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

Using timezone: Central European Time
* CIA-9 leaves09:15
* CIA-7 joins
* bonniot joins09:45
* arjanb joins13:45
<arjanb>had some discussion with cm here last week13:56
problems were ambiguity with function arguments, less type params with java classes/ retypings, performance of generic arrays13:59
<bonniot>what's the 1st one?
<arjanb>and nicec seems to behave different than javac on linux when leaving the first slash away in classpaths14:00
<bonniot>i see14:01
did u give a workaround?
<arjanb>he found that on his own14:03
<bonniot>which one?14:05
<arjanb>for all
* CIA-7 leaves20:56
* CIA-9 joins
* cm joins22:32
<cm>arjanb: There is some construct in the kiev language to compile specific instances of generic functions. Somewhat clumsy solution22:58
<arjanb>yes i don't think the user should to determine which specialization are needed23:07
<cm>I'm considering writing a paper on whether bioinformatic research could benefit on using functional languages. Nice could be an option here. 23:14
<arjanb>does bioinformatics includes crunching much data?23:17
<cm>A lot of the problems do. Most are however NP-complete and requires hence a lot of computational power. 23:21
There are no good libraries to solve problems today. I think a library in a functional language would be great23:25
<arjanb>hmm as Nice isn't complete yet it does only trivial optimizations
<bonniot>hi cm23:27
<arjanb>if the topic of the paper is comparative then would Nice fit well between Haskell and Java
<cm>Thats true but it could be as a proof of concept. Some research institutes have a lot of computational power. I have than I need
hi bonniot23:28
Yes I think so too. Perl isn't an option despite is it very common today
in conjunction with some obscure c++ code23:29
<bonniot>for numeric computations I don't think there is a problem with optimization, because that belongs to the JIT level, and the JVMs are doing quite well now
<cm>True. The problem I have is the tedious process of implementing things that are out of scope and the lack of reuse23:31
<arjanb>JVMs don't do well on generic containers with primitives
<cm>But nice could overcome that with specific instances. 23:32
what kind of things do you miss?
that you need to implement23:33
<cm>In nice?
Hidden markov models, aligment of sequences etc but also a library where I could combine functions as you could to in a functional language23:34
Most data oriented problems could be defined as evaluating some function over a set23:35
<bonniot>function combinations would be generally useful
i'm sure Bryn would be interested to help with that, maybe Isaac too23:36
<cm>Yes very. Object orientation isn't even could in terms of resuse
<bonniot>there is the nice.functional package that would be a good fit
for the more specific libraries, are there no implementations in Java?23:37
<cm>Yes BioJava but the design is really crazy23:38
<bonniot>then the work must be done somehow...23:40
<cm>Yes and you wan't to focus on the core problem.23:41
<arjanb>looking at BioJava i guess most bioinformatics don't have a cs background23:45
<cm>Neither do I... I studied engineering physics23:46
It is a general problem however. Computer scientist don't understand biology and vice versa23:48
Some tools are written from scratch in c without any libraries at all. Everything from lists to hashtables are impl.23:49
<arjanb>i'm sure making a libray in functional style is a big improvement but the choice of language is yours00:06
what do your fellow bioinformatics think of functional programming?00:08
<cm>They think that functional programming is defined as lisp...00:10
<cm>I showed them a proof of concept in haskell and they were quite impressed of the power in terms of writing concise code00:12
But the learing curve with Haskell is quite steep 00:13
And where the error messages look like some type theory expression00:14
The practical part of bioinformatic research is more or less a trial and error science and thats why it is import to be able to go from idea to impl. fast00:16
<arjanb>understandable type error messages and type inference don't go well toghether
could these libraries be implemented as DSL?00:18
<arjanb>domain specific language00:19
I have seen some pretty impressing circuit validators written in haskell00:20
As a DSL
<arjanb>a DSL is usefull if most users don't need to extend functionality00:27
<cm>You have a problem though if the DSL isn't powerful enough and/or when the problem isn't well defined enough00:28
<cm>No volunteers for the eclipse plug-in. I could contribute if someone took the lead. I have some eclipse plug-in dev. experience. We have started to use eclipse as a platform for our in-house developed programs00:32
<arjanb>daniel for you to answer00:35
i know too little about bioinformatics to help you directly but we can give comments on library prototypes00:36
<cm>I was thinking about the general eclipse plug-in00:37
<arjanb>anyway time to sleep for me, courses have started again :(00:38
<cm>Good night, time to sleep for me as well00:39
<arjanb>daniel knows more about the plug-in than I
good night
* arjanb leaves
<bonniot>there's a student who plans to work on the eclipse plugin as a project for his school00:53
that should start this september00:54
<cm>Ok, I think that it could help attract nice users01:00
Has anyone investigated if parts of JDT could be reused?01:02
And maybe implement the Nice plug-in as a project nature to the Java project.01:03
What is the status of the ambiguity problem when using functions as "function value", eg. list.app(println)01:05
* cm leaves01:33

Generated by Sualtam