Syntax of GNU Smalltalk class definitions

I'm trying to understand the syntax of GNU Smalltalk class definitions.

https://www.gnu.org/software/smalltalk/manual/html_node/The-syntax.html
documents the BNF of the bang syntax:

    methods: ``!'' id [``class''] ``methodsFor:'' string
             ``!'' [method ``!'']+ ``!''

But for the "new style" syntax, I can't see how it fits.
From the BNF point of view, it looks pretty much like a method call, but
not quite.

More specifically I thought that in Smalltalk [...] is just a block (aka
closure).  So how does Smalltalk know that when I do

    Package extend [ ... ]

this [...] should not be a block (or is it still treated as a block,
except that "Package extend" then uses introspection to look inside the
body of the closure and analyze its "source code" to treat it as
something different)?

Better yet: my current understanding of Smalltalk syntax tells me that
"Package extend" is a method invocation with the unary selector
"extend", i.e. it's

    (Package extend) [ ... ]

so what is the "operation" that passes [ ... ] to (Package extend)?
I doesn't look like any of the three syntaxes of method invocation
I know (unary, binary, and multiple-keywords)?


        Stefan

_______________________________________________
help-smalltalk mailing list
[hidden email]
https://lists.gnu.org/mailman/listinfo/help-smalltalk

Stefan Monnier writes:
This [ ... ] is not a block closure and the "extend" is more like a
keyword to indicate this special construct.

Derek

_______________________________________________
help-smalltalk mailing list
[hidden email]
https://lists.gnu.org/mailman/listinfo/help-smalltalk

> This [ ... ] is not a block closure and the "extend" is more like a
> keyword to indicate this special construct.

"like"?  Is it a reserved keyword?  Does that mean you can't define
a method called `extend`?

What about the case of "C1 :subclass C2 [...]", is "subclass:" also
a reserved keyword?  Does it also prevent using method called
`:subclass`?

What about uses of `extend` or `subclass:` elsewhere than at top-level?

Also within such a "set of methods", I see two different method
definitions:

    <method-selector> [<method-body>]
   
and

    <class> class >> <method-selector> [<method-body>]
   
what happens if <class> is another class than the one for which this
block is defining methods?

why do we need both the `class` and the `>>` keywords to define those
class methods?

Are there other possible valid elements than those two?

Where is this documented (other than by showing examples)?


        Stefan

_______________________________________________
help-smalltalk mailing list
[hidden email]
https://lists.gnu.org/mailman/listinfo/help-smalltalk

Stefan Monnier writes:

>> This [ ... ] is not a block closure and the "extend" is more like a
>> keyword to indicate this special construct.
>
> "like"?  Is it a reserved keyword?  Does that mean you can't define
> a method called `extend`?
You can. However, under this circumstance (very top level) the parser does
treat it differently.
>
> What about the case of "C1 :subclass C2 [...]", is "subclass:" also
> a reserved keyword?  Does it also prevent using method called
> `:subclass`?
subclass: is a method call. However, under this circumstance (very top
level) the parser is also recognizing the special construct as the body
of the class.

So in gst3 syntax to protect against these traps, if you do
want to have loose statements on the top level you will need to wrap
them in a Eval [ ... ] block; which is again a special construct only
valid at the top level.

>
> What about uses of `extend` or `subclass:` elsewhere than at top-level?
Then they lose the special treatment. Theoretically you can; however I think it is not encouraged.
>
> why do we need both the `class` and the `>>` keywords to define those
> class methods?
Theoretically you can even write:
<class> >> <method-selector> [<method-body>]
where <class> is a different class. However it is also not encouraged
for obvious reason.
The >> seperates the owner of this method and the method itself. Without
it it looks really awkward.
The outer block in extend or subclass: only gives default owner of
methods.
>
> Are there other possible valid elements than those two?
Other than what have been menthoned, no. Actually you don't strictly
need the second form if you put all class methods in a:

<class> class extend [ ... ]

block. There is no special priviledge for method or class method defined
in the original subclass: block.
>
> Where is this documented (other than by showing examples)?
>
I was wondering the same :)

_______________________________________________
help-smalltalk mailing list
[hidden email]
https://lists.gnu.org/mailman/listinfo/help-smalltalk