Which shells support ANSI-C quoting? e.g. $'string'

Question

I have a shell script that uses the following to print a green checkmark in its output:

col_green="\e[32;01m"
col_reset="\e[39;49;00m"

echo -e "Done ${col_green}✓${col_reset}"

After reading about Bash's ANSI-C Quoting, I realized I could use it when setting my color variables and remove the -e flag from my echo.

col_green=$'\e[32;01m'
col_reset=$'\e[39;49;00m'

echo "Done ${col_green}✓${col_reset}"

This seems appealing, since it means the message prints correctly whether it's passed to Bash's builtin echo or the external util /bin/echo (I'm on macOS).

But does this make the script less portable? I know Bash and Zsh support this style of quoting, but I'm not sure about others.

Answer 1

$'…' is a ksh93 feature that is also present in zsh, bash, mksh, FreeBSD sh and in some builds of BusyBox sh (BusyBox ash built with ENABLE_ASH_BASH_COMPAT). It isn't present in the POSIX sh language yet. Common Bourne-like shells that don't have it include dash (which is /bin/sh by default on Ubuntu among others), ksh88, the Bourne shell, NetBSD sh, yash, derivatives of pdksh other than mksh and some builds of BusyBox.

A portable way to get backslash-letter and backslash-octal parsed as control characters is to use printf. It's present on all POSIX-compliant systems.

esc=$(printf '\033') # assuming an ASCII (as opposed to EBCDIC) system
col_green="${esc}[32;01m"

Note that \e is not portable. It's supported by many implementations of printf but not by the one in dash¹. Use the octal code instead.

¹ _{It is supported in Debian and derivatives that ship at least 0.5.8-2.4, e.g. since Debian stretch and Ubuntu 17.04.}

Answer 2

The degree of $'...' support also needs to be taken into consideration when porting. The POSIX Folks' proposal to put this in POSIX sh mentions one in particular:

stephane: ksh93 is the shell $'...' comes from (while $'\uxxxx' [and $'\Uxxxxxxxx'] comes from zsh: http://www.zsh.org/mla/workers/2003/msg00223.html) [^]

From what I got here on my Debian bullseye, the ksh2020 understands $'\U1F600'. This is the only "official" Korn shell I can get on this new distro.

mksh parsed it but botched it entirely with a U+FFFE. Since it did not complain about a syntax error there's gotta be something wrong with it's understanding of Unicode. It handles $'\U01F60' just fine.