haskell - Ambiguous type variable when updating two record fields at a time (but not when updating one!) -

the following code fails "ambiguous type variable" error (bottom). however, alternate definition updates record twice works fine: why this? further, "broken" definition below similar 1 in trifecta source. i'm compiling ghc 7.10.3 against trifecta 1.5.2 , parsers 0.12.3.

module main  import text.trifecta import text.parser.token.style t  -- definition causes type error identstyle :: tokenparsing m => identifierstyle m identstyle =   t.emptyidents     { _stylestart    = letter     , _styleletter   = letter     } 

here working (alternative) definition

identstyle :: tokenparsing m => identifierstyle m identstyle = t.emptyidents { _stylestart = letter } { _styleletter = letter } 

the error generated first definition is:

main.hs:10:3:     not deduce (tokenparsing t0)       arising use of ‘emptyidents’     context (tokenparsing m)       bound type signature                  identstyle :: tokenparsing m => identifierstyle m       @ main.hs:8:15-49     type variable ‘t0’ ambiguous     note: there several potential instances:       instance attoparsec-0.13.0.1:data.attoparsec.internal.types.chunk                  t =>                tokenparsing                  (attoparsec-0.13.0.1:data.attoparsec.internal.types.parser t)         -- defined in ‘text.parser.token’       instance tokenparsing text.parsercombinators.readp.readp         -- defined in ‘text.parser.token’       instance tokenparsing m => tokenparsing (unhighlighted m)         -- defined in ‘text.parser.token’       ...plus 11 others     in expression: emptyidents     in expression:       emptyidents {_stylestart = letter, _styleletter = letter}     in equation ‘identstyle’:         identstyle           = emptyidents {_stylestart = letter, _styleletter = letter} failed, modules loaded: none. 

hah, kind of funny problem.

the issue here emptyidents class-polymorphic. when use it, part of type inference algorithm has define instance use.

when modify 1 field @ time, type of record isn't permitted change; is, type of \record -> record { _stylestart = undefined } identifierstyle m -> identifierstyle m. demanding final type of

emptyidents { _stylestart = letter } { _styleletter = letter } 

is identifierstyle m, can infer first emptyidents identifierstyle m with same type m argument.

on other hand, how record update works in haskell, when update both fields @ once (which happens fields types mention type argument m), update becomes polymorphic. is, type of \record -> record { _stylestart = undefined, _styleletter = undefined } identifierstyle m' -> identifierstyle m -- note prime!

so in case both updates @ once like

emptyidents { _stylestart = letter, _styleletter = letter } 

then fixing final type of update @ identifierstyle m not determine type of emptyidents.

there's half dozen ways fix this, basic idea should fix instance use while building emptyidents.