b2ceb12 · typeable/aeson

git.iko.soy/typeable/aeson

Microbenchmark time! Let's compare direct string encoding vs via text

Turns out that going via text is actually *faster* -- unless you inline the direct string function, in which case it wins. Inlining the text function has no effect on performance.

commitb2ceb12849ee8cf904efdbf6d52a7c4160aa73b6 authorBryan O'Sullivan <bos@serpentine.com> dateMon 6 Jul 2015, 19:54 (11 years ago) parent7c3a89c

2 files changed+18−5

compare with master

unifiedsplit

Data/Aeson/Encode/Builder.hs+1 benchmarks/aeson-benchmarks.cabal+17−5

Data/Aeson/Encode/Builder.hs+1

@@ -96,6 +96,7 @@ string :: String -> Builder

9696string t = B.char8 '"' <> foldMap go t <> B.char8 '"'

9797 where go c | c > '\x7f' = B.charUtf8 c

9898 | otherwise = BP.primBounded escapeAscii (c2w c)

99+{-# INLINE string #-}

99100

100101escapeAscii :: BP.BoundedPrim Word8

101102escapeAscii =

benchmarks/aeson-benchmarks.cabal+17−5

@@ -16,15 +16,13 @@ library

1616 exposed-modules:

1717 Data.Aeson

1818 Data.Aeson.Encode

19− Data.Aeson.Parser

20− Data.Aeson.Types

21− Data.Aeson.TH

22−

23− other-modules:

2419 Data.Aeson.Encode.Builder

2520 Data.Aeson.Encode.Functions

2621 Data.Aeson.Functions

22+ Data.Aeson.Parser

2723 Data.Aeson.Parser.Internal

24+ Data.Aeson.TH

25+ Data.Aeson.Types

2826 Data.Aeson.Types.Class

2927 Data.Aeson.Types.Generic

3028 Data.Aeson.Types.Instances

@@ -57,6 +55,20 @@ library

5755 ghc-options: -O2 -Wall

5856 cpp-options: -DGENERICS

5957

58+executable aeson-benchmark-micro

59+ main-is: Micro.hs

60+ hs-source-dirs: ../examples .

61+ ghc-options: -Wall -O2 -rtsopts

62+ build-depends:

63+ aeson-benchmarks,

64+ base,

65+ bytestring,

66+ criterion >= 1.0,

67+ deepseq,

68+ ghc-prim,

69+ text,

70+ time

71+

6072executable aeson-benchmark-typed

6173 main-is: Typed.hs

6274 hs-source-dirs: ../examples .