test(codegen): update scratch lifecycle gate for native-root decision

Author: Ralph Küpper

changelog.d/8072-inprocess-scratch-dir-leak.md

@@ -1,31 +1,8 @@
 ### Fixed
 
-- **In-process LLVM compiles no longer leak an empty scratch directory per compile.**
-  `compile_ll_inprocess_in`'s statepoint arm is the only in-process path that
-  *creates* the per-compile scratch directory — it writes the assembly there
-  before the compact-map rewrite — but its cleanup removed the two files it knew
-  about and left the directory. One `perry_llvm_scratch_<pid>_<counter>` husk
-  survived every compile.
-
-  The clang path has always done `remove_dir_all`, under a comment saying "the
-  directory cannot survive as an empty husk". Nothing went wrong while clang was
-  the default; the leak began when the in-process backend became it, and the
-  cleanup that only ever existed on the other path stopped running.
-
-  This is what turned `repsel-census` red on `main`: the job's Temp-directory
-  hygiene step reported `58 compile(s) left 58 entries in a temp directory that
-  started empty`. The leak is counted in *compiles* rather than in distinct IR,
-  which is what distinguishes it from the earlier `#7144` leak.
-
-  The consequence was larger than a pile of empty directories. Temp-directory
-  hygiene runs before `Promotion census` and `Sabotage check`, so both were
-  skipped for every red run — the census gate itself had not executed on `main`
-  since 2026-08-04. Because `repsel-census` is not a required status check, eight
-  consecutive red nightlies blocked nothing and nobody was forced to notice.
-
-  The regression test pins the statepoint lowering with `NativeRootsPin::native()`
-  rather than trusting the host default. Only the statepoint backends route
-  through assembly, and only that arm creates the directory, so without the pin
-  the test passes on a host that never enters the arm — the vacuous shape this
-  repo has been bitten by before. Verified by sabotage: restoring the two
-  `remove_file` calls turns the test red.
+- **Successful in-process LLVM statepoint compiles no longer leave an empty
+  scratch directory behind.** Cleanup now removes the per-compile directory as
+  a unit, matching the clang path and preventing temporary-directory growth
+  proportional to the number of compiles. A focused regression explicitly
+  selects native-root lowering and verifies that repeated compiles leave no
+  scratch entries.

crates/perry-codegen/src/linker_temp_lifecycle_tests.rs

@@ -455,17 +455,15 @@ fn inprocess_statepoint_compile_leaves_no_empty_scratch_dir() {
     // about and left the directory — harmless while clang was the default, a
     // leak per compile once in-process became it.
     //
-    // The pin is what makes this test non-vacuous. Only the statepoint
-    // backends route through assembly, and only that arm CREATES the scratch
-    // dir; without the pin this passes on a host that never enters the arm at
-    // all, which is the shape of gate the repo keeps getting bitten by.
+    // The explicit native-roots decision is what makes this test non-vacuous.
+    // Only the statepoint backends route through assembly, and only that arm
+    // CREATES the scratch dir; `false` would exercise a path that never creates
+    // the directory this regression is meant to observe.
     let Some(root) = temp_root_if_clang_available("inprocess_statepoint") else {
         return;
     };
-    let _pin = crate::codegen::helpers::NativeRootsPin::native();
-
     for nth in 0..3 {
-        let bytes = compile_ll_inprocess_in(&root, &test_ir(100 + nth), None, CLEAN)
+        let bytes = compile_ll_inprocess_in(&root, &test_ir(100 + nth), None, CLEAN, true)
             .unwrap_or_else(|e| panic!("in-process compile {nth} failed: {e:#}"));
         assert!(!bytes.is_empty(), "compile {nth} produced no object bytes");
         assert_eq!(