GeezerVison

LLM Description written by Claude Code:claude-opus-5-5

Fixes #845

Summary

Every crawler probe (gem env gemdir, python3 --version, npm root -g, composer global config home, ...) now runs under a 10 s deadline through one utils::process::output_within primitive. A wedged toolchain shim used to hang scan, and every other crawling command, forever with no output. The three sites that hand-rolled their own tokio::time::timeout + kill_on_drop (pipenv, hatch, self-update sanity_exec) now use the same primitive, and their inline blocks are deleted.

Why

• Issue #845, register row C48 (register), living document doc/07-infra-agent.md (process spawning).
• Leverage: B 1 (p1 hang), U 1 (the async-CommandRunner and npm_dir slices reuse the primitive), D ≈3 (three duplicate timeout blocks plus the unbounded runner, all on one primitive), R M. Score ≈5.
• Why this slice: it bounds run_resolved in utils/process.rs, so it fixes every probe without editing the crawler files that Fix hosted gem redirect ignoring Bundler mirror.all (#681) #684, sbt, Mill and scala-cli support in agent, hosted and vendored modes #690, Fix Hatch environments being invisible to stale-install checks and VEX (#335) #700 and Fix gem pair model ignoring custom lockfile and Bundler 1 twins (#749, #751) #768 change.

What changed

• utils::process::output_within(Command, budget) -> Result<Output, BoundedError>: null stdin, captured stdout, discarded stderr. stdout is read on a thread under recv_timeout, then the child's exit is polled until the deadline. At the deadline the child is killed and reaped, and the call returns TimedOut without waiting for a grandchild that still holds the pipe (a #!/bin/sh shim's sleep).
• PROBE_TIMEOUT = 10 s. That is the budget pipenv, hatch and sanity_exec already used.
• run_resolved, behind SystemCommandRunner and GlobalProbeRunner, goes through output_within. A probe that times out answers None ("no information"), the same as a failed probe, and logs one --debug line.
• pipenv::installed_major, pypi_hatch::require_environment_context_support_with and update::download::sanity_exec build a std::process::Command and run it through run_blocking(output_within). sanity_exec keeps its ETXTBSY retry loop around the call and its "hung during its --version self-check" error.
• Base-red port (7b2eabd): Route Gradle digests through utils::digest #878's fix, verbatim, so main's failing digest ratchet passes here. It moves the inline digests in gradle_cache.rs, jvm_jar.rs and sidecars/maven.rs onto utils::digest, and no-ops once Route Gradle digests through utils::digest #878 merges.

Deleted

Refactor commit only (excluding the #878 port): 4 files, +229 / −34.

• Production: ≈ +124 / −34. The deleted lines are the three tokio::time::timeout(..., cmd.output()) + kill_on_drop blocks and the unbounded command.output() in run_resolved.
• Tests: ≈ +105.

Behavior

• Crawler probes: used to wait with no limit. Now a probe that runs longer than 10 s is killed and treated as absent.
• Children's stderr: now goes to the null device instead of a pipe that was captured and then dropped. pipenv and hatch already threw their stderr away, and sanity_exec already nulled it.
• Unchanged: output, exit codes, JSON and the CLI contract.

Test evidence

• New tests in utils::process:
  • a_hung_probe_answers_none_within_its_budget: a gem shim that runs exec sleep 30, under a 300 ms budget. This can't be expressed on main, which has no budget.
  • output_within_does_not_wait_for_a_grandchild_holding_stdout.
  • output_within_reports_status_stdout_and_spawn_errors: stdout, stderr dropped, nonzero exit, null stdin, missing program.
• New test in pypi_hatch: a_hung_hatch_is_refused_within_the_probe_budget.
• The existing pipenv, hatch, sanity_exec (hang, ETXTBSY retry, wrong program) and process tests still pass.
• cargo test -p socket-patch-core --lib: everything passes except the 4 tests that fail on main too because the sandbox runs as root (copy_tree::relax_loop_must_not_traverse_symlinked_root, vlt_heal::an_unremovable_hidden_lock_keeps_every_store_entry, pypi_poetry::wire_write_failure_…, pypi_requirements::wire_failure_rolls_back_…). utils::digest::tests::production_digests_go_through_the_helpers is red on main @ 9c43dfc and green here after the Route Gradle digests through utils::digest #878 port.
• cargo test -p socket-patch-cli --all-features --test in_process_gem_apply --test cli_global_args --test in_process_alternate_installers: all pass.
• cargo clippy --workspace --all-features -- -D warnings: clean.
• CLI repro (debug build): a Bundler project (rack 2.2.8), with a gem shim on PATH that logs its args and runs exec sleep 3600, and the API pointed at a closed port. scan --json exits 1 after 10 s with the full JSON envelope (scannedPackages: 1). The shim log shows env gemdir and env gempath, and no sleep 3600 process is left afterwards. The issue's run of the same setup on main hung past 90 s with empty stdout.

Risk

M. Every probe's spawn path changes. The budgets match the old per-site values, and the crawler probes only gain a bound.

Remaining (#845 later slices)

• vendor/npm_dir.rs's git check-ignore exchange (after Fix vendored npm-family tarballs dropped by .gitignore (#831) #837, which edits that file).
• An async CommandRunner.
• The "no output() outside utils::process" architecture_tests guard.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Mr9d6u7vNrWPgoibe1ZRXd

Generated by Claude Code