This commit is contained in:
2026-08-18 09:20:44 +02:00
commit b826cd721a
522 changed files with 93730 additions and 0 deletions
+71
View File
@@ -0,0 +1,71 @@
# G3 Manual Test Setup
This setup supports only TC041-TC066, TC141, TC148, TC149, TC152, and
TC153. It prepares deterministic fixtures and native probes; it does not assign
runtime results.
## Host Preparation
From the `repo-pre-15` root, use an absent output directory:
```sh
git rev-parse HEAD
tests/prepare_g3_fixtures.sh /work/build/repo-pre15-g3
python3 -B tests/pre15/fixtures/g3.py verify \
--output /work/build/repo-pre15-g3
sha256sum /work/build/repo-pre15-g3/G3-MANIFEST.json
```
The maintained generator is `tests/pre15/fixtures/g3.py`. The compatibility
entry point `tests/g3_fixtures.py` delegates to it. Normal generation does not
read `tests/results/`; B01 separately proves that the migrated metadata and
final-review sets match both archived scripts byte for byte.
`G3-MANIFEST.json` records the generator SHA256, erofs-utils identity, exact
`mkfs.erofs` argv, canonical source-tree inventories, image hashes, and expected
directory sets. Record this manifest, all checksum files, the exact source
commit, and the KLD SHA256. Transfer the fixtures, sources, and only these native
probes to the FreeBSD 15 guest:
```text
readdir_probe.c g3_vfs_probe.c stat_special.c nfs_fh_tool.c
mmap_fault.c sparse_hole_probe.c
```
## Guest Preparation
Build the KLD natively from an archive exported from the recorded commit. Build
the probes directly:
```sh
cc -O2 -Wall -Wextra -Werror -std=c17 readdir_probe.c -o readdir_probe
cc -O2 -Wall -Wextra -Werror -std=c17 g3_vfs_probe.c -o g3_vfs_probe
cc -O2 -Wall -Wextra -Werror -std=c17 stat_special.c -o stat_special
cc -O2 -Wall -Wextra -Werror -std=c17 nfs_fh_tool.c -o nfs_fh_tool
cc -O2 -Wall -Wextra -Werror -std=c17 mmap_fault.c -o mmap_fault
cc -O2 -Wall -Wextra -Werror -std=c17 sparse_hole_probe.c \
-o sparse_hole_probe
```
Before the first test, require no EROFS mount, no md provider, and no stale
EROFS KLD. Load only the exact KLD under test. For each test, attach a fresh
dynamic vnode md, mount read-only, execute that numbered Markdown procedure,
then unmount and detach before continuing.
The report records the literal command, exit status or syscall errno, relevant
fixture/data/KLD hashes, dmesg delta, and cleanup state for every test.
## Common Mount Pattern
```sh
mkdir -p /tmp/repo-pre15-g3/mnt
unit=$(mdconfig -a -t vnode -f IMAGE)
mount -t erofs -o ro "/dev/$unit" /tmp/repo-pre15-g3/mnt
# Execute exactly one numbered test.
umount /tmp/repo-pre15-g3/mnt
mdconfig -d -u "$unit"
```
TC153 additionally creates and removes its documented multi-terabyte sparse
provider. Final cleanup requires zero EROFS mounts, zero owned md providers, the
test KLD unloaded, and all guest-generated sparse files removed.