From 421487affa84bf44a3de1fdb42b191fe1ac81ac6 Mon Sep 17 00:00:00 2001 From: bong-water-water-bong Date: Sun, 8 Nov 2026 13:07:51 -0400 Subject: [PATCH] =?UTF-8?q?pkg-proof:=20build=20a=20module=20against=20OUR?= =?UTF-8?q?=20headers=20package=20=E2=80=94=20the=20acceptance=20test=20#1?= =?UTF-8?q?487=20asks=20for?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit #1487 (the build-machine paths in the headers tree) closed with the right answer: two of the four offending files are REQUIRED — 'make prepare' runs tools/objtool and scripts/Makefile.lib uses it on every module compiled — so the fix is a prefix map at build time, and the acceptance test is a DKMS build against the mapped headers. That test was missing here, and it is not the same thing as what the harness already did: it installed our headers package and read a path out of it, but never compiled anything with them. So a future tidy that 'excludes the four files with /home paths in them' — which is literally what my own first issue comment proposed — would break every DKMS build on a user's machine while every row above stayed green. New row: dkms build -m amdxdna-1bit -v 0.18.0 -k 7.4.0-rc1-1bit, i.e. against the headers from linux-headers-7.4.0-rc1-1bit rather than the container's own. Measured: deb 34/34 rows PASS, rc=0, with 'a DKMS build works against OUR headers package amdxdna-1bit/0.18.0 for 7.4.0-rc1-1bit'. (The first run of this branch pointed at a package directory built before #1495's rename and failed 4 rows on the old 99- file name — the collision rows doing their job on packages that predate the fix, which is worth having seen.) Refs #1280, #1487. --- scripts/pkg-proof.sh | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/scripts/pkg-proof.sh b/scripts/pkg-proof.sh index f3e8e0afd..7f7e6e873 100755 --- a/scripts/pkg-proof.sh +++ b/scripts/pkg-proof.sh @@ -318,6 +318,17 @@ if [ -r "$img" ]; then else row FAIL "vmlinuz in the package" "$img missing" fi + # The acceptance test #1487 asks for, and a guard nothing had: build a module against OUR headers + # package. Installing them and reading a path out of them is not the same thing — `make prepare` runs + # tools/objtool and scripts/Makefile.lib uses it on every module compiled, so a future "tidy" that + # excludes it (it is one of the four files carrying build-machine paths) would break every DKMS build + # on a user's machine and every row above would still pass. + dkms build -m amdxdna-1bit -v 0.18.0 -k "$krel" >/tmp/hb.log 2>&1 + if [ $? -eq 0 ]; then + row PASS "a DKMS build works against OUR headers package" "amdxdna-1bit/0.18.0 for $krel" + else + row FAIL "a DKMS build works against OUR headers package" "$(grep -m2 -E 'error|Error' /tmp/hb.log | tr '\n' ' ')" + fi fi # [ -n "$krel" ] fi # have_kernel