10.3.05
POPL, Plan-X, FOOL
Here are notes on some of my favorite talks at POPL, Plan-X, and FOOL.
Plan-X, 11 Jan 2005.
Michael Carey, BEA, on BEA Liquid Data, an approach to data integration. He notes that BEA developers tend to ignore the strong typing features of XQuery, for two reasons, one trivial and one material. Trivial: developers find it inelegant to have to validate the data after it is created in order to attach type information to it. This is just a matter of writing "validate { ... }" around the XML that creates the data, but they strong resist this. Material: validation can change the data in various corner cases, and so interferes heavily with optimization. I took two points away from this. First, XQuery seems to have failed in the goal of integrating types with queries -- trying to fit well with Schema has inevitably thrown up a lot of road blocks. Second, sometimes seemingly trivial points can be just as significant as material ones; another example of this is how the C-like syntax of C++ and Java helped these languages to spread rapidly.
Alan Schmitt on Xtatic. The pattern matching style of Xduce/Xtatic does not do well with data evolution, unlike XPath, as adding more elements to the data causes matching to fail. I would have thought of this as a bad thing, but Schmitt claims it is good, as the failure tells you where you need to change your code.
The demo session worked well. I particularly liked Fisher and Fernandez's PADX, which I had not seen before.
POPL, 12-14 Jan 2005.
Rob Pike on a programming language for searching Google's data. His neatest example was an animated map, with one pinprick of light for each query sent to Google in one day, correctly located in space and time -- one could watch the pattern of queries flow across the world. The data for the map was generated by a program just a few tens of lines long. The key to the query language is, as Rob noted, a simple idea from the functional programming community: use map and fold as the basis of parallel programming, where the fold is over over associative and commutative operators. Pike's language is based on Google's map-reduce system. (Added 27 Aug 2007: Pike's language is called Sawzall. Later a paper on it appeared here.)
Peter Selinger on a quantum programming language. A lovely presentation of tricky ideas, the first time I had made it through to the end. I finally understood that quantum programming involves two levels of probability, within a state there is the probability attached to each qbit, and then there is the probability of being in a particular entangled state. Selinger explained a clever matrix representation due to von Neuman that captures both levels. He gave a nice explanation of quantum teleportation/quantum compression as two halves of a "use once" isomorphism -- a pair of inverse functions that contain a free linear variable so each can be used only once, where the order of composition determines whether one has teleportation or compression. When I asked him why one should bother with programming languages for computers that might never be built, he pointed to the ability of the language to give precise descriptions of tricky thought experiments, as in the teleportation/compression example.
Pat Hanrahan argued that game processors pose a worthy challenge for language designers. He observes that game processors have a steeper "Moore's law" curve than traditional processors, and hence will outperform them by an exponentially increasing factor.
Dinner on the Queen Mary was fun. Mads Tofte hasn't been active in research lately, and it was great that he showed up to receive the award for Most Influential POPL paper of 1994, for his work on regions.
FOOL, 15 Jan 2005.
There was some question as to whether FOOL would continue after this year, but the meeting was lively and there was a strong concensus in favor of continuing. Benjamin Pierce commented "Most years there is at least one talk at FOOL that, after you hear it, you wish it wasn't accepted. But that wasn't the case this year." Special thanks to Benjamin for serving as chair of the FOOL steering committee for many years, and to Kathleen Fisher for picking up this role going forward.
My two favorite comments, both from the final panel on "Extreme Typing".
Robby Findler: Programmers are motivated to write strong contracts, because the stronger the contract the more likely that blame for a failing program will be assigned to the other guy. This explains why the notion of blame attached to higher-order contracts is important.
Jan Vitek: A recipe for stealth adoption: don't change the syntax of the language, but reinterpret the semantics of the existing constructs. (No one will notice the change anyway!) He was following the mandate to be provocative, but there is a knotty kernel of truth here.
Plan-X, 11 Jan 2005.
Michael Carey, BEA, on BEA Liquid Data, an approach to data integration. He notes that BEA developers tend to ignore the strong typing features of XQuery, for two reasons, one trivial and one material. Trivial: developers find it inelegant to have to validate the data after it is created in order to attach type information to it. This is just a matter of writing "validate { ... }" around the XML that creates the data, but they strong resist this. Material: validation can change the data in various corner cases, and so interferes heavily with optimization. I took two points away from this. First, XQuery seems to have failed in the goal of integrating types with queries -- trying to fit well with Schema has inevitably thrown up a lot of road blocks. Second, sometimes seemingly trivial points can be just as significant as material ones; another example of this is how the C-like syntax of C++ and Java helped these languages to spread rapidly.
Alan Schmitt on Xtatic. The pattern matching style of Xduce/Xtatic does not do well with data evolution, unlike XPath, as adding more elements to the data causes matching to fail. I would have thought of this as a bad thing, but Schmitt claims it is good, as the failure tells you where you need to change your code.
The demo session worked well. I particularly liked Fisher and Fernandez's PADX, which I had not seen before.
POPL, 12-14 Jan 2005.
Rob Pike on a programming language for searching Google's data. His neatest example was an animated map, with one pinprick of light for each query sent to Google in one day, correctly located in space and time -- one could watch the pattern of queries flow across the world. The data for the map was generated by a program just a few tens of lines long. The key to the query language is, as Rob noted, a simple idea from the functional programming community: use map and fold as the basis of parallel programming, where the fold is over over associative and commutative operators. Pike's language is based on Google's map-reduce system. (Added 27 Aug 2007: Pike's language is called Sawzall. Later a paper on it appeared here.)
Peter Selinger on a quantum programming language. A lovely presentation of tricky ideas, the first time I had made it through to the end. I finally understood that quantum programming involves two levels of probability, within a state there is the probability attached to each qbit, and then there is the probability of being in a particular entangled state. Selinger explained a clever matrix representation due to von Neuman that captures both levels. He gave a nice explanation of quantum teleportation/quantum compression as two halves of a "use once" isomorphism -- a pair of inverse functions that contain a free linear variable so each can be used only once, where the order of composition determines whether one has teleportation or compression. When I asked him why one should bother with programming languages for computers that might never be built, he pointed to the ability of the language to give precise descriptions of tricky thought experiments, as in the teleportation/compression example.
Pat Hanrahan argued that game processors pose a worthy challenge for language designers. He observes that game processors have a steeper "Moore's law" curve than traditional processors, and hence will outperform them by an exponentially increasing factor.
Dinner on the Queen Mary was fun. Mads Tofte hasn't been active in research lately, and it was great that he showed up to receive the award for Most Influential POPL paper of 1994, for his work on regions.
FOOL, 15 Jan 2005.
There was some question as to whether FOOL would continue after this year, but the meeting was lively and there was a strong concensus in favor of continuing. Benjamin Pierce commented "Most years there is at least one talk at FOOL that, after you hear it, you wish it wasn't accepted. But that wasn't the case this year." Special thanks to Benjamin for serving as chair of the FOOL steering committee for many years, and to Kathleen Fisher for picking up this role going forward.
My two favorite comments, both from the final panel on "Extreme Typing".
Robby Findler: Programmers are motivated to write strong contracts, because the stronger the contract the more likely that blame for a failing program will be assigned to the other guy. This explains why the notion of blame attached to higher-order contracts is important.
Jan Vitek: A recipe for stealth adoption: don't change the syntax of the language, but reinterpret the semantics of the existing constructs. (No one will notice the change anyway!) He was following the mandate to be provocative, but there is a knotty kernel of truth here.
