Allright. But the problem happens for any literal you mutate, not only byte arrays. For example this one: foo #(true nil 10) at: 2 put: 50. ^ #(true nil 10) Answers... #(true 50 10) The most common problems are with Array, not ByteArray. Even worse, it can happen with LargeIntegers, Floats, ScaledDecimal, etc. Hopefully read-only objects will solve most of the problems. On Sun, Jul 10, 2016 at 10:51 AM, Peter Uhnák <i.uhnak@gmail.com> wrote:

Thanks for the explanation!

I've this byte array syntax for the first time and I don't think that I will be making heavy use of it, so it shouldn't be a too big problem — since now I understand the behavior, so I can keep it in mind.

Thanks, Peter

On Sun, Jul 10, 2016 at 7:43 AM, Clément Bera <bera.clement@gmail.com> wrote:

...

This is the normal behavior. You are mutating the method's literal, which becomes different. The instance is per method, so even if you do:

foo #[1 2 3] at: 2 put: 50. ^ #[1 2 3]

You get #[1 50 3], which is easy to understand in this example, but can be very tricky to understand in some cases.

I've introduced recently "read-only" objects (Object>>#isReadOnlyObject and co). I am waiting for the VM repo merge which should happen in the next few weeks to have this feature in the default Pharo VM, then I will be able to compile method literals "read-only" by default. With this feature, the #at:put: store will fail telling you that you cannot store into a read-only object, which will solve the common problem, when one does not mutate a literal on purpose.

Now as in Pharo everything is reflective, it will always be possible to mark back the literal as "writable", mutate it, and still have this behavior.

Best, Clement

On Sat, Jul 9, 2016 at 10:22 PM, Peter Uhnák <i.uhnak@gmail.com> wrote:

...

Is this normal behavior?

cls := Object subclass: #Something.

cls compile: 'array ^ #[1 2 3 4]'.

cls new array. "#[1 2 3 4]"

cls new array at: 1 put: 55.

cls new array. "#[55 2 3 4]"

(cls>>#array) sourceCode "'array ^ #[1 2 3 4]'"

So I can permanently change the compiled byte array in the method, which is weird.

Shouldn't the behavior be either

a) don't change the original byte array

or

b) change the source code

Thanks, Peter

Thanks, indeed! Doru

On Jul 10, 2016, at 7:28 PM, stepharo <stepharo@free.fr> wrote:

thanks for all the energy you put in getting non mutable objects. I will bring a nicer system.

Stef

Le 10/7/16 à 12:06, Clément Bera a écrit :

...

Allright. But the problem happens for any literal you mutate, not only byte arrays.

For example this one:

foo #(true nil 10) at: 2 put: 50. ^ #(true nil 10)

Answers... #(true 50 10)

The most common problems are with Array, not ByteArray.

Even worse, it can happen with LargeIntegers, Floats, ScaledDecimal, etc.

Hopefully read-only objects will solve most of the problems.

On Sun, Jul 10, 2016 at 10:51 AM, Peter Uhnák <i.uhnak@gmail.com> wrote: Thanks for the explanation!

I've this byte array syntax for the first time and I don't think that I will be making heavy use of it, so it shouldn't be a too big problem — since now I understand the behavior, so I can keep it in mind.

Thanks, Peter

On Sun, Jul 10, 2016 at 7:43 AM, Clément Bera <bera.clement@gmail.com> wrote: This is the normal behavior. You are mutating the method's literal, which becomes different. The instance is per method, so even if you do:

foo #[1 2 3] at: 2 put: 50. ^ #[1 2 3]

You get #[1 50 3], which is easy to understand in this example, but can be very tricky to understand in some cases.

I've introduced recently "read-only" objects (Object>>#isReadOnlyObject and co). I am waiting for the VM repo merge which should happen in the next few weeks to have this feature in the default Pharo VM, then I will be able to compile method literals "read-only" by default. With this feature, the #at:put: store will fail telling you that you cannot store into a read-only object, which will solve the common problem, when one does not mutate a literal on purpose.

Now as in Pharo everything is reflective, it will always be possible to mark back the literal as "writable", mutate it, and still have this behavior.

Best, Clement

On Sat, Jul 9, 2016 at 10:22 PM, Peter Uhnák <i.uhnak@gmail.com> wrote: Is this normal behavior?

cls := Object subclass: #Something.

cls compile: 'array ^ #[1 2 3 4]'.

cls new array. "#[1 2 3 4]"

cls new array at: 1 put: 55.

cls new array. "#[55 2 3 4]"

(cls>>#array) sourceCode "'array ^ #[1 2 3 4]'"

So I can permanently change the compiled byte array in the method, which is weird.

Shouldn't the behavior be either

a) don't change the original byte array

or

b) change the source code

Thanks, Peter

-- www.tudorgirba.com www.feenk.com "We cannot reach the flow of things unless we let go."