Builder.fromString and Builder.singleton don't substitute surrogate code points

[minoki]

Unlike Data.Text.pack, Data.Text.Lazy.Builder.fromString doesn't perform replacement on surrogate code points. Therefore, one can create an invalid Text value with it.

Data.Text.Lazy.Builder.singleton also have the problem.

Example:

λ> import qualified Data.Text as T
λ> import qualified Data.Text.Lazy as TL
λ> import qualified Data.Text.Lazy.Builder as TB
λ> TL.pack "\xD800" -- correctly returns "\xFFFD"
"\65533"
λ> TL.fromChunks [T.pack "\xD800"] -- correctly returns "\xFFFD"
"\65533"
λ> TB.toLazyText (TB.fromString "\xD800") -- the Builder builds a Text containing [0xD800], whichi is invalid
"\9980"
λ> TB.toLazyText (TB.fromString "\xD800") -- a garbage is read when decoding [0xD800], thus a random value is shown
"\9571"
λ> TB.toLazyText (TB.singleton '\xD800') -- the Builder builds a Text containing [0xD800], whichi is invalid
"\10262"
λ> TB.toLazyText (TB.singleton '\xD800')
"\10269"

[hvr]

Good catch. This is indeed inconsistent semantics and falsifies the stated property in the docstring:

-- | /O(1)./ A @Builder@ taking a single character, satisfying
--
--  * @'toLazyText' ('singleton' c) = 'L.singleton' c@
--
singleton :: Char -> Builder