Explicitly set text encoding to utf-8?

[wz1000]

I've been running into a build failure with ghc-lib-parser-9.12.2.20250421, where on a fresh docker image I get

*** Exception: compiler/GHC/Parser.y: hGetContents: invalid argument (cannot decode byte sequence starting from 226`:

The locale on this image is:

LANG=
LANGUAGE=
LC_CTYPE="POSIX"
LC_NUMERIC="POSIX"
LC_TIME="POSIX"
LC_COLLATE="POSIX"
LC_MONETARY="POSIX"
LC_MESSAGES="POSIX"
LC_PAPER="POSIX"
LC_NAME="POSIX"
LC_ADDRESS="POSIX"
LC_TELEPHONE="POSIX"
LC_MEASUREMENT="POSIX"
LC_IDENTIFICATION="POSIX"
LC_ALL=

Setting LC_ALL=C.UTF-8 gets rid of the error, but since Haskell source files (and transitively happy files) are specified to be UTF-8, I wonder if the right thing to do could be to explicitly set the encoding for both happy input and output to UTF-8.

[phadej]

The right fix is to not use locale dependent hGetContents derivatives (like Prelude.readFile) at all.

c.f. when you read a json file -- which are specified to be UTF8 -- with aeson or about anything else, your results wont depend on the locale. I don't see how haskell/happy files are different. The situation would been different if no encoding were specified, but instead source files had to be treated as "text files" (then the locale tells what encoding is used).