Yet Another Shell

Fork

It seems that variables cannot hold byte values unless LC_CTYPE=C and it seems like it is not possible to change that for a script.

Consider this script test.sh:

 a="$(printf '\037\213\010')"
 b="$(printf '\037\213\020')"
  
 if [ "$a" = "$b" ]; then
         echo "shouldnt happen..."
 fi
  
 if [ "$(printf '\213')" = "$(printf '\214')" ]; then
         echo "this too shouldn't happen..."
 fi
  
 awk -v a="$(printf '\213')" -v b="$(printf '\214')" 'END { if (a==b) print "awk shouldnt happen"}' /dev/null

When executing it in dash, bash or busybox ash, it will not print anything. But with yash it prints:

ncopa-desktop:~$ yash test.sh
shouldnt happen...
this too shouldn't happen...
awk shouldnt happen
ncopa-desktop:~$ 

It seems that variables can only hold 7-bit ascii values.

Now, if I set LC_CTYPE=C when spawning yash, it does pass:

ncopa-desktop:~$ LC_CTYPE=C yash -c ". ./test.sh"
ncopa-desktop:~$ 

But if I set LC_CTYPE=C within the a spawned shell, it still fails:

ncopa-desktop:~$ yash -c "LC_CTYPE=C ; . ./test.sh"
shouldnt happen...
this too shouldn't happen...
awk shouldnt happen
ncopa-desktop:~$ 

This effectively means that it is impossible to write or execute portable shells for yash that uses 8-bit bytes in variables. There is no way a portable script can control what locale the user has set in its yashrc.

This happens on Alpine Linux which uses musl libc and I believe the default locale in musl is C.utf8.

This was discovered when debugging yash for alpine linux' tiny-cloud script: https://gitlab.alpinelinux.org/alpine/cloud/tiny-cloud/-/blob/fda9a350a1dfb4a33e9e4bf9e5272d5b4f74f541/lib/tiny-cloud/init-main#L69