> Similar reasoning applies to the -I${.CURDIR} in citrus/Makefile.inc
> which is responsible for the second -I/usr/src/lib/libc as far as I
> can see. Should such an -I not come from the top-level libc/Makefile.inc?

Doh! I had missed this one.

How about this better diff, then?

Index: Makefile
===================================================================
RCS file: /OpenBSD/src/lib/libc/Makefile,v
diff -u -p -r1.42 Makefile
--- Makefile	12 Nov 2019 04:38:27 -0000	1.42
+++ Makefile	6 Aug 2025 20:28:34 -0000
@@ -24,6 +24,9 @@ SYMBOL_LISTS=	${LIBCSRCDIR}/Symbols.list
 
 
 LIBCSRCDIR=${.CURDIR}
+# needed by a few files in gen/ and string/ to reach other directories, and
+# by sys/stack-protector.c to reach ../csu
+CFLAGS+=-I${LIBCSRCDIR}
 .include "${LIBCSRCDIR}/Makefile.inc"
 
 all: tags
Index: citrus/Makefile.inc
===================================================================
RCS file: /OpenBSD/src/lib/libc/citrus/Makefile.inc,v
diff -u -p -r1.2 Makefile.inc
--- citrus/Makefile.inc	5 Sep 2016 09:47:02 -0000	1.2
+++ citrus/Makefile.inc	6 Aug 2025 20:28:34 -0000
@@ -4,4 +4,3 @@
 .PATH: ${LIBCSRCDIR}/citrus
 
 SRCS+=	citrus_none.c citrus_utf8.c
-CFLAGS+=-I${.CURDIR}
Index: locale/Makefile.inc
===================================================================
RCS file: /OpenBSD/src/lib/libc/locale/Makefile.inc,v
diff -u -p -r1.28 Makefile.inc
--- locale/Makefile.inc	25 Aug 2023 12:45:45 -0000	1.28
+++ locale/Makefile.inc	6 Aug 2025 20:28:34 -0000
@@ -24,4 +24,4 @@ MAN+=	nl_langinfo.3 setlocale.3 newlocal
 	wcstol.3 wcstombs.3 wcsxfrm.3 wctob.3 wctomb.3 \
 	wctype.3 iswctype.3 wctrans.3 towctrans.3 wcwidth.3
 
-CFLAGS+=-I${.CURDIR} -I${LIBCSRCDIR}/citrus
+CFLAGS+=-I${LIBCSRCDIR}/citrus