vshabanov

{-# LANGUAGE OverloadedStrings, BangPatterns, FlexibleInstances #-}
-- | Тупейший (и, надеюсь, быстрый) парсер xml -> теги.
--
-- Выдает где-то от 40МБ/сек (а где-то и 80 и 200)
--
-- Все теги и атрибуты возвращаются в lowercase (строка меняется INPLACE,
-- для уменьшения числа аллокаций и фрагментации)
--
-- После профилирования выяснилось,
-- что TagSoup отжирает 75% памяти и >50% времени. Более того,

vshabanov

Core was generated by `/tank/riak-1.3.1/erts-5.9.1/bin/beam.smp -K true -A 64 -W w -- -root /tank/riak'.
Program terminated with signal 11, Segmentation fault.
#0  0x00007f162547fa30 in MurmurHash64A(void const*, int, unsigned int) ()
   from /tank/riak-1.3.1/lib/eleveldb-1.3.0/priv/eleveldb.so

(gdb) bt
#0  0x00007f162547fa30 in MurmurHash64A(void const*, int, unsigned int) ()
   from /tank/riak-1.3.1/lib/eleveldb-1.3.0/priv/eleveldb.so
#1  0x00007f162547833c in leveldb::(anonymous namespace)::BloomFilterPolicy2::CreateFilter(leveldb::Slice const*, int, std::string*) const ()
   from /tank/riak-1.3.1/lib/eleveldb-1.3.0/priv/eleveldb.so

vshabanov

import qualified Data.ByteString.Lazy.Char8 as B
import qualified Data.Text.Lazy as T
import qualified Data.Text.Lazy.Encoding as T
import qualified Data.Aeson as JSON
import Data.Char
import Text.Printf
import Control.Monad

-- wget http://www.unicode.org/Public/9.0.0/ucd/UnicodeData.txt

vshabanov

{-# LANGUAGE GADTs #-}

class Clickable a where
    click :: a -> String
class Renderable a where
    render :: a -> String

data Interface a where
    Clickable :: Clickable a => Interface a
    Renderable :: Renderable a => Interface a

vshabanov

resolveA :: DNSCache -> String -> IO (Either T.Text HostAddress)
resolveA c d =
    handle Left $ checkDomain id d $ fmap errorStr $ resolve c (B.pack d)

resolveCachedA :: DNSCache -> String -> IO (Maybe (Either T.Text HostAddress))
resolveCachedA c d =
    handle (Just . Left) $ checkDomain Just d $ fmap (fmap errorStr) $ resolveCache c (B.pack d)

handle f a = a `E.catch` handleError f

vshabanov

{-# LANGUAGE GHC2021, LambdaCase, PatternSynonyms, RecordWildCards,
             ViewPatterns, OverloadedStrings, GADTs, NoMonoLocalBinds,
             UnboxedTuples, MagicHash #-}

{-# OPTIONS_GHC -O2 -fspec-constr-count=100 -fllvm #-}
-- need at least 8 to completely remove all allocations
-- (I don't know what change lead to it)

{-# OPTIONS_GHC -Wall -Wno-gadt-mono-local-binds -Wno-type-defaults #-}

vshabanov

{-# LANGUAGE TypeFamilies #-}

import Control.Concurrent
import Control.Concurrent.Async
import Control.Concurrent.Chan
import Control.Concurrent.MVar
import Control.Concurrent.QSem
import Control.Exception qualified as E
import Control.Monad
import Control.Monad.IO.Class

vshabanov

/* Experiment with a uniform representation of values that can hold
   references to C++ objects and be used polymorphically. Examples of
   the Rank2 IO monad dictionary and relation extend are provided at
   the bottom.

   The approach is the same as in dynamically typed languages.
   Primitive values are coupled with types at runtime. Functions only
   keep their arity to know when to apply. Empty values ([] or
   Nothing) are untyped.

vshabanov

// g++ -g -O3 -fopenmp -std=c++20 -Iboost_1_89_0 -Wall -Wpedantic hashtable.cpp && perf stat -e branches,branch-misses,bus-cycles,cache-misses,cache-references,cycles,instructions,ref-cycles ./a.out
#include <atomic>
#include <functional>
#include <utility>
#include <string>
#include <vector>
#include <stdexcept>
#include <map>
#include <unordered_map>
#include <bit>

vshabanov

// g++ -g -fopenmp -O2 -std=gnu++2c -Wall -Wpedantic -Iboost_1_89_0 Graph.cpp && ./a.out

/* [WIP]

   TODO

   - removing unused nodes?
     - ref data use case -- evaluate immediately and only keep results
     - exceptions in recipes -- do not keep unused nodes
     - recipes that add nodes but do not use them