tests: bonsai/npc.mk: fix harrowing ordeal of a Linux error

tests/bonsai/npc.mk

@@ -32,11 +32,30 @@ npc_ascii: npc_ascii_controls npc_ascii_symbols npc_ascii_uppers # \
 .PHONY: npc_ascii_controls
 # (control characters)
 npc_ascii_controls:
+	# The following test prints the bytes 0x00 (inclusive) through 0x20
+	# (exclusive) and pipes them through npc(1). npc(1) should then replace all
+	# non-printing, non-space (in the isspace(3p) sense) characters with their
+	# graphical carat-char counterparts (see the npc(1) man page). The head(1p)
+	# invocation then strips off everything past the first line (or past the
+	# first newline byte, 0x0A) and xargs(1p) is used to test(1p) the output
+	# against the known good answer.
+
+	# Immediately before that newline, 0x09 is printed - in ASCII, the
+	# horizontal tab. If xargs' -I option is used, tr(1p) should used to delete
+	# that tab. If the tab is left as part of input, OpenBSD's xargs(1)
+	# implementation has been observed to strip it along with the other
+	# trailing whitespace (the newline), but Busybox's and GNU's xargs(1)
+	# implementations have been observed to leave the tab in. All three
+	# implementations strip off the trailing tab if `-I` is not used. The POSIX
+	# specification for `-I` is ambiguous as to which behavior is correct.
+	# This comment is the result of much bewilderment and debugging.
+
 	# ASCII 0x00 to 0x0a (before the newline, due to xargs(1p) issues)
 	awk 'BEGIN{ for (i = 0; i < 32; ++i) printf("%c", i); }' \
 		| $(BIN)/npc \
 		| head -n 1 \
-		| xargs -I out test "^@^A^B^C^D^E^F^G^H" = out
+		| tr -d '\t' \
+		| xargs test "^@^A^B^C^D^E^F^G^H" =
 
 	# ASCII 0x0a (otherwise the head|tail sequence won't work) to 0x1f
 	awk 'BEGIN{ for (i = 0; i < 32; ++i) printf("%c", i); print }' \