fix(wasi): component get-directories returns a guest-usable preopen list (SR-37, #405)

[avrabe]

What

Fixes the get-directories layer of #405: a WASI Preview2 component run via kilnd --wasi --component --wasi-fs <dir> could not use any preopen — wasi:filesystem/preopens::get-directories returned a list backed by memory the guest never allocated, so preview2 libc registered no usable preopen and every file op returned ENOENT without ever calling open-at.

Root cause (measured, not guessed)

For the maintainer's single wasm32-wasip2 component:

• nested_component_instances.len() == 0 → the InterComponentHandler override does not fire (ruling out the dispatcher-swap hypothesis).
• the queried dispatcher already held the preopen (preopens.len() == 1).

So the defect was the canonical-ABI return encoding: get-directories wrote its list<tuple<descriptor,string>> entries at retptr+8 and path strings at retptr+20, but the return area for func() -> list<…> is only 8 bytes. The list must be backed by cabi_realloc'd guest memory — exactly how the working get-arguments path does it.

Fix (mirrors the proven get-arguments machinery, 5 sites)

• kiln-foundation HostImportHandler: get_preopens + set_preopens_allocation.
• kiln-runtime: pre_allocate_wasi_preopens + allocate_wasi_preopens_memory (N×12 entry buffer + per-path string buffers via cabi_realloc) + CapabilityAware wrapper.
• kiln-wasi WasiDispatcher: preopens_alloc; get-directories writes into the allocation and fails loud if it was not set (no more unowned scribble).
• kiln-component ComponentInstance::pre_allocate_wasi_preopens.
• kilnd: call it before the entry point, beside pre_allocate_wasi_args.

Verification

• 2 new unit oracles (// rivet: verifies SR-37): the encoding uses the allocation (header points at list_ptr, not retptr+8) and fails loud without it.
• Integration trace confirms the guest now registers the . preopen and successfully open-at + stat's a file under it (before: rejected immediately).

Scope / honesty

This lands the prerequisite. Full end-to-end file read for #405 is still blocked on SR-38 (input-stream.blocking-read writes list<u8> to a fixed unowned address — a distinct, harder defect needing on-demand cabi_realloc). #405 stays open; v0.4.1 will cut once SR-38 also lands.

Trace: SR-37

🤖 Generated with Claude Code

[github-actions]

🔍 Build Diagnostics Report

Summary

Metric	Base Branch	This PR	Change
Errors	0	0	0
Warnings	5	5	0

🎯 Impact Analysis

Issues in Files You Modified

• 0 new errors introduced by your changes
• 0 new warnings introduced by your changes
• 0 total errors in modified files
• 0 total warnings in modified files
• 0 files you modified

Cascading Issues (Your Changes Breaking Other Files)

• 0 new errors in unchanged files
• 0 new warnings in unchanged files
• 0 unchanged files now affected

Note: "Cascading issues" are errors in files you didn't modify, caused by your changes (e.g., breaking API changes, dependency issues).

✅ No Issues Detected

Perfect! Your changes don't introduce any new errors or warnings, and don't break any existing code.

---

📊 Full diagnostic data available in workflow artifacts

🔧 To reproduce locally:

# Install cargo-kiln
cargo install --path cargo-kiln

# Analyze your changes
cargo-kiln build --output json --filter-severity error
cargo-kiln check --output json --filter-severity warning

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!