[pyrex41]

Two low-risk fixes from the capcov cross-port run, one commit each.

Fixes #86
Fixes #87

#86: relative file names resolve against *home-directory* (82afe27)

(cd Dir) sets *home-directory* (sys.kl:164), but open and the native fast paths for read-file, read-file-as-string, read-file-as-bytelist and the fasl load wrapper passed relative names straight to io.open, so they still read from the process cwd.

• New P.home_path(name) in prims.lua prepends *home-directory* to every name, as a plain string join. open and all four fast paths use it. The fast paths therefore can no longer open a same-named file in the cwd that open would not.
• The fasl load wrapper (boot.lua) reads and records the resolved path.
• Absolute names (decided, c349fff): *home-directory* is prepended unconditionally, matching shen-scheme d037ae6 and shen-go 1f587de. After (cd "/a"), (read-file "/b/x.shen") opens /a//b/x.shen and fails, as on both of them.

#87: kernel cache moves to the per-user cache dir (7ad7ae4)

When SHEN_KERNEL_CACHE was unset, cache_path() returned the relative .shen-kernel-cache.<build>.bin. Every directory shen ran from got about 1.8 MB of cache plus the .stlib sidecar, and recompiled the kernel.

• The default is now ${XDG_CACHE_HOME:-$HOME/.cache}/shen-lua/kernel-<build>.bin, with the sidecar next to it. The directory is created on first write (open_for_write).
• If the location is unwritable, caching is skipped, as before. If neither XDG_CACHE_HOME nor HOME is set, there is no cache.
• SHEN_KERNEL_CACHE=<path>|off is unchanged. The fasl default (~/.cache/shen-lua-fasl, SHEN_FASL_DIR) is untouched.
• The old cleanup of ./.shen-kernel-cache.bin only ran for the relative default, so it was removed along with it. Stale .shen-kernel-cache.* files in project dirs are gitignored and can be deleted.

Tests

Base is e3dec8b (main) and head is 7ad7ae4. Commands:

make test      # scripts/run-tests.lua
make certify   # run-kernel-tests.lua

Both were run with GIT_DIR/GIT_WORK_TREE/GIT_INDEX_FILE unset, a private TMPDIR and a private XDG_CACHE_HOME.

	make test	make certify (43.1 suite)
base e3dec8b	TOTAL: 1210 pass, 0 fail across 23 specs	passed ... 154 / failed ... 0
head 7ad7ae4	TOTAL: 1225 pass, 0 fail across 23 specs	passed ... 154 / failed ... 0

New checks (15):

• test/io_spec.lua (+7). cd, then each of read-file, read-file-as-string, read-file-as-bytelist, open and load, with a decoy file of the same name in the cwd. Also an absolute name after cd (joined, so it fails as on shen-scheme), and (cd "") restoring cwd resolution. On base the spec dies at the first open (open: cannot open ...).
• test/boot_cache_spec.lua (+8). Boot in an empty cwd with SHEN_KERNEL_CACHE unset: nothing lands in the cwd, the cache and sidecar land in $XDG_CACHE_HOME/shen-lua, else $HOME/.cache/shen-lua, a warm boot works, and an unwritable dir still boots. On base, 4 of these fail (got .shen-kernel-cache.bdd98551.bin .shen-kernel-cache.bdd98551.bin.stlib).

Repros from the issues, at head:

cd/repro.shen :: home=".../cd/sub/"|read-file: [[+ 1 2]]|open: ok|loaded data|load: ok|
cache-cwd: cwd listing: hello.shen home   (base: .shen-kernel-cache.bdd98551.bin .shen-kernel-cache.bdd98551.bin.stlib hello.shen home)
           D/home/.cache/shen-lua/kernel-bdd98551.bin(.stlib)

Cross-port check (capcov workloads)

Payloads were extracted with capcov's bench.py payload filter and compared byte for byte with shen-go 1f587de (kernel 42.2).

• f-modelcheck does (cd ".../model/") and then (read-file "shen/load.shen").
  • On base it fails with shen: open: cannot open shen/load.shen (shen/load.shen: No such file or directory).
  • On this branch alone it is identical to shen-go (sha256 93ccb82e6c2849e0…, all 25 MC lines).
• I also ran b-typed, c-untyped, d-g5, a2-libs-typed, e-draft and f-modelcheck on a local merge of this PR with the Generated Lua exceeds LuaJIT's 200-locals limit: datatype rules with nested conclusion patterns, list literals with 200+ elements #84/Reader fails on list literals over ~1550 elements with a misleading reader error near here: (suspected Lua stack overflow) #88 and Native Prolog engine overflows the Lua stack on modest typecheck depth (~400), where the legacy engine reaches about 1000 and shen-go has no limit #85 PRs. All six are identical to shen-go 1f587de, so there are no kernel 43.1 vs 42.2 differences to classify.

🤖 Generated with Claude Code