summaryrefslogtreecommitdiffstats
path: root/coccinelle/run-coccinelle.sh
diff options
context:
space:
mode:
authorYu Watanabe <watanabe.yu+github@gmail.com>2023-12-25 21:45:58 +0100
committerGitHub <noreply@github.com>2023-12-25 21:45:58 +0100
commitfe3fcb94924877a362bc9799e0c8092be876c9a8 (patch)
tree1ba5d6cbdd09f936923c66ea542ec76f9df1863e /coccinelle/run-coccinelle.sh
parentcore/unit: don't log 0 values in unit_log_resources (diff)
parentsystemctl: use SYNTHETIC_ERRNO() (diff)
downloadsystemd-fe3fcb94924877a362bc9799e0c8092be876c9a8.tar.xz
systemd-fe3fcb94924877a362bc9799e0c8092be876c9a8.zip
Merge pull request #30633 from mrc0mmand/cocci-shenanigans
coccinelle: rework how we run the Coccinelle transformations
Diffstat (limited to 'coccinelle/run-coccinelle.sh')
-rwxr-xr-xcoccinelle/run-coccinelle.sh51
1 files changed, 47 insertions, 4 deletions
diff --git a/coccinelle/run-coccinelle.sh b/coccinelle/run-coccinelle.sh
index cd951790b9..bb72a493f0 100755
--- a/coccinelle/run-coccinelle.sh
+++ b/coccinelle/run-coccinelle.sh
@@ -2,6 +2,14 @@
# SPDX-License-Identifier: LGPL-2.1-or-later
set -e
+# FIXME:
+# - Coccinelle doesn't like our TEST() macros, which then causes name conflicts; i.e. Cocci can't process
+# that TEST(xsetxattr) yields test_xsetxattr() and uses just xsetxattr() in this case, which then conflicts
+# with the tested xsetxattr() function, leading up to the whole test case getting skipped due to
+# conflicting typedefs
+# - something keeps pulling in src/boot/efi/*.h stuff, even though it's excluded
+# - Coccinelle has issues with some of our more complex macros
+
# Exclude following paths from the Coccinelle transformations
EXCLUDED_PATHS=(
"src/boot/efi/*"
@@ -10,13 +18,17 @@ EXCLUDED_PATHS=(
# Symlinked to test-bus-vtable-cc.cc, which causes issues with the IN_SET macro
"src/libsystemd/sd-bus/test-bus-vtable.c"
"src/libsystemd/sd-journal/lookup3.c"
+ # Ignore man examples, as they redefine some macros we use internally, which makes Coccinelle complain
+ # and ignore code that tries to use the redefined stuff
+ "man/*"
)
TOP_DIR="$(git rev-parse --show-toplevel)"
+CACHE_DIR="$(dirname "$0")/.coccinelle-cache"
ARGS=()
# Create an array from files tracked by git...
-mapfile -t FILES < <(git ls-files ':/*.[ch]')
+mapfile -t FILES < <(git ls-files ':/*.c')
# ...and filter everything that matches patterns from EXCLUDED_PATHS
for excl in "${EXCLUDED_PATHS[@]}"; do
# shellcheck disable=SC2206
@@ -37,12 +49,43 @@ fi
[[ ${#@} -ne 0 ]] && SCRIPTS=("$@") || SCRIPTS=("$TOP_DIR"/coccinelle/*.cocci)
+mkdir -p "$CACHE_DIR"
+echo "--x-- Using Coccinelle cache directory: $CACHE_DIR"
+echo "--x--"
+echo "--x-- Note: running spatch for the first time without populated cache takes"
+echo "--x-- a _long_ time (15-30 minutes). Also, the cache is quite large"
+echo "--x-- (~15 GiB), so make sure you have enough free space."
+echo
+
for script in "${SCRIPTS[@]}"; do
echo "--x-- Processing $script --x--"
TMPFILE="$(mktemp)"
echo "+ spatch --sp-file $script ${ARGS[*]} ..."
- parallel --halt now,fail=1 --keep-order --noswap --max-args=20 \
- spatch --macro-file="$TOP_DIR/coccinelle/macros.h" --smpl-spacing --sp-file "$script" "${ARGS[@]}" ::: "${FILES[@]}" \
- 2>"$TMPFILE" || cat "$TMPFILE"
+ # A couple of notes:
+ #
+ # 1) Limit this to 10 files at once, as processing the ASTs is _very_ memory hungry - e.g. with 20 files
+ # at once one spatch process can take around 2.5 GiB of RAM, which can easily eat up all available RAM
+ # when paired together with parallel
+ #
+ # 2) Make sure spatch can find our includes via -I <dir>, similarly as we do when compiling stuff
+ #
+ # 3) Make sure to include includes from includes (--recursive-includes), but use them only to get type
+ # definitions (--include-headers-for-types) - otherwise we'd start formating them as well, which might be
+ # unwanted, especially for includes we fetch verbatim from third-parties
+ #
+ # 4) Use cache, since generating the full AST is _very_ expensive, i.e. the uncached run takes 15 - 30
+ # minutes (for one rule(!)), vs 30 - 90 seconds when the cache is populated. One major downside of the
+ # cache is that it's quite big - ATTOW the cache takes around 15 GiB, but the performance boost is
+ # definitely worth it
+ parallel --halt now,fail=1 --keep-order --noswap --max-args=10 \
+ spatch --cache-prefix "$CACHE_DIR" \
+ -I src \
+ --recursive-includes \
+ --include-headers-for-types \
+ --smpl-spacing \
+ --sp-file "$script" \
+ "${ARGS[@]}" ::: "${FILES[@]}" \
+ 2>"$TMPFILE" || cat "$TMPFILE"
+ rm -f "$TMPFILE"
echo -e "--x-- Processed $script --x--\n"
done