction a labelObject should be a morph and it should answer enabled: and the case where this is not the case should be fixed.
>>
>>
>> Using submorphs to avoid one single inst var makes code quite ugly to read:
>>
>>
>> labelMorph
>> "Answer the actual label morph."
>>
>> self hasSubmorphs ifFalse: [^nil].
>> self firstSubmorph hasSubmorphs ifFalse: [^nil].
>> ^self firstSubmorph firstSubmorph
>>
>>
>>
>>
>>
>> _______________________________________________
>> Pharo-project mailing list
>> [hidden email]
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>>
>> _______________________________________________
>> Pharo-project mailing list
>> [hidden email]
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>

--
Best regards,
Igor Stasenko AKA sig.

_______________________________________________
Pharo-project mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

Stéphane Ducasse

Re: some patterns I would like to kill

In reply to this post by Stéphane Ducasse

On Oct 23, 2010, at 1:58 PM, Igor Stasenko wrote:

> On 23 October 2010 14:33, Stéphane Ducasse <[hidden email]> wrote:
>> I think that if you have a container that contains morphs probably findA and other queries are useful.
>> Now for a button that is also at the same place a field is the best.
>>
>
> <sarcasm>
> if morph having a random submorphs, inserted out of nowhere, just for
> fun to see what happen. then yes.. findA is userful, as well as
> respondsTo:
> </sarcasm>

I like your sarcasm, we are all in sync on that
Open a 3.8 or 3.9 and check findA:

> but if we are talking about specific morphs, like Pluggable and
> friends, who serving a certain purpose (building UI from spec),
> where no place for 'random', then this truly a bad design.

:)

>
>> Stef
>>
>>
>> On Oct 22, 2010, at 11:14 PM, Schwab,Wilhelm K wrote:
>>
>>> Stef,
>>>
>>> #respondsTo: - no argument. More defensive programming (aka masked bugs). Tests like this have their place, but are over-used in Squeak.
>>>
>>> Using #submorphs *might* be easier to defend. Morphs were designed to be usable in large numbers, and one might argue (playing Devil's Advocate here) that lookups are cheaper than the additional gc load. An instance variable could also cache stale information. Of course, for something that is accessed frequently, an instance variable avoids the lookup and I agree that it is cleaner and easier to read.
>>>
>>> Bill
>>>
>>>
>>>
>>> ________________________________________
>>> From: [hidden email] [[hidden email]] On Behalf Of stephane ducasse [[hidden email]]
>>> Sent: Thursday, October 21, 2010 4:47 PM
>>> To: Pharo Development
>>> Subject: [Pharo-project] some patterns I would like to **kill**
>>>
>>> Hi
>>>
>>> I was reading Pluggable and friends
>>> I identified some patterns.
>>>
>>>
>>> The respondsTo plague
>>> Examples:
>>>
>>> color := self fillStyle asColor.
>>> (self labelMorph respondsTo: #enabled:)
>>> ifTrue: [self labelMorph enabled: self enabled].
>>> (self labelMorph respondsTo: #interactionState:)
>>> ifTrue: [self labelMorph interactionState: self interactionState]
>>>
>>> (self labelMorph respondsTo: #enabled:) ifTrue: [
>>> self labelMorph enabled: aBoolean].
>>>
>>> (self enabled not and: [self label isMorph and: [(self label respondsTo: #enabled:) not]])
>>>
>>>
>>> by construction a labelObject should be a morph and it should answer enabled: and the case where this is not the case should be fixed.
>>>
>>>
>>> Using submorphs to avoid one single inst var makes code quite ugly to read:
>>>
>>>
>>> labelMorph
>>> "Answer the actual label morph."
>>>
>>> self hasSubmorphs ifFalse: [^nil].
>>> self firstSubmorph hasSubmorphs ifFalse: [^nil].
>>> ^self firstSubmorph firstSubmorph
>>>
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> Pharo-project mailing list
>>> [hidden email]
>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>>>
>>> _______________________________________________
>>> Pharo-project mailing list
>>> [hidden email]
>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>>
>>
>> _______________________________________________
>> Pharo-project mailing list
>> [hidden email]
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>>
>
>
>
> --
> Best regards,
> Igor Stasenko AKA sig.
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

_______________________________________________
Pharo-project mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

Igor Stasenko

Re: some patterns I would like to kill

On 23 October 2010 15:04, Stéphane Ducasse <[hidden email]> wrote:

>
> On Oct 23, 2010, at 1:58 PM, Igor Stasenko wrote:
>
>> On 23 October 2010 14:33, Stéphane Ducasse <[hidden email]> wrote:
>>> I think that if you have a container that contains morphs probably findA and other queries are useful.
>>> Now for a button that is also at the same place a field is the best.
>>>
>>
>> <sarcasm>
>> if morph having a random submorphs, inserted out of nowhere, just for
>> fun to see what happen. then yes.. findA is userful, as well as
>> respondsTo:
>> </sarcasm>
>
> I like your sarcasm, we are all in sync on that
> Open a 3.8 or 3.9 and check findA:
>

Actually this is a serious problem: how to keep flexibility, but at same time
enforce some specific rules over 'random' cloud of data.

My thought about it, that if some morph expects some hierarchical organization
from submorphs (like label in button morph), then it should explicitly
state that it expects
a certain protocol to be supported by its submorphs, like #enabled: #contents:
etc.
So, then it could send these messages without any checks, and if DNU happens,
then its a responsibility of those who made a mistake by putting wrong
things inside
a button.

Another idea is to use a button as a 'decorator' to another morph,
called 'label'.
A label could have any number of arbitrary submorphs, but button could
have only single submorph,
up to the point that it could override #submorphs with:
submorphs
^ { label }

and dispatch #addSubmorph to a label:

addSubmorph: aMorph
^ label addSubmorph: aMorph

then it having total control over its contents (only single submorph,
known that it understand a certain protocol),
and everyone is happy:
- user could put anything inside button
- developer can be sure it will work without any problems, whatever
you may put in it

There could be other ways to solve that.

>> but if we are talking about specific morphs, like Pluggable and
>> friends, who serving a certain purpose (building UI from spec),
>> where no place for 'random', then this truly a bad design.
>
> :)
>>
>>> Stef
>>>
>>>
>>> On Oct 22, 2010, at 11:14 PM, Schwab,Wilhelm K wrote:
>>>
>>>> Stef,
>>>>
>>>> #respondsTo: - no argument. More defensive programming (aka masked bugs). Tests like this have their place, but are over-used in Squeak.
>>>>
>>>> Using #submorphs *might* be easier to defend. Morphs were designed to be usable in large numbers, and one might argue (playing Devil's Advocate here) that lookups are cheaper than the additional gc load. An instance variable could also cache stale information. Of course, for something that is accessed frequently, an instance variable avoids the lookup and I agree that it is cleaner and easier to read.
>>>>
>>>> Bill
>>>>
>>>>
>>>>
>>>> ________________________________________
>>>> From: [hidden email] [[hidden email]] On Behalf Of stephane ducasse [[hidden email]]
>>>> Sent: Thursday, October 21, 2010 4:47 PM
>>>> To: Pharo Development
>>>> Subject: [Pharo-project] some patterns I would like to **kill**
>>>>
>>>> Hi
>>>>
>>>> I was reading Pluggable and friends
>>>> I identified some patterns.
>>>>
>>>>
>>>> The respondsTo plague
>>>> Examples:
>>>>
>>>> color := self fillStyle asColor.
>>>> (self labelMorph respondsTo: #enabled:)
>>>> ifTrue: [self labelMorph enabled: self enabled].
>>>> (self labelMorph respondsTo: #interactionState:)
>>>> ifTrue: [self labelMorph interactionState: self interactionState]
>>>>
>>>> (self labelMorph respondsTo: #enabled:) ifTrue: [
>>>> self labelMorph enabled: aBoolean].
>>>>
>>>> (self enabled not and: [self label isMorph and: [(self label respondsTo: #enabled:) not]])
>>>>
>>>>
>>>> by construction a labelObject should be a morph and it should answer enabled: and the case where this is not the case should be fixed.
>>>>
>>>>
>>>> Using submorphs to avoid one single inst var makes code quite ugly to read:
>>>>
>>>>
>>>> labelMorph
>>>> "Answer the actual label morph."
>>>>
>>>> self hasSubmorphs ifFalse: [^nil].
>>>> self firstSubmorph hasSubmorphs ifFalse: [^nil].
>>>> ^self firstSubmorph firstSubmorph
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> Pharo-project mailing list
>>>> [hidden email]
>>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>>>>
>>>> _______________________________________________
>>>> Pharo-project mailing list
>>>> [hidden email]
>>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>>>
>>>
>>> _______________________________________________
>>> Pharo-project mailing list
>>> [hidden email]
>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>>>
>>
>>
>>
>> --
>> Best regards,
>> Igor Stasenko AKA sig.
>>
>> _______________________________________________
>> Pharo-project mailing list
>> [hidden email]
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>

--
Best regards,
Igor Stasenko AKA sig.

_______________________________________________
Pharo-project mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

Schwab,Wilhelm K

Re: some patterns I would like to kill

In reply to this post by Stéphane Ducasse

Stef,

Of course. But in terms of avoiding unintended consequences, just consider the points I raised. Some of the avoidance of instance variables (extensions, etc.) might have been a good idea with the gc at the time but is now an ugly hack, or that sparse storage tricks were over-used.

Bill

________________________________________
From: [hidden email] [[hidden email]] On Behalf Of Stéphane Ducasse [[hidden email]]
Sent: Saturday, October 23, 2010 7:33 AM
To: [hidden email]
Subject: Re: [Pharo-project] some patterns I would like to **kill**

I think that if you have a container that contains morphs probably findA and other queries are useful.
Now for a button that is also at the same place a field is the best.

Stef

On Oct 22, 2010, at 11:14 PM, Schwab,Wilhelm K wrote:

_______________________________________________
Pharo-project mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

_______________________________________________
Pharo-project mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

Igor Stasenko

Re: some patterns I would like to kill

On 23 October 2010 17:13, Schwab,Wilhelm K <[hidden email]> wrote:
> Stef,
>
> Of course. But in terms of avoiding unintended consequences, just consider the points I raised. Some of the avoidance of instance variables (extensions, etc.) might have been a good idea with the gc at the time but is now an ugly hack, or that sparse storage tricks were over-used.
>
Well, one way or another you have to tame the complexity somehow.
Be it extra instvar, or putting it into extensions doesn't changes the
model's concept: you need a reference to some object
which is responsible for some specific stuff.

> Bill
>
>
>
>
> ________________________________________
> From: [hidden email] [[hidden email]] On Behalf Of Stéphane Ducasse [[hidden email]]
> Sent: Saturday, October 23, 2010 7:33 AM
> To: [hidden email]
> Subject: Re: [Pharo-project] some patterns I would like to **kill**
>
> I think that if you have a container that contains morphs probably findA and other queries are useful.
> Now for a button that is also at the same place a field is the best.
>
> Stef
>
>
> On Oct 22, 2010, at 11:14 PM, Schwab,Wilhelm K wrote:
>
>> Stef,
>>
>> #respondsTo: - no argument. More defensive programming (aka masked bugs). Tests like this have their place, but are over-used in Squeak.
>>
>> Using #submorphs *might* be easier to defend. Morphs were designed to be usable in large numbers, and one might argue (playing Devil's Advocate here) that lookups are cheaper than the additional gc load. An instance variable could also cache stale information. Of course, for something that is accessed frequently, an instance variable avoids the lookup and I agree that it is cleaner and easier to read.
>>
>> Bill
>>
>>
>>
>> ________________________________________
>> From: [hidden email] [[hidden email]] On Behalf Of stephane ducasse [[hidden email]]
>> Sent: Thursday, October 21, 2010 4:47 PM
>> To: Pharo Development
>> Subject: [Pharo-project] some patterns I would like to **kill**
>>
>> Hi
>>
>> I was reading Pluggable and friends
>> I identified some patterns.
>>
>>
>> The respondsTo plague
>> Examples:
>>
>> color := self fillStyle asColor.
>> (self labelMorph respondsTo: #enabled:)
>> ifTrue: [self labelMorph enabled: self enabled].
>> (self labelMorph respondsTo: #interactionState:)
>> ifTrue: [self labelMorph interactionState: self interactionState]
>>
>> (self labelMorph respondsTo: #enabled:) ifTrue: [
>> self labelMorph enabled: aBoolean].
>>
>> (self enabled not and: [self label isMorph and: [(self label respondsTo: #enabled:) not]])
>>
>>
>> by construction a labelObject should be a morph and it should answer enabled: and the case where this is not the case should be fixed.
>>
>>
>> Using submorphs to avoid one single inst var makes code quite ugly to read:
>>
>>
>> labelMorph
>> "Answer the actual label morph."
>>
>> self hasSubmorphs ifFalse: [^nil].
>> self firstSubmorph hasSubmorphs ifFalse: [^nil].
>> ^self firstSubmorph firstSubmorph
>>
>>
>>
>>
>>
>> _______________________________________________
>> Pharo-project mailing list
>> [hidden email]
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>>
>> _______________________________________________
>> Pharo-project mailing list
>> [hidden email]
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>

--
Best regards,
Igor Stasenko AKA sig.

_______________________________________________
Pharo-project mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

Schwab,Wilhelm K

Re: some patterns I would like to kill

Hey, I want some credit here - for coming to Squeak's possible defense :)

All I am saying is that there is a possible space/speed tradeoff with a twist. There was a fair amount written about these decisions, which is why I know anything at all about them. Adding an instance variable for a rarely used sub-morph of a widely-instantiated morph might be an opportunity to fail to learn from history. Most likely, the patterns Stef has noticed are simply mis-uses of sparse storage. The best way to proceed would be to find an example of it that has many instances in the image, profile, fix, and profile again to see if the fix hurts performance (or perhaps helps it). The usual caveats about benchmarking being really hard to do well apply, of course.

Bill

________________________________________
From: [hidden email] [[hidden email]] On Behalf Of Igor Stasenko [[hidden email]]
Sent: Saturday, October 23, 2010 11:00 AM
To: [hidden email]
Subject: Re: [Pharo-project] some patterns I would like to **kill**

On 23 October 2010 17:13, Schwab,Wilhelm K <[hidden email]> wrote:
> Stef,
>
> Of course. But in terms of avoiding unintended consequences, just consider the points I raised. Some of the avoidance of instance variables (extensions, etc.) might have been a good idea with the gc at the time but is now an ugly hack, or that sparse storage tricks were over-used.
>
Well, one way or another you have to tame the complexity somehow.
Be it extra instvar, or putting it into extensions doesn't changes the
model's concept: you need a reference to some object
which is responsible for some specific stuff.

--
Best regards,
Igor Stasenko AKA sig.

_______________________________________________
Pharo-project mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

_______________________________________________
Pharo-project mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

Igor Stasenko

Re: some patterns I would like to kill

On 23 October 2010 21:47, Schwab,Wilhelm K <[hidden email]> wrote:
> Hey, I want some credit here - for coming to Squeak's possible defense :)
>
> All I am saying is that there is a possible space/speed tradeoff with a twist. There was a fair amount written about these decisions, which is why I know anything at all about them. Adding an instance variable for a rarely used sub-morph of a widely-instantiated morph might be an opportunity to fail to learn from history. Most likely, the patterns Stef has noticed are simply mis-uses of sparse storage. The best way to proceed would be to find an example of it that has many instances in the image, profile, fix, and profile again to see if the fix hurts performance (or perhaps helps it). The usual caveats about benchmarking being really hard to do well apply, of course.
>

I did some experiment(s) with Morphic, placing many morph aspects into
properties dictioary (such as color, border, borderwidth etc), and
these values are used in rendering all the time.
And i didn't noticed heavy speed degradation.

> Bill
>
>
> ________________________________________
> From: [hidden email] [[hidden email]] On Behalf Of Igor Stasenko [[hidden email]]
> Sent: Saturday, October 23, 2010 11:00 AM
> To: [hidden email]
> Subject: Re: [Pharo-project] some patterns I would like to **kill**
>
> On 23 October 2010 17:13, Schwab,Wilhelm K <[hidden email]> wrote:
>> Stef,
>>
>> Of course. But in terms of avoiding unintended consequences, just consider the points I raised. Some of the avoidance of instance variables (extensions, etc.) might have been a good idea with the gc at the time but is now an ugly hack, or that sparse storage tricks were over-used.
>>
> Well, one way or another you have to tame the complexity somehow.
> Be it extra instvar, or putting it into extensions doesn't changes the
> model's concept: you need a reference to some object
> which is responsible for some specific stuff.
>
>
>> Bill
>>
>>
>>
>>
>> ________________________________________
>> From: [hidden email] [[hidden email]] On Behalf Of Stéphane Ducasse [[hidden email]]
>> Sent: Saturday, October 23, 2010 7:33 AM
>> To: [hidden email]
>> Subject: Re: [Pharo-project] some patterns I would like to **kill**
>>
>> I think that if you have a container that contains morphs probably findA and other queries are useful.
>> Now for a button that is also at the same place a field is the best.
>>
>> Stef
>>
>>
>> On Oct 22, 2010, at 11:14 PM, Schwab,Wilhelm K wrote:
>>
>>> Stef,
>>>
>>> #respondsTo: - no argument. More defensive programming (aka masked bugs). Tests like this have their place, but are over-used in Squeak.
>>>
>>> Using #submorphs *might* be easier to defend. Morphs were designed to be usable in large numbers, and one might argue (playing Devil's Advocate here) that lookups are cheaper than the additional gc load. An instance variable could also cache stale information. Of course, for something that is accessed frequently, an instance variable avoids the lookup and I agree that it is cleaner and easier to read.
>>>
>>> Bill
>>>
>>>
>>>
>>> ________________________________________
>>> From: [hidden email] [[hidden email]] On Behalf Of stephane ducasse [[hidden email]]
>>> Sent: Thursday, October 21, 2010 4:47 PM
>>> To: Pharo Development
>>> Subject: [Pharo-project] some patterns I would like to **kill**
>>>
>>> Hi
>>>
>>> I was reading Pluggable and friends
>>> I identified some patterns.
>>>
>>>
>>> The respondsTo plague
>>> Examples:
>>>
>>> color := self fillStyle asColor.
>>> (self labelMorph respondsTo: #enabled:)
>>> ifTrue: [self labelMorph enabled: self enabled].
>>> (self labelMorph respondsTo: #interactionState:)
>>> ifTrue: [self labelMorph interactionState: self interactionState]
>>>
>>> (self labelMorph respondsTo: #enabled:) ifTrue: [
>>> self labelMorph enabled: aBoolean].
>>>
>>> (self enabled not and: [self label isMorph and: [(self label respondsTo: #enabled:) not]])
>>>
>>>
>>> by construction a labelObject should be a morph and it should answer enabled: and the case where this is not the case should be fixed.
>>>
>>>
>>> Using submorphs to avoid one single inst var makes code quite ugly to read:
>>>
>>>
>>> labelMorph
>>> "Answer the actual label morph."
>>>
>>> self hasSubmorphs ifFalse: [^nil].
>>> self firstSubmorph hasSubmorphs ifFalse: [^nil].
>>> ^self firstSubmorph firstSubmorph
>>>
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> Pharo-project mailing list
>>> [hidden email]
>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>>>
>>> _______________________________________________
>>> Pharo-project mailing list
>>> [hidden email]
>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>>
>>
>> _______________________________________________
>> Pharo-project mailing list
>> [hidden email]
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>>
>> _______________________________________________
>> Pharo-project mailing list
>> [hidden email]
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>>
>
>
>
> --
> Best regards,
> Igor Stasenko AKA sig.
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>

--
Best regards,
Igor Stasenko AKA sig.

_______________________________________________
Pharo-project mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

Schwab,Wilhelm K

Re: some patterns I would like to kill

If I understand you, you took things normally stored in instance variables, and (redundantly?) stored them as properties, right?

________________________________________
From: [hidden email] [[hidden email]] On Behalf Of Igor Stasenko [[hidden email]]
Sent: Saturday, October 23, 2010 10:47 PM
To: [hidden email]
Subject: Re: [Pharo-project] some patterns I would like to **kill**

On 23 October 2010 21:47, Schwab,Wilhelm K <[hidden email]> wrote:
> Hey, I want some credit here - for coming to Squeak's possible defense :)
>
> All I am saying is that there is a possible space/speed tradeoff with a twist. There was a fair amount written about these decisions, which is why I know anything at all about them. Adding an instance variable for a rarely used sub-morph of a widely-instantiated morph might be an opportunity to fail to learn from history. Most likely, the patterns Stef has noticed are simply mis-uses of sparse storage. The best way to proceed would be to find an example of it that has many instances in the image, profile, fix, and profile again to see if the fix hurts performance (or perhaps helps it). The usual caveats about benchmarking being really hard to do well apply, of course.
>

I did some experiment(s) with Morphic, placing many morph aspects into
properties dictioary (such as color, border, borderwidth etc), and
these values are used in rendering all the time.
And i didn't noticed heavy speed degradation.

some patterns I would like to **kill**

some patterns I would like to **kill**

Re: some patterns I would like to **kill**

Re: some patterns I would like to **kill**

Re: some patterns I would like to **kill**

Re: some patterns I would like to **kill**

Re: some patterns I would like to **kill**

Re: some patterns I would like to **kill**

Re: some patterns I would like to **kill**

Re: some patterns I would like to **kill**

Re: some patterns I would like to **kill**

Re: some patterns I would like to **kill**

some patterns I would like to kill

some patterns I would like to kill

Re: some patterns I would like to kill

Re: some patterns I would like to kill

Re: some patterns I would like to kill

Re: some patterns I would like to kill

Re: some patterns I would like to kill

Re: some patterns I would like to kill

Re: some patterns I would like to kill

Re: some patterns I would like to kill

Re: some patterns I would like to kill

Re: some patterns I would like to kill