[M3devel] A question for our language lawyers

[Daniel Alejandro Benavides D.]

Hi all:
I think if we are to type define initialization, we need a kernel to type more fun than rigid Modula-3 semantics:
http://www.hpl.hp.com/techreports/Compaq-DEC/SRC-RR-95.pdf

That said, we can define a m3kernel sort of type minimal abstraction of a Modula-3 Object, and built on top of that. Advantages are we can type theorize  in every wanted way with it and still protect us from incompatible type systems, by branding the type system to allow smooth transitions. Besides parallelization implicitly in the abstract machine (kernel) and check the type safety of it.
Also rewrite the type system in terms of this kernel might get us to a new language in the sense of a language definition smoothly

If someone steems this good I can make my try.

Thanks in advance

--- El vie, 6/7/12, Daniel Alejandro Benavides D. <dabenavidesd at yahoo.es> escribió:

De: Daniel Alejandro Benavides D. <dabenavidesd at yahoo.es>
Asunto: Re: [M3devel] A question for our language lawyers
Para: m3devel at elegosoft.com, "Rodney M. Bates" <rodney_bates at lcwb.coop>
Fecha: viernes, 6 de julio, 2012 14:17

Hi all:
English men say array is a sequence of elements (of a common type), and a BOOLEAN is an enumeration so you might attack that distinction to define what is an initialized boolean or array of boolean in common compilers, gcc javac, etc, which if is java-like is really undefined:
 
http://www.drdobbs.com/architecture-and-design/the-humble-boolean-deserves-help/232900836?cid=DDJ_nl_upd_2012-05-02_h&elq=cef656ee4d6c4bca996b337620b98f85

So I prefer non-uniform rules for records different of Sets, Arrays, and records as that, note that NEW expression doesn't allow constructors to be used, so the only thing you can use is array of uninitialized variables (but current gcc or javac, etc are really wrong in that)

This means we need to address this by either a native backend (NT386) or by another language for that matter.

Thanks in
 advance for any comments you may have

--- El vie, 6/7/12, Rodney M. Bates <rodney_bates at lcwb.coop> escribió:

De: Rodney M. Bates <rodney_bates at lcwb.coop>
Asunto: Re: [M3devel] A question for our language lawyers
Para: m3devel at elegosoft.com
Fecha: viernes, 6 de julio, 2012 13:27



On 07/06/2012 04:23 AM, Dirk Muysers wrote:
> The report says (2.6.9)
> "The values in the array will be arbitrary values of their type."

> Now, ParseParams in its "init" method allocates an array of BOOLEANs
> and relies on the fact that it is supposedly initialised with FALSE values.

> At the other hand the report says (2.2.4)
> "The constant |default| is a default value used when a record is constructed or allocated"

> If I allocate an array
 of records, which statement is stronger:
> - the array contains arbitray record values ?
> - the array record fields will be initialised to their default values?

Admittedly unclearly if not misleadingly worded.  Better wording might be
to say each element is initialized as it would if it were a scalar variable
of its type.

I think the way to interpret this is that the array itself does not impose
any initialization, but this fact will not eliminate initialization
imposed by other rules, specifically, the type of the array's elements.

This is a language quirk that I have always been deeply ambivalent about.
The type safety would go down the drain if variables were not initialized
to a bit pattern that represents some value of the type, so we have to pay
the performance penalty of executing initialization code.  So why not define
which value of the type is initialized-to and get behavioral
 predictability
for free?  And further save redundant initialization in the likely event
that the compiler's chosen arbitrary value happens to match what the
programmer wants?

(OK, a smart enough optimizer might figure this out, but we could have
had it even with a naive compiler.)

The contrary case is a type whose compiler-chosen representation happens
to use every bit pattern in the allocated space for a value of the type.
Here, no compiler-generated runtime initialization is needed.

Also, the rule we have might sometimes encourage programmers to at least give a
millisecond's thought to whether they need to do some explicit initialization.


> The ParseParams "init" method is obviously erroneous and works only
> by virtue of a happy combination of circumstances.
> But how is the report to be interpreted in the second case?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://m3lists.elegosoft.com/pipermail/m3devel/attachments/20120706/d9dd7a28/attachment-0002.html>