FTfJP and the DLS

So on to day 2 of the conference, where I mainly attended FTfJP and briefly popped into the Dynamic Languages Symposium.

First of all I'd like to bing attention to my colleague, Alex Summers' talk, which was a good talk, but more importantly the work is interesting. Actually I have been a little bit involved with it (only a little though) so I have heard a lot of teh stuff he talked about. But, I think it is very interesting, it is about adding static fields and methods to the universe type system. Frankly, I think this is a terrible idea, we should just avoid static anything like the plague (see Gilad's summer school talk and (I think) an earlier blog post), but if you're going to do it, then it turns out to be an interesting problem with no easy solution (the best kind, or worst, depending on your point of view).

The next great talk was by Stefen Wehr who talked about wildcards (a topic very close to my heart). He is concerned with decidability, admitedly not directly. In his work he shows that _a_ model for wildcards and another for interfaces in JavaGI, both of which use a similar kind of existential type, are undecidable. This is obviously interesting since decidability of wildcards is still an open question, which I'd love to look at if I had the time and skills, but I have neither, so I have to make do with reading interesting papers like this one.

The next talk of note was at the DLS, it was an invited talk given by Laurence Tratt entitled Language Design: Back to the Future?. The gist of the talk was that all current languages are almost identical and language designers are not really being creative. He also claimed there are some valuable things that people have done and that have been ignored. I kind of agree, although I don't agree that it is due to a lack of creativity on the part of language designers - there are plenty of fairly wild ideas floating around, but it is true that most never make it into real life languauges. He seemed to concentrate on 'micro' level language features such as new expressions, which is an area (I thinK) that has been neglected in research lately in favour of more 'macro' ideas - such as ownership types, different object meta-models etc. I was a little inspired by the speaker to think a bit more about language culture. His point is that programmers, lanaguage designers, and researchers are all absorbed in a certain culture of a language - the paradigm, the programming idioms, way of thinking, etc. It would be interesting to see some research on this - to charecterise and compare these programming cultures and the effects they have on programming languages, and importantly, vice versa. A quote I liked was that innovative language features were destined to have "some spotty twat at the front take the piss out of it".

I'm afraid that by this point I was a little drained and failed to make any meaningful notes, but I'd like to draw attention to the last two talks of the day that I thought were interesting - Rok Strnisa on the Java Module System and Samir Genaim and Fausto Spoto on constancy analysis

And an interesting thought from one of the talks - if we have an underlying database implementation (for example) that only has a next method as an accessor that returns null at the end of the table. If we wish to implement a wrapper with a hasNext method, then this method will have an internal side effect (it must cache the returned row, if it is not null), but logically it should be pure. Can this be handled elegantly in the various ownership/invariants/effects systems that are around?

ECOOP '08, Cyprus

This is turning into nothing more than a conference blog, and I apologise if anyone actually reads this (actually, at least one or two people do, as I found out - hello!). Its not that I've stopped having ideas, its just that I've been very busy. I'm now in Israel, en route to Egypt to write my thesis (yes, it is a good idea, actually). I have a little time to write a little in the blog, so here is my summary of ECOOP...

ECOOP this year was in Cyprus, not an altogether popular choice, since getting there from anywhere except London was a nightmare, but I came from London so I was happy. Also happy to be very close to the beach and for it to be very warm, thus making for a very enjoyable conference. Well, except for the fact that I was working a little and had a walk to the hotel and it was hot and I inevitably sweat so much I spent most of the conference smelling like a tramp, but hey.

Anyway, the first day was IWACO, which was great. I really like this little workshop and enjoyed nearly all the papers, so you could go worse than go to http://www.cs.purdue.edu/homes/wrigstad/iwaco08/programme.html and read them all! Some comments on some of the talks follow:

First was Jonathan Aldritch's invited talk - the theme of which was making ownership types systems usable by real programmers. It was a great talk and he discussed a lot of aspects that make type systems more or less likely to ever make it into real langauges. He talked about the adoptability and applicability of ownership systems - adoptability concentrated on making these systems easier to use: less annotations, more inference, etc. And how to get the programmer to give the right guidance to the compiler to do inference. Applicability was basically about offering less strict organisation than classical ownership types - such as existential types (!), multiple ownership (!), domains, etc. He also encouraged us to _actually_ study more code to find the strucure of large software.

Alex Potanin then talked about unifying ownership and immutability by using generics, which is a very interesting idea at its early stages.

I gave a talk on existential types for ownership variance, which uses the nice clean framework of existential types to give owner variance without the kind of hackery we use in, for eg, MOJO.

The talk by Nicholas D. Matsakis and Thomas R. Gross was also good. They parameterise types with partitions that restrict effects and use this information to ensure thread safety. It seemed to me like a more fine-grained version of atomic sections, since they restrict the atomic-ness (i.e., the non interfernce of effects) to a parition, while atomic sections apply to the whole heap. I also wonder what would happen if you allowed some kind of variant partitions, like wildcards (not that I'm developing a one-track mind for wildcards or anything).

John Boyland gave a good position paper, which basically said non-linear reasoning (ownership) is better for reasoning about non-linear problems, specifically volatile fields, than linear reasoning (seperation logic).

The talks were rounded off by some interesting demos and a talk about J* by Dino from Queen Mary's, which inspired me to read the seperation logic literature, which I've been meaning to do for a while, but never got round to...

James Noble made the astute comment that some kind of ownership was creeping into seperation logic techniques in order to address certain problems. Also the composite pattern is the new subject-observer - it is difficult to model since you go bottom-up, breaking all the invariants as you go, and then repair them all. I wonder what we will do for motivation when we finish the design patterns book.

Existential types everywhere!

I am certainly biased in this respect, since most of my research work has been with existential types, but I think that existential types are vastly important to the world of OO programming languages, and probably the rest of PLT too. Existential types give us a way of expressing data hiding (i.e., abstraction) in the type system, and seeing as this is what most of software engineering is all about, it shouldn't be surprising that existential types start cropping up all over the place.

My own work has so far focused on Java wildcards, which are a (relatively) simple way of using existential types to deal with the mismatch between parametric (i.e., generics) and inclusion (i.e., subclassing) polymorphism. In the same way I have found existential types to be useful in the world of ownership types. I believe Werner Dietl is having a similar experience with universes.

Existential types are also a foundational model for objects themselves in OO languages. And, I suspect, are a better way of thinking of generic classes than the universal types that people often assume.

Furthermore, I've found some pretty elegant tricks that can be played using existential types mechanisms such as packing and unpacking types, in the realms of path dependent types and virtual classes. I'm pretty sure the same techniques can be used elsewhere.

More than a whiff of existential types can be had all over the place in current research, delayed types, ownership domains, and MOJO all come to mind, and I'm sure I've come across more that I can't remember.

And so what? Well I'm not sure, but I guess I'm glad I've managed to get pretty deep into existential types. I'm sure they'll keep coming in useful for some time to come.

OOPSLA Day 4

The last day at OOPSLA had the session on ownership types, which was great. It also had my talk (paper and slides available here) which went well and I got some excellent feedback. I was also a bit more relaxed once it was out of the way! The talks that I founf interesting today were:

Ownership Transfer in Universe Types Seemed to take a lot from the ideas of external uniqueness and ownership domains, also doesn't seem to work with concurrency. But, the ideas are very interesting and I'm really motivated to read the paper.

Lost in Translation: Formalising Proposed Extensions to C# Dealt manly with LINQ (again), and had a lot of formal stuff in the talk, which is normally a bad sign, but it was pretty easy to follow and the formalism looks interesting

The Java Module System: Core Design and Semantic Definition Was a good talk, all about formalising the proposed Java Module system. Its kind of interesting seeing how it will all work and the work has obviously been well thought through. The size of the formalism is staggering!


And that was the end of OOPSLA, well actually I had quite a few very interesting conversations along the way, especially oni the last day it seems. Had an ice cream and went home, well back to my hotel anyway. All in all it was a great conference, and I look forward to coming back, insha'allah.

OOPSLA Day 3

And so to day 3, and without further ado, the papers I found most interesting:

StreamFlex: High throughput stream programming in Java A mini-solution to real time stream programming, because the general problem, that is real time programming is horrible. Streaming apps have no buffering and no messages can be dropped. Plus all the usual real time constraints. In particular, garbage collection is the big headache. They have an interesting combination of channels, capsules and transient and stable regions of the heap, and an interesting, simple variation on ownership types. Plus region based memory management and transactional memory.

Can programming be liberated from the two-level style? Multi-level programming with DeepJava Presented a cool meta-type system, but I wasn't sure how it was different from other meta-type systems. Nice though. The interesting bits were their potency annotations that give a nice way to specify types in classes lower down the chain of meta-types than the immediate implementer. They also suggest that the system can be used as an alternative to generics, which looks kind of virtual types like, I must look closer at that.

Modular typestate checking of aliased objects Can prevent a lot (apparently) of dynamic errors (or exceptions) statically (always a good thing) using fractional permissions. Seems a nice, little system and deserves a closer look because of the reference to a pack/unpack system, possibly similar to existential types.

Establishing object invariants with delayed types This seems like a big, clever solution to a small, dumb problem, however, the motivation and evaluation seemed convincing. The problem seems to be handling constructors and establishing invariants. The solution is interesting, using delayed types (obviously). James noted an interesting connection to ownership types that deserves further investigation.

OOPSLA Day 2

Day 2 started with two long talks, fortunately both speakers were excellent and the subject matter interesting enough to keep my attention for one and a half hours. The first was by Erik Meijer, titled confessions of a used programming language salesman. The talk (and accompanying essay) is part autobiography, part introduction to LINQ, and part sales pitch for functional languages; which you might have thought would be a hard sell at an OO conference, but hey. Erik started by talking about his background in category theory and functional programming, and reminded me that I wish I could understand this stuff. He then went on to slate C Omega as a a bad job with too many types, but seemed happy with the LINQ stuff, which he revealed are monad comprehensions in disguise. He gave his theory for non-adoption of functional languages and good science in general, by saying the perceived pain of adoption was too high. He thus strives to reduce this to close to zero. He then went off on a bit of a tangent - the aforementioned sales pitch/rant for pure functional languages. on the way he gave the best metaphor for monads I've come across - in the context of a getTime function, the time changes and so isn't a function, but if you return the clock then you're OK, because the clock does not change. Monads are like clocks. He explained it much better than that though. Another good metaphor was making the comparison between functional and OO and nouveax cuisine and McDonalds, although I think this is a little unfair on OO, its a nice image.

The second talk was Second Life: The World's Biggest Programming Environment by Jim Pubrick and Mark Lentczner. This was very cool, it ticked all the boxes - games, 3d graphics, distributed systems, massively parallel and concurrent, and user definable everything! The basic idea is they have this 'game' (or world for creativity and business, depending on your point of view) with huge numbers of users, 15% of whom write scripts. The numbers for programmers, scripts, lines of code, etc. were staggering, never mind the sheer number of potential users for each script. They weren't shy in admitting that there scripting language is an abomination, but they claim the distribution model is great. In effect they have very real encapsulation, each unit is tiny and self-contained, but there are millions of them and they all run concurrently on thousands of processors, so the kind of stuff we've been promised for the future in countless motivation chapters. On a side note they have an explicit of state in their language and I think this is a great opportunity to plug some work I participated in - StateJ, paper available from the usual place. Anyway, the long and short of things is that they are porting it all to Mono so it all works properly at scale and they employ a cornucopia of neat tricks to do this, most of which are hacks to get poor design decisions in the CLR, eg. lightweight threads, serialisation, etc. Basically they have a very cool problem on their hands. They finished off by selling Second Life as a development environment, showing how their dev team collaborate virtually. It was a pretty cool demo, everyone loves pretty graphics at the end of the day! On another note, their security model is cool, looks like security is more interesting than I thought.

The last session I attended was research papers Atushi Igarashi presented Variant Path Types for Scalable Extensibilty, which is an extension of the FOOL paper I've described before, so won't do so again, its still very cool stuff.

Next up was Dependent Classes by Vaidas Gasiunas, a development of virtual classes that subsumes multiple dispatch, it is one of the most interesting papers I've read in a long time. But we've discussed it in SLURP, so I won't write more.

The last paper was User-Changable Visibilty: Resolving Unanticipated Name Clashes in Traits, the idea was adding access modifiers to traits and being able to change them on trait composition by freezing and defrosting. It looked like interesting stuff, but the talk was quite basic, mostly taken up with an introduction to traits. The paper looks worth reading though.

OOPSLA Day 1 - Dynamic Languages Symposium

The Dynamic Language Symposium was interesting, the invited talk was on security and this (in particular capabilities) seemed to permeate a lot of the questions. I found a lot of the talks (esp the invited talk) thought provoking, but not directly interesting, which is a shame because dynamic languages are a hot and interesting area. In particular the invited talk got me thinking about security as an interesting topic, which takes some doing!

This is pretty lazy, but I'm not going to write anything specific about the talks, mostly because I don't have anything deep to say. But here's a list of the ones I found most enjoyable/interesting/thought provoking:

Tradeoffs in Retrofitting Security: An experience Report

OMeta: An Object Oriented Language for Pattern Matching

Dynamic Ownership in a Dynamic Language

Highly Dynamic Behaviour through Prototypes with Subjective Multimethods

Mirages: Behvioural Intercession in a Mirror Based Architecture

OOPSLA 07, Montreal - First Impressions

So, I'm at OOPSLA and I'll be blogging about it. I'm talking on Thursday, so I can't relax until the end, but hey, should still be fun.

Montreal is beautiful at the moment, all the trees are gold, red and green, really stunning stuff. I've seen a few parks and wondered around the city, and its all very nice. Not much in the way of history though, the old town is only old relatively. In fact the whole place feels slightly soulless, but still a nice town.

The conference venue is pretty impressive, very posh and huge - a real difference from the usual academic conference in a university! Also there are a lot of people and a lot of stuff going on, it certainly feels like a pretty unique kind of conference (at least compared to what I'm used to).

Anyway, the next posts will deal with the interesting stuff - the talks...

ECOOP day 3

And so the last and shortest day began. It started with Jonathan Aldrich's invited talk. This was another good talk, describing the research he has been involved in with software architecture. The general gist is that the architecture of software should be part of the code and enforced by the compiler. To this end he talked about ownership, and his variant - ownership domains, and also ArchJava. He motivated ownership domains with the MVC pattern, showing that owners-as-dominators style ownership is too restrictive to be used there. The whole thing sounded very ambitous and also extremely practical, so pretty exciting stuff. Also mentioned existential types in the context of ownership types, which is something I want to look at.

The first talk on friday was MAO: Ownership and Effects for more Effective Reasoning about Aspects - motivated by the difficulty of reasoning about aspects (their greatest drawback in my opinion), the authors used ownership and effects to aid this reasoning and claimed to get good results.

I'm afraid that by friday I was suffering conference fatigue and my notes were getting pretty flakey, but I did manage to write down which talks I found intersting, so I can read the papers:

Joinpoint Inference from Behavioral Specification to Implementation

A Machine Model for Aspect-Oriented Programming

Tracking Linear and Affine Resources with Java(X)

Attribute Grammar-based Language Extensions for Java

And that was the end of ECOOP. Now I have to find some time to read a lot of papers in more detail, I fear the list never gets shorter.

ECOOP day 2

The day kicked off with Luca Cardelli's invited talk. This was a personal history of his involvement with OO theory (and practice to some extent) and type theory, which turns out to be a pretty good history indeed. Once again concurrency reared its ugly head as its Luca's current research area and he thinks its the next intersting thing.

Yi Lu - Validity Invariants and Effects - Lu and Potter's annual ECOOP paper was as interesting as expected. This year they extended ownership types with invariants, the idea being it is easier to reason about invariants in the presence of ownership types.

Non-null references by default in Java - was a case study on delarations that could be declared non-null or nullable and found that most could be non-null. They then argued that non-null should be the default, which seems fair, but is unlikely to happen in Java.

Burak Emir - Matching Objects with Patterns - Compares various OO pattern mathcing techniques and suggests a new one. Pattern matching seems like a good idea, but is sometimes criticised for being non-OO, will read the paper and decide...

I liked the talks on DirectFlow: a Domain-Specific Language for Information-Flow Systems and A Relational Model of Object Collaborations and its Use in Reasoning about Relationships, but don't have much to say about them, sorry.

I missed the talk on JavaGI, but I think I read an earlier version of the work (I seem to remember it being very powerful, but complex) and liked it so will go back and read the paper.

Aaron Turon - Metaprogramming with Traits - Uses a flexible version of traits to do metaprogramming (just like it says on the tin). They have an interesting type system that is hybrid structural/nominal. On a side note, I like traits, I think they are cool.

Shan Shan Huang - Morphing: Safely Shaping a Class in the Image of Others - more metaprogramming than morphing (classes do not change during execution), this seems powerful and generic and (relatively) easy to use. I've written "label checking" in my notes, but can't remember why now. An interesting view on this work is that they have a form of type-checked reflection, which is still a big open question at the moment. The claim that the language was more intuitive (has more "synergy" ) than the competition, was unsurprisingly provocative.

A Higher Abstraction Level using First-Class Inheritance Relations - An interesting idea that basically 'does components' by having a complex and novel form of inheritance.