3 ms·

You just described the entire bank! I prefer my single codebase/single build definition. I forget the current figure, but StanChart has at least 3M lines of (de

by willtim 10y ago

You just described the entire bank! I prefer my single codebase/single build definition. I forget the current figure, but StanChart has at least 3M lines of (dense) Haskell. More details here: https://skillsmatter.com/skillscasts/9098-haskell-in-the-large-the-day-to-day-practice-of-using-haskell-to-write-large-systems https://skillsmatter.com/skillscasts/9098-haskell-in-the-lar...

pron 10y ago
> I forget the current figure, but StanChart has at least 3M lines of (dense) Haskell Oh, thank god, finally we get a number. It turns out that I was completely wrong, and that in its 20 years of hyped existence, someone has built something big with Haskell once. You may think I'm sneering, but only a little bit, because while one is almost nothing, it is much better than actually nothing because at least it is a first anecdote. Now, who do I have pester to get some more metrics? > More details here: Sadly, there are no more relevant details in that talk.
- willtim 10y ago
  Just out of interest, are you skeptical of static type systems in general? Because AFAIK there has never been a quantitative study proving their effectiveness, only empirical studies.
  - sheepmullet 10y ago
    There is a lot of evidence that "primitive" static type systems add value via enabling better tooling. Low cost of adoption + clear minor tooling benefits is usually enough evidence to adopt Java/C# style static types. Haskell style static typing has a very high cost of adoption and so must conclusively show a strong benefit in order to be adopted by industry. At the moment I feel Haskell companies can get away with using Haskell because PL enthusiasts are willing to absorb the training costs.
  - pron 10y ago
    I am not skeptical of type systems in general and TBH I've never worked on a large piece of software written in an untyped language (I often program in Clojure, but so far never a large system -- and yeah, I am skeptical about Clojure too even though I love it). But you're mistaken if you think I'm looking for quantitative studies; that's far too high a bar. I am looking for well-researched anecdotes, and there are plenty of them for type systems in general. I am, however, very skeptical of the interpretation some people, especially in typed FP, give to types and the reasons they believe types provide a benefit. For example, the relationship between useful type systems and software correctness is not direct. I am currently using a formal verification tool that is completely logic based (and it has an interactive theorem prover, a model checker, and is backed by decades of careful mathematical analysis of its soundness) which is completely untyped, and yet it is as powerful as Coq for proving correctness of algorithms. If direct, formal, proofs of correctness is what you're after, types might not be the best solution. Types, however, have other clear benefits that are not related to formal correctness, or, at least, not correctness of global program properties.

by willtim 10y ago

pron 10y ago

> I forget the current figure, but StanChart has at least 3M lines of (dense) Haskell Oh, thank god, finally we get a number. It turns out that I was completely wrong, and that in its 20 years of hyped existence, someone has built something big with Haskell once. You may think I'm sneering, but only a little bit, because while one is almost nothing, it is much better than actually nothing because at least it is a first anecdote. Now, who do I have pester to get some more metrics? > More details here: Sadly, there are no more relevant details in that talk.

willtim 10y ago
Just out of interest, are you skeptical of static type systems in general? Because AFAIK there has never been a quantitative study proving their effectiveness, only empirical studies.
- sheepmullet 10y ago
  There is a lot of evidence that "primitive" static type systems add value via enabling better tooling. Low cost of adoption + clear minor tooling benefits is usually enough evidence to adopt Java/C# style static types. Haskell style static typing has a very high cost of adoption and so must conclusively show a strong benefit in order to be adopted by industry. At the moment I feel Haskell companies can get away with using Haskell because PL enthusiasts are willing to absorb the training costs.
- pron 10y ago
  I am not skeptical of type systems in general and TBH I've never worked on a large piece of software written in an untyped language (I often program in Clojure, but so far never a large system -- and yeah, I am skeptical about Clojure too even though I love it). But you're mistaken if you think I'm looking for quantitative studies; that's far too high a bar. I am looking for well-researched anecdotes, and there are plenty of them for type systems in general. I am, however, very skeptical of the interpretation some people, especially in typed FP, give to types and the reasons they believe types provide a benefit. For example, the relationship between useful type systems and software correctness is not direct. I am currently using a formal verification tool that is completely logic based (and it has an interactive theorem prover, a model checker, and is backed by decades of careful mathematical analysis of its soundness) which is completely untyped, and yet it is as powerful as Coq for proving correctness of algorithms. If direct, formal, proofs of correctness is what you're after, types might not be the best solution. Types, however, have other clear benefits that are not related to formal correctness, or, at least, not correctness of global program properties.