World: r3wp

?

If Carl indeed builds from the same codebase, then a binary release 
is identical with a core-only hostkit release except for above polish.

Oh, I get it. We didn't have those comments in the first place (public 
codebase).

For instance, we only have a Windows build for now. A host-kit release 
includes the sources for all supported platforms, even if it doesn't 
include the binaries. And that source would need to be adapted to 
match the new system model to be included in a release. As of when 
last I checked, we hadn't adapted the Linux-specific stuff yet. No 
changes may be needed, but without test builds, internal testing 
of those builds, and debugging we can't say for sure that the extra 
code in the host kit release is of even alpha release quality, no 
matter if we mark it as experimental.

You clearly have not looked into the hostkit much.

A host-kit release includes the sources for all supported platforms


Yes, except that there are separate hostkit releases for each platform. 
In a given hostkit release, the sources for all platforms but the 
one matching the release are dysfunctional.

No, core works for me for now, and all of my contributions in the 
recent round have been mezzanine, and native testing. I'll start 
working on the higher-level host kit code more when the lower-level 
code I have been working on is done.

Maybe you want to make less bold assertions about the hostkit until 
you actually find time to look into it.

That said, let's not get the actual good news swallowed by this discussion:

R3 "Core" A108 for Windows is out! Rejoice!
http://www.rebol.com/r3/downloads.html

I have been following the discussions of the people designing and 
implementing it. And I have read the host code source.

Please redownload the Windows build. An obscure bug and a last-minute 
convenience fix were just added.

Rather, the obscure bug was fixed, not added :)

I have to navigate long paths sometimes ... in heavily nested objects.


How feasible would it be to use wild cards in the path  so instead 
of Obj/Problem/Description/Code/Value I could write Obj/P*/D*/C*/Value 
because I know that there are only one possible match?

so the * finds the first match and uses it ...

Should be easy using a simple custom dialect.

and as part of the language?

built in ... I mean

Makes more sense to draft this in a simple function first, imho.

Ok, ... can try it out

Not likely in direct syntax for two reasons:
- P* is a valid word itself.

- Word lookup is done by pointer comparison, not string comparison.

so I could use a function that traverses the object doing string 
comparisons and returns the full path for me so I don't have to keep 
typing it

walk-path 'Obj/P*/D*/C*/Value

Exactly.

Test results:
A107: Total: 3822 Succeeded: 3769 Failed: 53
A108: Total: 3829 Succeeded: 3750 Failed: 79

These your tests? And did you redownload the build?

The publicly available testsuite. Yes.

Ah, we should check the tests to see if they are affected by the 
changes in semantics in A108.

I am looking thru the regressions now.

Also, see if there are any new tests succeeding.

https://gist.github.com/9457aa5b41bfd0b7edcb

If anyone wants to join.

Are the red lines failed tests?

Also, # means none now.

The green lines are new failures.

The red lines are tests which no longer fail.

Yes, that's a case where the testsuite needs updating.

Nice.

The first few new regressions indicate that the "make error!" semantics 
have changed.

DISARM has finally been removed. It has never been necessary in R3.

Yes. Would maybe make sense to keep it around for a while to warn 
that it is now gone.

We did that for more than a year. Now it is gone.

It only had a deprecation notice, no hard error (like e.g. LOAD/next 
has now).

I personally don't care either way.

The deprecation notice was in CureCode, when someone complained that 
it didn't work. From early 2009.

Yeah, and in the docstring.

Just looked back and we made that change in 2008. Wow.

It predated CureCode :)

The `make image!` errors in lines 42-43 seem to be real regressions.

Seems complement no longer works on images and silently fails.

The test on line 51 is a bug. See bug#1471, fixed in a108.