Ruby Forum IronRuby > Friday Status Update

Like we talked about earlier, let's give an update on what we are 
working on

I'm working on hacking optparse to work so that we can update to the 
newest Rubinius spec's. Then I'll be closing out a whole slew of bugs.

Jim Deville

I'm still working on the shelveset that I sent out for code review 
earlier today (bugfixes9). We found a few more issues when we did the 
code review. The most interesting case is this one:

We normally define methods that accept symbol using a signature that 
takes a SymbolId, eg:


[RubyMethod("attr", RubyMethodAttributes.PrivateInstance)]
public static void Attr(CodeContext/*!*/ context, RubyModule/*!*/ self, 
SymbolId name) {
    DefineAccessor(context, self, name, true, false);
}


Methods like these will blow up if you pass it a 'string-like' thing. We 
have a Protocol to convert these things - Protocols.CastToSymbol(), 
which does the right thing, which is call to_str on the target.

We have an open question right now which is whether we should bake in 
object to SymbolId conversions in the binder or whether we should handle 
them in a case by case basis by defining an overload that accepts an 
Object, and calling Protocols.CastToSymbol() on that object.

Once I get this shelveset past the troll, I'll work next on getting 
socket.cs and the ironi regex library into the tree.

Thanks,
-John

From: ironruby-core-bounces@rubyforge.org 
[mailto:ironruby-core-bounces@rubyforge.org] On Behalf Of Jim Deville
Sent: Friday, May 02, 2008 2:37 PM
To: ironruby-core@rubyforge.org
Subject: [Ironruby-core] Friday Status Update

Like we talked about earlier, let's give an update on what we are 
working on

I'm working on hacking optparse to work so that we can update to the 
newest Rubinius spec's. Then I'll be closing out a whole slew of bugs.

Jim Deville

I've been looking at what it would take to start using byte arrays 
rather
then strings for the backing store for MutableString.  It's one heck of 
a
rabbit hole, I can get it /almost/ working... either I fail one spec, or
three... and they're totally different.  Anyway, if noone else is 
working on
it I'll continue to see what I can do, this is related to the zlib issue
with binary files (or actually any file that contains bytes over 127
potentially) as data is passed around as strings and not byte arrays.

Oh and I hope to implement actual compression sometime soon.. just need 
to
block out some time.

On Fri, May 2, 2008 at 6:24 PM, John Lam (IRONRUBY) 
<jflam@microsoft.com>

Michael Letterle:

> need to block out some time.
Tomas has spent quite some time looking at a dual byte-array + 
char-array MutableString implementation. It was sitting in his work 
queue to tackle after his current work on eval and blocks. Perhaps you 
could share some of your findings with him around the zlib integration 
effort so that we don't duplicate our efforts?

Thanks!
-John

I might be right off track here and since I haven't seen any of the 
detailed
discussion about the question below I might be being completely naive.

The open question should really be targeting all standard conversion
protocols in Ruby.

In my mind it seems that the way Ruby works generally is to have a load 
of
these "protocols" for type conversion, but that it is not enforced by 
the
language at all and that it is entirely up the library developer to 
follow
suit.  While this allows ultimate freedom for the developer it obviously
does allow inconsistency to creep in.
I would feel uncomfortable baking these protocols into the binder, 
unless
there was some way of overriding them in library code, since this is
perfectly possible in MRI.  For instance, it must be possible in at 
least
some cases to monkey patch code that by default uses the standard 
conversion
protocol so that it does something different (e.g. converts strings to
symbols in some other way???)

Clearly this is not a good thing to do in general but since Ruby allows 
it
as a language, it seems that it should be supported in IronRuby.

Happy to be completely wrong here.

Pete

Peter Bacon Darwin wrote:
> does allow inconsistency to creep in.
> I would feel uncomfortable baking these protocols into the binder, unless
> there was some way of overriding them in library code, since this is
> perfectly possible in MRI.  For instance, it must be possible in at least
> some cases to monkey patch code that by default uses the standard conversion
> protocol so that it does something different (e.g. converts strings to
> symbols in some other way???)
> 
> Clearly this is not a good thing to do in general but since Ruby allows it
> as a language, it seems that it should be supported in IronRuby.

FYI, there are a few cases in the core classes that use e.g to_s or to_a
instead of to_str or to_ary to coerce. I don't remember them off the top
of my head.

- Charlie

Yes, we have Protocols class that implements various flavors of 
conversions that are used in libraries.
The binder idea is that there would be an attribute you can apply on C# 
methods that would disable the default conversions implemented in 
binder. So if 90% of methods do the conversion one way their 
implementation could be simplified. The rest would be marked by the 
attribute.

Tomas

Tomas Matousek wrote:
> Yes, we have Protocols class that implements various flavors of conversions that are used in libraries.
> The binder idea is that there would be an attribute you can apply on C# methods that would disable the default conversions implemented in binder. So if 90% of methods do the conversion one way their implementation could be simplified. The rest would be marked by the attribute.

Nice, I'll have to add that. At the moment, everything generifies to a
common type, and the method itself calls the appropriate coercion
mechanism. But it would be nice to eliminate that.

- Charlie

Charles Oliver Nutter:

> > The binder idea is that there would be an attribute you can apply on
> C# methods that would disable the default conversions implemented in
> binder. So if 90% of methods do the conversion one way their
> implementation could be simplified. The rest would be marked by the
> attribute.
>
> Nice, I'll have to add that. At the moment, everything generifies to a
> common type, and the method itself calls the appropriate coercion
> mechanism. But it would be nice to eliminate that.

The other nice thing about this is that it would be trivial to generate 
a report on all of the exceptions. This way we can drive some meaningful 
design discussions about all of the exceptional cases and get them fixed 
in the libs.

-John