Smalltalk › Squeak › Squeak - Dev

Serious bug in 3.10

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

9 messages Options

askoh

Serious bug in 3.10

Administrator

| block |
block := [:i | Transcript cr; show: i].
1 to: 10 do: [:i | block value: i+1]

The above should print ten times. But it only prints the five even numbers. That is a bug isn't it?

Aik-Siong Koh

Ramon Leon-5

RE: [squeak-dev] Serious bug in 3.10

> Sent: Thursday, July 31, 2008 3:33 PM
> To: [hidden email]
> Subject: [squeak-dev] Serious bug in 3.10
>
>
> | block |
> block := [:i | Transcript cr; show: i].
> 1 to: 10 do: [:i | block value: i+1]
>
> The above should print ten times. But it only prints the five
> even numbers.
> That is a bug isn't it?
>
> Aik-Siong Koh

Try this..

block := [:f | Transcript cr; show: f].
(1 to: 10) do: [:f | block value: f+1]

And you'll see that likely, to:do: is being inlined by the compiler and
causing something funny like allowing you to increment the loops index var
causing it to skip a number each iteration.

Ramon Leon
http://onsmalltalk.com

keith1y

Re: [squeak-dev] Serious bug in 3.10

In reply to this post by askoh

askoh wrote:
> | block |
> block := [:i | Transcript cr; show: i].
> 1 to: 10 do: [:i | block value: i+1]
>
> The above should print ten times. But it only prints the five even numbers.
> That is a bug isn't it?
>
> Aik-Siong Koh
>
More of a "known limitation" than a bug... due to the fact that squeak
does not have proper closures.

try

| block |
block := [:j | Transcript cr; show: j].
1 to: 10 do: [:i | block value: i+1]

Keith

Ramon Leon-5

RE: [squeak-dev] Serious bug in 3.10

> askoh wrote:
> > | block |
> > block := [:i | Transcript cr; show: i].
> > 1 to: 10 do: [:i | block value: i+1]
> >
> > The above should print ten times. But it only prints the
> five even numbers.
> > That is a bug isn't it?
> >
> > Aik-Siong Koh
> >
> More of a "known limitation" than a bug... due to the fact
> that squeak
> does not have proper closures.
>
> try
>
> | block |
> block := [:j | Transcript cr; show: j].
> 1 to: 10 do: [:i | block value: i+1]
>
> Keith

I was going to say that, but why does it work using (1 to: 10) do: rather
than 1 to: 10 do:, seems related to #to:do: being inlined?

Ramon Leon
http://onsmalltalk.com

Igor Stasenko

Re: [squeak-dev] Serious bug in 3.10

2008/8/1 Ramon Leon <[hidden email]>:

>
>> askoh wrote:
>> > | block |
>> > block := [:i | Transcript cr; show: i].
>> > 1 to: 10 do: [:i | block value: i+1]
>> >
>> > The above should print ten times. But it only prints the
>> five even numbers.
>> > That is a bug isn't it?
>> >
>> > Aik-Siong Koh
>> >
>> More of a "known limitation" than a bug... due to the fact
>> that squeak
>> does not have proper closures.
>>
>> try
>>
>> | block |
>> block := [:j | Transcript cr; show: j].
>> 1 to: 10 do: [:i | block value: i+1]
>>
>> Keith
>
> I was going to say that, but why does it work using (1 to: 10) do: rather
> than 1 to: 10 do:, seems related to #to:do: being inlined?
>

#to:do: is a hacky compiler optimization, which replaces true messages. :)
Someday its can be removed, when VM will be powered by good JIT

> Ramon Leon
> http://onsmalltalk.com
>
>
>

--
Best regards,
Igor Stasenko AKA sig.

Eliot Miranda-2

Re: [squeak-dev] Serious bug in 3.10

In reply to this post by askoh

On Thu, Jul 31, 2008 at 3:32 PM, askoh <[hidden email]> wrote:

| block |
block := [:i | Transcript cr; show: i].
1 to: 10 do: [:i | block value: i+1]

The above should print ten times. But it only prints the five even numbers.
That is a bug isn't it?

yes, but its ancient. It is to do with the implementation of blocks and specifically block temporaries. Both blocks bind to the same "i". So when you evaluate

block value: i + 1

the 1 to: do:'s i gets incremented.

Try this with my closure implementation (http://www.mirandabanda.org/cogblog/downloads/) and it is fixed.

Aik-Siong Koh
--
View this message in context: http://www.nabble.com/Serious-bug-in-3.10-tp18764850p18764850.html
Sent from the Squeak - Dev mailing list archive at Nabble.com.

Eliot Miranda-2

Re: [squeak-dev] Serious bug in 3.10

In reply to this post by Ramon Leon-5

On Thu, Jul 31, 2008 at 3:54 PM, Ramon Leon <[hidden email]> wrote:

> askoh wrote:
> > | block |
> > block := [:i | Transcript cr; show: i].
> > 1 to: 10 do: [:i | block value: i+1]
> >
> > The above should print ten times. But it only prints the
> five even numbers.
> > That is a bug isn't it?
> >
> > Aik-Siong Koh
> >
> More of a "known limitation" than a bug... due to the fact
> that squeak
> does not have proper closures.
>
> try
>
> | block |
> block := [:j | Transcript cr; show: j].
> 1 to: 10 do: [:i | block value: i+1]
>
> Keith

I was going to say that, but why does it work using (1 to: 10) do: rather
than 1 to: 10 do:, seems related to #to:do: being inlined?

Yes. In the activation of to:do: there is a different counter. Look at the implementation of Number>>to:do:. So while both i's are still the same lvalue, i is not used by Number>>to:do: either to control the loop iterations or to to derive the values for the loop. Only when the loop is inlined does i become the loop control variable.

HTH

Ramon Leon
http://onsmalltalk.com

Nicolas Cellier-3

[squeak-dev] Re: Serious bug in 3.10

In reply to this post by Ramon Leon-5

Ramon Leon <ramon.leon <at> allresnet.com> writes:

>
> > Sent: Thursday, July 31, 2008 3:33 PM
> > To: squeak-dev <at> lists.squeakfoundation.org
> > Subject: [squeak-dev] Serious bug in 3.10
> >
> >
> > | block |
> > block := [:i | Transcript cr; show: i].
> > 1 to: 10 do: [:i | block value: i+1]
> >
> > The above should print ten times. But it only prints the five
> > even numbers.
> > That is a bug isn't it?
> >
> > Aik-Siong Koh
>
> Try this..
>
> block := [:f | Transcript cr; show: f].
> (1 to: 10) do: [:f | block value: f+1]
>
> And you'll see that likely, to:do: is being inlined by the compiler and
> causing something funny like allowing you to increment the loops index var
> causing it to skip a number each iteration.
>
> Ramon Leon
> http://onsmalltalk.com
>
>

Yes but:

block := [:f | Transcript cr; show: f].
(1 to: 10) do: [:f | block value: f+1. block value: f].

does not behave much better...
the 10 loops are performed, but since the blocks share same temporary f, things
are still a bit strange.

You need to use different argument names, NewCompiler or Eliot's closures.

Ramon Leon-5

RE: [squeak-dev] Re: Serious bug in 3.10

> >
> > And you'll see that likely, to:do: is being inlined by the
> compiler and
> > causing something funny like allowing you to increment the
> loops index var
> > causing it to skip a number each iteration.
> >
> > Ramon Leon
> > http://onsmalltalk.com
> >
> >
>
> Yes but:
>
> block := [:f | Transcript cr; show: f].
> (1 to: 10) do: [:f | block value: f+1. block value: f].
>
> does not behave much better...
> the 10 loops are performed, but since the blocks share same
> temporary f, things
> are still a bit strange.
>
> You need to use different argument names, NewCompiler or
> Eliot's closures.

Oh I agree, I wasn't suggesting a solution, I know the scope of the args is
shared because of the lack of real closures. I was more inquiring why the
behavioral difference between (to:) do: and to:do: since the latter is
inlined but normally you'd expect them to be equivalent.

Ramon Leon
http://onsmalltalk.com