[squeak-dev] Units package

I'm doing some test with the Units package and the results are puzzling me:

In a pristine 3.10.2-7179-basic image, I installed Units version 4 via
Universe.

I run the UnitsTest via SUnit Runner and all the 8 tests run green. So
far OK!

Now if I copy to a Workspace from the source of the UnitsTest>>testAdd
the following snippet:

u :=  (14 units: #feet) + (10 units: #meters).

If I print the whole expression I get '46.8083989501312 feet' as expected:

Now when I attempt to make any other thing with the "u" object, like
inspecting it, I get a stack, with the Title "Error: My subclass should
have overriden #hash.

Now, the tests run OK in the SUnit and fail in the Workspace! Even
attempting to use the object gets the same stack trace...

I probably doing something grossly wrong here as it seems to basic to
have eluded other users, but I cannot see what it is it!

Can you please advise?

--
Cesar Rabak
GNU/Linux User 52247.
Get counted: http://counter.li.org/

Hi

>
> u :=  (14 units: #feet) + (10 units: #meters).

same for me, if just evaluate this line twice... the parser use the
hash value...
a hash value has to be defined for Unit

I added in UnitValue (comparing):

hash

        ^value hash

and it seems to work... Units tests pass

hth

Cédrick

Hi!

> Now, the tests run OK in the SUnit and fail in the Workspace! Even
> attempting to use the object gets the same stack trace...
I met this problem recenty. It is something to do with, how the compiler
handles global variables. It involves a dictionary lookup, where your UnitValue
object is (part of) the key.

> Can you please advise?
Implement UnitValue >> #hash eg:

hash

        ^value hash bitXor: self class hash

Balázs

>
> hash
>
>        ^value hash bitXor: self class hash
>
There, (60 units: #seconds) hash = (60 units: #meters) returns true
I'm not sure this could lead to problem but what about:

hash
        ^value hash bitXor: unit hash

and

testEquality

        self deny: (60 units: #seconds) hash = (60 units: #meters) hash.
        self deny: (60 units: #seconds) = (60 units: #meters).
        self deny: (60 units: #seconds) hash = (1 units: #minutes) hash.
        self assert: (60 units: #seconds) = (1 units: #minutes).


Cédrick

Much better :)

Cheers, Balázs

ps: cc-d to Marcus Denker, because he seems to be maintainer of the package.

cdrick escreveu: Thanks Cédrick!

I figure out that I could circumvent this using the 'create' button in
the debugger, but I thought I must be doing something wrong (perhaps
missing some dependency?).

If my reading of the code is correct, we have to define hash and =
methods. . .

I'll get back to this next week when I follow other tests.

Thanks again,

--
Cesar Rabak
GNU/Linux User 52247.
Get counted: http://counter.li.org/

2008/6/19 Gabriel Cotelli <[hidden email]>:
> Hi,
> I believe is not a good idea to test de denial of two hashes...
> I two objects are equal then the hash must be the same, but te inverse is
> not true

I agree, I wrote this test to show the behavior (and I should have
left the failing test :) ). The hash function is probably not good.
the last one fails with the provided hash.

this hash seems ok:

UnitValue>>hash
|baseUnit|

        baseUnit := (self baseUnits).
  ^ baseUnit value hash bitXor: baseUnit unit hash


> Gabriel
>
Thanks ;)

Cédrick

Hello,