Suboptimal parseInstalledPackageInfo performance

[blackgnezdo]

My search for ghc-pkg memory consumption described in https://ghc.haskell.org/trac/ghc/ticket/9885 led me to parseInstalledPackageInfo. The following simple program will eat up 8G of memory:

import Distribution.InstalledPackageInfo

main = print $ length $ show $ parseInstalledPackageInfo x
where x = "ld-options: x" ++ concat (take 20000 $ repeat ","x" ")

The memory profile looks something like this:
(953)>>=...(...)
>>=..
>>=.
>>=
many1
many
sepBy1
sepBy
parseSepList
parseOptCommaList
listField
installedFieldDescrs
accumFields.setField
accumFields
parseFieldsFlat.
parseFieldsFlat
parseInstalledPackageInfo
parsePackageInfo
registerPackage
runit
main 794023536

[23Skidoo]

Bad performance of the parser is a known issue. @dcoutts has written a new Parsec-based parser, but it hasn't been merged yet.

[blackgnezdo]

@dcoutts, I wouldn't mind applying the patch locally if it's easy enough to backport to ghc 7.8.3.

[blackgnezdo]

My workaround for this was to split the long comma separated list of ld-options into multiple ld-options with one item per line as the following program doesn't exhibit the pathology:

import Distribution.InstalledPackageInfo

main = print $ length $ show $ parseInstalledPackageInfo $ unlines $ take 20000 (repeat "ld-options: x")

[phadej]

Fixed in #5017