Log of the #nice channel on irc.freenode.net

Using timezone: Central European Time
<CIA-3>03bonniot * 10Nice/src/ (3 files in 2 dirs): 03:55
Remove nice.tools.code.ClassLoader (which was used to load java.util.* classes
from a version with type parameter). For now, all imported classes are loaded
using reflexion.
* bonniot leaves04:01
* CIA-3 leaves11:24
* CIA-3 joins11:25
* arjanb joins12:01
* bonniot joins12:46
<bonniot>what do you think about the string matching question?12:54
<arjanb>i agree with the need for a local multiple dispatch construct12:55
the string specific optimizations he propose look quite advanced12:57
<bonniot>but is there any drawback?
<arjanb>creating hashtables goes a bit too far12:59
but i don't see an drawback if the optimization code is cleanly seperated13:00
* CIA-3 leaves
<bonniot>you mean it does not clutter the compiler?13:01
and i think there are other less specific optimations that would improve all Nice code13:02
this is a specific optimization, but it can make a big difference when it is needed13:04
and this is interesting because it would be a huge improvement over using Java
generic optimizations are good, but at this point I think we should still focus on features13:05
i don't think we got any complaints about speed, while features are requested13:06
<arjanb>*away for half an hour*13:10
* CIA-3 joins13:29
* bonniot leaves13:45
* bonniot joins13:46
<arjanb>i think it makes sense for a switch statement not to use compiler ordered alternative13:50
<bonniot>for strings/primites/enums order does not matter, does it?13:58
but when you have match all patterns or multiple argument it can matter
<bonniot>like foo(n,0) and foo(0,n)14:06
<bonniot>that could be handled by textual ordering of patterns that partially overlap and are defined in the same source14:09
<arjanb>i only want textual ordering of switch patterns14:11
i think it would be nice to be able to write:
f(0,n) = ...;
f(n,0) = ...;
f(n,m) = ...;
(that's ackerman ;-)
and that would avoid a special case14:14
<arjanb>i think it's confusing having both for methods
<arjanb>textual and compiler ordered pattern14:18
<bonniot>they work together14:19
the compiler ordering is always the logical one14:20
in addition, partial ambiguities are resolved using the textual ordering
do you see any case where that's a problem?14:23
<arjanb>i think textual ordering is only ok when the alternatives are always together14:24
<bonniot>yes, that's why the ordering would only be partial, for those defined in the same source14:25
<arjanb>i can't explain yet but partial textual ordering of methods feels bad14:33
<bonniot>ok, we can discuss it another time14:36
what are _your_ plans for the summer?14:37
<arjanb>local type inference, new code generation library, looking into a Nice backend for antlr, doing something on the experimental functional language here, and other things i come across14:41
<bonniot>for antlr, do you need to wait for 3.0, or is there already a prototype available?14:43
<arjanb>i need to wait for the 3.0 prototype14:44
<bonniot>i got contacted by somebody who is writing an article about Nice, to be published on IBM's developerWorks portal15:55
<bonniot>I'll have to go. I'll be less present on the net next week, but I will read my mail16:35
<arjanb>you mean coming week?16:36
<bonniot>yes, from today
always mix that one up16:37
<arjanb>i don't know any urgent issue to discuss16:38
* bonniot leaves

Generated by Sualtam