Sign in with Twitter

Username:

Occasional OCaml programmer

95 Following   2,383 Followers   1,443 Tweets

Joined Twitter 6/21/09

Reverse Tweets
Ah, the mysteries of Amazon's categorization system... http://t.co/0ecqGrFz9dJust got my copy of "The Little Prover", co-authored by Jane Street's own Carl Eastlund. It looks really fun! http://t.co/K3U0vnDm5u@ccshan There's no Haskell. Haskell has no monopoly on monads, or on infix operators!
7/23
2015
Another nice post about self adjusting computations, this time from Neelkatan Krishnaswami: http://t.co/USSU9nKLy3@PhilipWadler Perhaps we can make an exception for any author of "Theorems for Free". Anyone qualifying should email me their address.
7/22
2015
So sad... https://t.co/2R1MjAZzFR@zittrain So instead they need to spend a lot doing the more difficult job of post-hoc security engineering.@zittrain It seems like the car companies are too cost-conscious to spend any time doing security engineering in advance.@zittrain What's especially shocking is that this has all been clear for years. I heard about this from DARPA folk 2-3 years ago.@zittrain The car companies are the least cost avoiders, and the avoiding isn't even that hard: just don't connect the drivetrain to the net@zittrain I find the whole thing shocking. Do you have any thoughts as to why the obvious liability didn't incentivize more carefulness?@yminsky and this result was so predictable. It really feels like a case where liability concerns should have helped the decision-making.@jakedonham those are key parts of the design. I think my favorite thing about the shirt is that it doesn't 100% make sense.That cars don't have an air gap between control systems and the internet should be considered security malpractice http://t.co/80cJRRpRfh@satnamsingh I'm sorry Satnam, but you'll have to wait for ICFP like everyone else...If you're going to ICFP, fear not: we'll be distributing a few hundred of these there.The latest Jane Street T-shirt. I pretty much love the gold foil. http://t.co/DJuclH8unX
7/21
2015
@jhaberku every single one is moving to ppx.@ndm_haskell My personal experience with kids and adults alike is contrary to that, but we're now in the realm of anecdote.@sjfloat @ndm_haskell Maybe, but I'm skeptical. Once experienced with types, they're more productive immediately. But it takes time to learn@craigstuntz @danluu I started there. But all the studies on that list one way or another fail to account for experience and scale.The first patch to switch our tree off of camlp4 has (finally) landed! I'm hoping our next public release will have no dependency on camlp4.@ndm_haskell Days seems crazy to me. Dynamic languages are really more effective early on. Types shine with scale and experience.@sperbsen the most implausible part of your conjecture is the first 48 hours.And existing studies on all sides strike me as deeply problematic because of their inability to cope with experience and scale.My point really is that studying such a question at reasonable cost seems extremely hard. An unfortunate fact for PL research.OH: Occam's Taser: the most painful explanation is usually the correct one.
Retweeted by Yaron MinskyConjecture: dynamic PLs are more productive for the first 6mo, static ones for the next 10yrs. how would you design a study to confirm?
7/20
2015
@johanatan @rtfeldman @deech @jonsterling it's also theoretically possible that the quote is slightly tongue in cheek...@gravbeast @ambrosebs https://t.co/0XGOINjZ0k Start at minute 27. He seems particularly sad about reified generics at runtime.@johanatan @rtfeldman @deech @jonsterling from the point of view of a GC paper, it's pretty conventional...@johanatan @rtfeldman @deech @jonsterling A Leroy and Doligez GC papers says: "The ML language is a conventional imperative language..."@johanatan @rtfeldman @deech @jonsterling I don't quite revere ML, but I do like it. That said, I don't hate impurity in other languages...@gravbeast @ambrosebs I saw it in this talk: https://t.co/0XGOINjZ0kAnd types can help you slim down runtime representations massively, as with mlton. The nature of his concern is unclear to me.In particular, untyped languages are forced into heavier representations so you can discover the correct behavior at runtime.But I don't understand his view that types shouldn't affect runtime. He says types weight down the runtime, but that's backwards.I normally find myself disagreeing with @Gilad_Bracha, but the point that documentation is one of the primary benefits of types rings true.@marius I'm curious: How much of SAC does Var support? Does it give you bind-like dynamism, or just static graphs?@marius With one master process to orchestrate stabilization of the multi-process graph. This is not part of Incremental proper, though.@marius Parallel applications are broken up into multiple processes, each with its own incremental universe.@marius Typically one universe per process per application of incremental, but sometimes multiple per program.@lambda_calculus @krismicinski @matthewhammer I'll be mostly at the workshops after, but yes!@krismicinski Indeed. I'm quite curious about how adapton differs conceptually from incremental, actually.@rshestakov you can do that with Incremental, without any special support. Am I missing something?A blog post about the newly released Incremental. I'm hoping this gives people a sense of what it's useful for. https://t.co/3tgfElstAmReal World OCaml With Obj.magic http://t.co/kPXCZOxDjb
Retweeted by Yaron Minsky
7/18
2015
two postdoc positions available in rigorous engineering of mainstream systems: http://t.co/kE6yLsPxHC fixing the real world now :D plz rt
Retweeted by Yaron Minsky
7/17
2015
@krismicinski @recursecenter http://t.co/MdxC6vtpwTMy favorite bit about talking at @recursecenter is that when you say "ask questions", they do. Good ones, too. https://t.co/NUlwgDhmYwNew tech report on a probabilistic network programming language #SDN #Frenetic #NetKAT http://t.co/LkM0JplEIn
Retweeted by Yaron Minsky
7/14
2015
@recursecenter @yminsky sporting emacs slides like a boss :-)
Retweeted by Yaron MinskySO DISAPPOINTED that "Go Set A Watchman" does not end with Atticus Finch battling Ozymandias.
Retweeted by Yaron MinskyWe're excited to have @yminsky back tonight to talk about distributed systems! http://t.co/6h78Q1M0Dv
Retweeted by Yaron Minsky
7/13
2015
@superfunc First Scratch, now Racket.@marius @avsm @cmeik (though he wouldn't agree...)@marius @avsm @cmeik Actually, this is a better link. His work on hierarchical laws to me is mostly about modularity http://t.co/Bd70uftpLWParenting achievement unlocked: I just had to admonish my son to write somewhat fewer comments in his code.@marius @avsm @cmeik but there's a ton more to do...@marius @avsm @cmeik My dad's work on law governed architecture makes some interesting nips at the problem, actually http://t.co/g2QB5kp5N6@marius @avsm @cmeik Nice summary, particularly the list of uncovered topics. I'm especially interested in modularity in distributed systems@ID_AA_Carmack some (though only some) of the same win can be had through heavy cross module inlining.@ID_AA_Carmack mlton is a whole program SML compiler that gets a lot of juice out of this. The key win is zero-cost abstractions.
7/12
2015
@onigoetz ssh
7/11
2015
@chewedwire Both us and OCamlPro are making a big push to get the patch ready. Exciting stuff.@chewedwire For example, with it you can add an optimization so a variant returned by a function directly into a match needs no allocation.@chewedwire flambda is a new IR for OCaml that comes along with a much better inliner. It brings you much closer to zero-cost abstractions.Signs are good for an Flambda release around year end, and it's going to totally change how you write performance sensitive OCaml.
7/10
2015
@whitequark @avsm @das_kube I'm glad to hear you guys are all happy with the outcome. The intent was never to cause a split...@agumonkey @shajra I think the strongest argument is the easier extensibility due to Python.@deanwampler @bos31337 @shajra It's a shame, really. I wish for a more distributed infrastructure for DVCSs, but there's no push for it.@deanwampler @bos31337 @shajra Indeed. But the centralization is worrisome. And other services (e.g., bitbucket) are pretty good too.@shajra @bos31337 @kmarekspartz But changes like this were not something the git community was interested in https://t.co/POQ14crD2q@shajra @bos31337 @kmarekspartz Mercurial certainly isn't perfect. e.g., highly branchy workflows within a big repo can be inefficient.@bos31337 @shajra the killer app for git really seems to be github. I think that explains a lot of the difference.@bos31337 @shajra And I've heard rumors of Google's growing interest in hg. Interesting seeing this as the open source world flocks to git.@shajra and @bos31337 has described problems in adapting git to deal with performance issues with large monorepos at Facebook.@shajra More concretely, the ability to customize hg with python send to help in adapting it to custom workflows.@shajra Mostly, it's the edge of despair in the Git engineer's voice that makes me feel good about using hg.The more I talk to people at major tech firms about their internal use of git, the happier I am that we use mercurial.
7/9
2015
Converting a code base from camlp4 to ppx: As with many projects in the OCaml world, at Jane Street we ... http://t.co/YVSy27mVKH #ocaml
Retweeted by Yaron Minsky@kragen @rob_pike @jbuberel It has been from the start. It was based on a parallel design that was later simplified to an incremental one.@rob_pike @jbuberel Interesting. I'm used to incremental collectors, like OCaml's, where latencies are measured in 100's of mics.@jbuberel @rob_pike Ouch. Kind of stunning that the pauses were measured in seconds. We're those compactions?
7/8
2015
@lambda_foo just read Umut Acar's work on self adjusting computations. Incremental is pretty much an implementation of that idea.Given the Incremental release, this post explaining the connection between SAC and FRP might be useful: https://t.co/u7dCJWJLmm@jordwalke If the thing that is computed is itself a pure value, don't you get what you want for free, on top of Incremental?@michael_w_hicks as for implementation, perhaps take a look at our use of gadts to make the representation more explicit and lighter.
7/7
2015
@michael_w_hicks And they seem to both avoid the use of Sleator-Dietz, though for somewhat different reasons I think.@michael_w_hicks The idea of explicitly tracking what outputs are observed seems to show up in both.@michael_w_hicks Having just released Incremental, I happened to take a look at that Adapton paper. I think there are a lot of connections.@codeclimate @mattmight @plumlab You might want to take a look at the just released Incremental library for OCaml. https://t.co/i9QWFGWbELI'm pleased to announce that we just open-sourced Incremental, an OCaml library for self-adjusting computations. https://t.co/i9QWFGWbEL
7/6
2015
@sperbsen What about it is uncomposeable?
6/26
2015
Who is writing the law review article: "From Argle-Bargle to Jiggery-Pokery: Toward a Theory of Scalian Textualist Putdowns" ?
Retweeted by Yaron Minsky
6/25
2015
@das_kube we had to rewrite a dozen syntax extensions and verify precise ast compatibility over a 3 million line codebase. It takes time...
6/24
2015
Don't get me wrong: camlp4 has been incredibly valuable. But having two separate parser implementations just isn't worth the complexity.We're on the verge of a massive de-camlp4-ization of the Jane Street codebase. I can't express how much I'm looking forward to it.
6/23
2015
@emilybazelon is one of my favorite legal commentators, and this article of hers in the Times is especially good. https://t.co/lRfu9FQ7aI
6/21
2015
So how do you debug a stack overflow when using js_of_ocaml? It's always on line 1....
6/19
2015
« Prev1234567

0