Demi Marie Obenour <demiobenour@gmail.com> writes:

> Use builtin string manipulation instead.
>
> Signed-off-by: Demi Marie Obenour <demiobenour@gmail.com>
> ---
>  scripts/make-erofs.sh | 14 +++++++++++++-
>  1 file changed, 13 insertions(+), 1 deletion(-)
>
> diff --git a/scripts/make-erofs.sh b/scripts/make-erofs.sh
> index cf942972910c76e1835dc5b0084c2d04bf084a9d..93cb3245f409b24c24be05e9307a1b2e12c867fe 100755
> --- a/scripts/make-erofs.sh
> +++ b/scripts/make-erofs.sh
> @@ -78,7 +78,19 @@ while read -r arg1; do
>  
>  	check_path "$arg1" "$arg2"
>  
> -	parent=$(dirname "$arg2")
> +	# The below simple version of dirname(1) can only handle
> +	# a subset of all paths, but this subset includes all of
> +	# the paths that check_path doesn't reject.

Are any of the paths it would mishandle paths that would actually be
likely to show up?  I feel like we don't really need to worry about
people putting silly things in the Makefile, especially since that's
going to be generated going forward, and in the case of Nix store paths
we know those will always be the store directory, a slash, and then a
single component.  I don't really want to be overly defensive,
especially since we're not in other places in the build system — as a
consequence of using make, which doesn't handle spaces well, for
example.

> +	case $arg2 in
> +	(*/*)
> +		# Create the parent directory if it doesn't already
> +		# exist.
> +		parent=${arg2%/*}
> +		;;
> +	(*)
> +		parent=.
> +		;;
> +	esac
>  	mkdir -p -- "$root/$parent"
>  	cp -RT -- "$arg1" "$root/$arg2"
>  done
>
> -- 
> 2.51.0