Pursuing the Strong, Not So Silent Type: A Haskell Story
•
1 gostou•1,258 visualizações
In recent months Katie's Facebook team has completely replaced an in-house interpreted language, moving to a strong and statically typed Haskell DSL called Haxl. Dozens of Facebook developers have become functional programmers, using the open-source Haxl framework to battle spam at scale. This talk will explain how Haskell shines in this context, bust a few myths about the language, and highlight lessons Rubyists and Haskellers could learn from each other.
Recomendados
Mais conteúdo relacionado
Mais procurados
Mais procurados (19)
Destaque
Destaque (19)
Semelhante a Pursuing the Strong, Not So Silent Type: A Haskell Story
Semelhante a Pursuing the Strong, Not So Silent Type: A Haskell Story (20)
Último
Último (20)
Pursuing the Strong, Not So Silent Type: A Haskell Story
- 1. Pursuing the strong, not so silent type A Haskell story by Katie Miller (@codemiller) Software Engineer at Facebook
- 3. "The limits of my language mean the limits of my world" - Ludwig Wittgenstein
- 5. Source: Ruin Raider on Flickr, CC BY-NC-ND 2.0
- 11. more than 1 million requests/second
- 12. def haskell_spammer(user, friend, post) if talking_about_haskell(post) && num_common_friends(user, friend) < 5 && most_friends_like_ruby(friend) block_post else do_nothing end end
- 14. FXL
- 15. functional
- 18. HaskellSpammer = If (TalkingAboutHaskell(postContent) && NumCommonFriends(userId, friendId) < 5 && MostFriendsLikeRuby(friendId)) Then BlockAction Else @[] NumCommonFriends(x, y) = Length(Intersect(FriendsOf(x), FriendsOf(y)))
- 19. 20x speedup
- 21. Haskell
- 25. haskellSpammer :: Haxl Bool haskellSpammer = talkingAboutHaskell .&& numCommonFriends .< 5 .&& mostFriendsLikeRuby where mostFriendsLikeRuby = do friends <-‐ getFriends rubyFriends <-‐ filterM friendLikesRuby friends return (length rubyFriends >= length friends `div` 2) friendLikesRuby friend = hasAssoc friend likeAssoc rubyPage
- 27. hasAssoc :: Int -‐> Int -‐> Int -‐> Bool hasAssoc id assoc target = ...
- 28. hasAssoc :: Int -‐> Int -‐> Int -‐> Bool hasAssoc id assoc target = ... newtype Id = Id Int newtype AssocId = AssocId Int
- 29. hasAssoc :: Int -‐> Int -‐> Int -‐> Bool hasAssoc id assoc target = ... newtype Id = Id Int newtype AssocId = AssocId Int hasAssoc :: Id -‐> AssocId -‐> Id -‐> Bool hasAssoc id assoc target = ...
- 30. data Language = Ruby | Haskell | Php
- 31. data Language = Ruby | Haskell | Php likesLanguage :: Language -‐> Id -‐> Bool
- 32. data Language = Ruby | Haskell | Php likesLanguage :: Language -‐> Id -‐> Bool likesLanguage Ruby userId = hasAssoc userId likesAssoc 1995 || hasAssoc userId likesAssoc 2005
- 33. data Language = Ruby | Haskell | Php likesLanguage :: Language -‐> Id -‐> Bool likesLanguage Ruby userId = hasAssoc userId likesAssoc 1995 || hasAssoc userId likesAssoc 2005 likesLanguage Haskell userId = hasAssoc userId likesAssoc 42
- 34. data Language = Ruby | Haskell | Php likesLanguage :: Language -‐> Id -‐> Bool likesLanguage Ruby userId = hasAssoc userId likesAssoc 1995 || hasAssoc userId likesAssoc 2005 likesLanguage Haskell userId = hasAssoc userId likesAssoc 42 likesLanguage Php _ = False
- 35. Is Haskell difficult to learn?
- 36. expectations
- 38. Is Haskell a panacea?
- 39. blockAustralians :: Haxl SyncResponses blockAustralians = do textMap <-‐ textArr let text = HashMap.lookupDefault "" "main_text" textMap numBadWords = length $ filter (`Text.isInfixOf` text) aussieTerms numBadPhrases = length $ filter (`Text.isInfixOf` text) aussieSayings if numBadWords < 2 && numBadPhrases <= 0 then return noResponses else if numBadWords < 4 && numBadPhrases < 2 then return requireCaptcha else if numBadWords < 5 && numBadPhrases < 3 then return $ responses [warnUser, requireCaptcha] else if numBadWords < 7 && numBadPhrases < 4 then return warnUser else if numBadWords < 8 && numBadPhrases < 5 then return $ responses [warnUser, blockAccess] else if numBadWords < 10 && numBadPhrases < 6 then return blockAccess else if numBadWords < 13 && numBadPhrases < 7 then return $ responses [ blockAccess , enrollInFakeAccountCheckpoint ] else return $ responses [ blockAccess , enrollInFakeAccountCheckpoint , requireCaptcha ] where aussieTerms = [ "Acca Dacca" , "ambo" , "arvo" , "Aussie" , "bangaroo"
- 40. blockAustralians :: Haxl SyncResponses blockAustralians = do textMap <-‐ textArr let text = HashMap.lookupDefault "" "main_text" textMap numBadWords = length $ filter (`Text.isInfixOf` text) aussieTerms numBadPhrases = length $ filter (`Text.isInfixOf` text) aussieSayings if numBadWords < 2 && numBadPhrases <= 0 then return noResponses else if numBadWords < 4 && numBadPhrases < 2 then return requireCaptcha else if numBadWords < 5 && numBadPhrases < 3 then return $ responses [warnUser, requireCaptcha] else if numBadWords < 7 && numBadPhrases < 4 then return warnUser else if numBadWords < 8 && numBadPhrases < 5 then return $ responses [warnUser, blockAccess] else if numBadWords < 10 && numBadPhrases < 6 then return blockAccess else if numBadWords < 13 && numBadPhrases < 7 then return $ responses [ blockAccess , enrollInFakeAccountCheckpoint ] else return $ responses [ blockAccess , enrollInFakeAccountCheckpoint , requireCaptcha ] where aussieTerms = [ "Acca Dacca" , "ambo" , "arvo" , "Aussie" , "bangaroo"
- 41. results
- 43. community
- 44. Haxl team past and present Louis Brandy Jonathan Coens Andrew Farmer Kubo Kováč Jake Lengyel Simon Marlow Katie Miller Bartosz Nitka Jon Purdy Aaron Roth Zejun Wu Noam Zilberstein
- 45. More about Haxl Haxl on GitHub 'Fighting spam with Haskell' blog post 'There is no Fork' ICFP paper and presentation 'The Road to Running Haskell at Facebook Scale' presentation Wired article
- 46. The End by Katie Miller (@codemiller) Software Engineer at Facebook