When S=${WORKDIR}, cmake.eclass would create its build directory (if CMAKE_USE_DIR
is unset) above WORKDIR(!) as ${WORKDIR}_build. Creating directories above WORKDIR is not legal.

Portage has its own bug (bug #889418) in that it doesn't clean up unknown directories
above WORKDIR in PORTAGE_TMPDIR, so combined, you get a problem where "ebuild ... clean" doesn't
actually clean things up at all, and you get very confusing issues if e.g. changing
CC between runs.

The explicit S=WORKDIR check isn't truly needed but it makes explicit our intent here.

Bug: https://bugs.gentoo.org/889418
Closes: https://bugs.gentoo.org/889420
Signed-off-by: Sam James <sam@gentoo.org>
---
eclass/cmake.eclass | 9 +++++++++
1 file changed, 9 insertions(+)

diff --git a/eclass/cmake.eclass b/eclass/cmake.eclass
index 1cdbc123a243..4050beb22ba3 100644
--- a/eclass/cmake.eclass
+++ b/eclass/cmake.eclass
@@ -293,6 +293,15 @@ _cmake_check_build_dir() {
BUILD_DIR="${CMAKE_USE_DIR}"
else
: "${BUILD_DIR:=${CMAKE_USE_DIR}_build}"
+
+ # Avoid creating ${WORKDIR}_build (which is above WORKDIR).
+ # TODO: For EAPI > 8, we should ban S=WORKDIR for CMake.
+ # See bug #889420.
+ if [[ ${S} == ${WORKDIR} && ${BUILD_DIR} == ${WORKDIR}_build ]] ; then
+ eqawarn "QA notice: S=WORKDIR is deprecated for cmake.eclass."
+ eqawarn "Please relocate the sources in src_unpack."
+ BUILD_DIR="${WORKDIR}"/${P}_build
+ fi
fi

einfo "Source directory (CMAKE_USE_DIR): \"${CMAKE_USE_DIR}\""
--
2.41.0

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On Mon, 26 Jun 2023, Sam James wrote:

+
+ # Avoid creating ${WORKDIR}_build (which is above WORKDIR).
+ # TODO: For EAPI > 8, we should ban S=WORKDIR for CMake.
+ # See bug #889420.
+ if [[ ${S} == ${WORKDIR} && ${BUILD_DIR} == ${WORKDIR}_build ]] ; then

I'd suggest adding quotes to the RHS of the expression, to prevent
globbing.

But I think what you really want is to check whether ${BUILD_DIR}
(whatever its name is) is a subdirectory of ${WORKDIR}? Maybe a test
like this would make that intent clearer:

if [[ ${BUILD_DIR} != "${WORKDIR}"/* ]]; then

+ eqawarn "QA notice: S=WORKDIR is deprecated for cmake.eclass."
+ eqawarn "Please relocate the sources in src_unpack."
+ BUILD_DIR="${WORKDIR}"/${P}_build
+ fi

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Hi,

On 2023/06/26 12:36, Ulrich Mueller wrote:

On Mon, 26 Jun 2023, Sam James wrote:

+
+ # Avoid creating ${WORKDIR}_build (which is above WORKDIR).
+ # TODO: For EAPI > 8, we should ban S=WORKDIR for CMake.
+ # See bug #889420.
+ if [[ ${S} == ${WORKDIR} && ${BUILD_DIR} == ${WORKDIR}_build ]] ; then

I'd suggest adding quotes to the RHS of the expression, to prevent
globbing.

But I think what you really want is to check whether ${BUILD_DIR}
(whatever its name is) is a subdirectory of ${WORKDIR}? Maybe a test
like this would make that intent clearer:

if [[ ${BUILD_DIR} != "${WORKDIR}"/* ]]; then

BUILD_DIR="${WORKDIR}/../build"

I know it's pathological ... but still.  readlink -f should be
considered here unless it can be guaranteed that BUILD_DIR will not
contain .. components at this stage.

Kind Regards,
Jaco

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On Mon, 26 Jun 2023, Jaco Kroon wrote:

if [[ ${BUILD_DIR} != "${WORKDIR}"/* ]]; then

BUILD_DIR="${WORKDIR}/../build"

Oh right, I hadn't thought about the case that the ebuild would override
it. (AFAICS cmake.eclass itself doesn't do .. in BUILD_DIR.)

I know it's pathological ... but still.  readlink -f should be
considered here unless it can be guaranteed that BUILD_DIR will not
contain .. components at this stage.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Ulrich Mueller <ulm@gentoo.org> writes:

On Mon, 26 Jun 2023, Sam James wrote:

+
+ # Avoid creating ${WORKDIR}_build (which is above WORKDIR).
+ # TODO: For EAPI > 8, we should ban S=WORKDIR for CMake.
+ # See bug #889420.
+ if [[ ${S} == ${WORKDIR} && ${BUILD_DIR} == ${WORKDIR}_build ]] ; then

I'd suggest adding quotes to the RHS of the expression, to prevent
globbing.

I'm fine with quoting the RHS, but I don't think it's worth making the
other changes given this works and it's already an edge case.

-----BEGIN PGP SIGNATURE-----

iOUEARYKAI0WIQQlpruI3Zt2TGtVQcJzhAn1IN+RkAUCZJmtuF8UgAAAAAAuAChp c3N1ZXItZnByQG5vdGF0aW9ucy5vcGVucGdwLmZpZnRoaG9yc2VtYW4ubmV0MjVB NkJCODhERDlCNzY0QzZCNTU0MUMyNzM4NDA5RjUyMERGOTE5MA8cc2FtQGdlbnRv by5vcmcACgkQc4QJ9SDfkZDglgD/fN4/k3Klbwuzz2h5sxN4IAVYi4kkDEZlpAaf tQHwWEcBAPdNdKPO4rrXYP4zLlLNpn0iialVcEFYsnNgwOIETnUI
=tVoU
-----END PGP SIGNATURE-----

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)