core: the prepared environment -- RequestContext.env, registerPreparedEnv, and the $_SERVER merge

[melihucar]

Gate: default
Agent: opencode

Nothing in this port can deliver a per-request/per-worker environment variable
to PHP. Upstream calls this the prepared environment, and it is not a nicety:
it is the only mechanism by which FRANKENPHP_WORKER=1 reaches
$_SERVER['FRANKENPHP_WORKER'], which is what
vendor/frankenphp/testdata/_executor.php:4 branches on to decide whether to
call frankenphp_handle_request() at all. Without it, 44 upstream fixtures
cannot be run in worker mode, because every one of them silently takes the
regular-mode branch.

Split out of #14, which cannot implement this without touching four files
outside its lane.

Why there is no back door

$_SERVER in this SAPI is built only by frankenphp_register_variables
(vendor/frankenphp/frankenphp.c:1371-1383), which calls
go_register_server_variables and then
frankenphp_register_variables_from_request_info. The OS environment reaches
$_ENV and getenv() (frankenphp.c:1117 swaps php_import_environment_variables
for get_full_env), and never $_SERVER. So exporting FRANKENPHP_WORKER=1
from the process, or putenv()-ing it, does not work. The prepared env is the
mechanism, and there is no substitute.

What exists upstream

• frankenphp.c:127 — static THREAD_LOCAL HashTable *prepared_env.
• frankenphp_add_to_prepared_env(name, name_len, val, val_len, size)
  (frankenphp.c:1461-1469, declared frankenphp.h:205-206) — populates it.
  Upstream calls it from registerPreparedEnv (cgi.go), reached from
  go_update_request_info (cgi.go:295-297) when len(fc.env) != 0.
• frankenphp_merge_with_prepared_env(track_vars_array)
  (frankenphp.c:1270-1275, declared frankenphp.h:207) — copies it into
  $_SERVER. Upstream calls it from go_register_server_variables
  (cgi.go:185-187) last, after the known variables and the headers:
  "The Prepared Environment is registered last and can overwrite any previous
  values."
• Keys are NUL-terminated in the Go map (worker.go:138 writes
  "FRANKENPHP_WORKER\x00"), because registerPreparedEnv hands the key
  straight through to C. Decide deliberately whether our env map stores the
  NUL or appends it at the boundary, and say which in a comment — do not
  reproduce the Go representation by accident.

What to build

1. crates/frankenrust-sys/build.rs — additive only: two
   .allowlist_function entries, frankenphp_add_to_prepared_env and
   frankenphp_merge_with_prepared_env, with a comment naming this issue the
   way the existing frankenphp_update_local_thread_context entry names server: regular-thread handler, hyper HTTP/1.1 front end, and the async/pthread bridge #13.
   Hazard: core: worker mode -- the frankenphp_handle_request park/resume loop #14 adds one entry (frankenphp_shutdown_dummy_request) to the
   same list. Append yours at the end of the function block rather than
   interleaving with the existing comment blocks, so the two diffs do not
   collide on the same hunk.
2. crates/frankenrust-core/src/context.rs — RequestContext gains
   pub env: Option<...> (upstream fc.env, a PreparedEnv; nil and empty
   both mean "do not call into C", per the len(fc.env) != 0 guards at both
   call sites — pick one representation and pin it in the doc comment).
   Set through RequestContext::new's existing shape or a builder-style
   setter; do not break the existing four-argument constructor's call sites in
   crates/frankenrust-server/src/server.rs and the existing tests.
3. crates/frankenrust-core/src/callbacks/servervars.rs —
   go_update_request_info calls frankenphp_add_to_prepared_env once per
   entry when env is non-empty, before the rest of its work, mirroring
   cgi.go:295-297. The two "fc.env (PreparedEnv) is not part of this issue's RequestContext ... there is nothing to check" comments (around :584 and
   :639) are now false and must be replaced, not left standing.
4. crates/frankenrust-sys/shim.c's go_register_server_variables — call
   frankenphp_merge_with_prepared_env(track_vars_array) after the batch
   the Rust side supplied, guarded on the batch reporting a non-empty prepared
   env. Ordering is the behaviour: last wins.

Acceptance

• A unit test in servervars.rs in the shape of the existing slot tests:
  install a RequestContext carrying env = {"FRANKENPHP_WORKER": "1"}, call
  go_update_request_info, and assert the C side saw the add (a test double
  over the extern, or the same technique the neighbouring tests already use for
  C-facing calls — do not weaken an existing test to make room).
• An end-to-end test that actually reads $_SERVER from PHP: a fixture PHP
  script that echoes $_SERVER['FRANKENPHP_WORKER'] ?? 'unset', served with
  a RequestContext carrying that prepared env, asserting the body is 1; and
  the same script with no prepared env asserting unset. This is the criterion
  that can actually fail — the unit test alone cannot see the $_SERVER merge.
• A test pinning the ordering: a prepared-env key that collides with a
  known CGI variable (e.g. SERVER_SOFTWARE) overwrites it, not the reverse
  (cgi.go:186).
• bash scripts/gate.sh default passes.

Out of scope

• getenv()/$_ENV sandboxing (frankenphp.c:661-665's sandboxed_env),
  go_putenv, and the putenv()-per-worker-script-run release.
• WithRequestPreparedEnv as a public request-option API surface — a plain
  public field on RequestContext is enough for now.
• Worker mode itself (core: worker mode -- the frankenphp_handle_request park/resume loop #14), and wiring the worker's FRANKENPHP_WORKER=1 into
  a running server — that belongs to the server-side worker configuration
  issue, which consumes this.

Recoveries: 1

[melihucar]

Blocked by the loop.

Failed 3 attempts. Last failure:

--- gate: test-suite-intact
SELFTEST PASS (27 cases)
    PASS test-suite-intact
--- gate: orchestrator-runnable
    PASS orchestrator-runnable
--- gate: doc-citations
ok - missing file reports resolution failure
ok - start line past EOF is rejected
ok - end line past EOF is rejected
ok - reversed range is rejected
ok - zero start line is rejected
ok - negative start line is rejected
ok - zero end line is rejected
ok - negative end line is rejected
ok - basename matching two vendor subdirs is rejected
ok - vendor-root basename shadowing a subdir copy is rejected
ok - repo-root file shadowing a vendor file is rejected
ok - './' prefix resolves to the repo-root file and is not ambiguous
ok - same fixture cited bare (no './') is still ambiguous
ok - '..' traversal is rejected before touching disk
ok - symlink out of the repository is rejected
ok - anchor absent from its named doc is rejected
ok - bare-form doc citation satisfies a normalized full-form anchor
ok - anchor whose token left the cited range is rejected
ok - zero citations is a specific failure
ok - two docs with the same basename are rejected

SELFTEST PASS
checked 109 citations across 2 doc(s) and 10 anchor(s): all resolve
    PASS doc-citations
--- gate: dev-env
dev-env: 49 passed
    PASS dev-env
--- gate: build
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.06s
    PASS build
--- gate: fmt
    PASS fmt
--- gate: clippy
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.08s
    PASS clippy
--- gate: test
    Finished `test` profile [unoptimized + debuginfo] target(s) in 0.08s
     Running unittests src/lib.rs (target/debug/deps/frankenrust_core-27133b74fa98a933)

running 169 tests
test callbacks::input::tests::go_read_cookies_joins_multiple_cookie_headers_with_semicolon_space ... frankenrust[Level(4)]: putenv() was called, but frankenrust does not sandbox the process environment: the change is visible to this PHP interpreter's own getenv()/$_ENV, but not to any process this server spawns afterwards
frankenrust[Level(4)]: opcache_reset() was called, but frankenrust does not implement rebooting PHP threads to pick up a fresh opcache; the request continues without one
frankenrust[Level(4)]: mercure_publish() was called, but frankenrust is not built with Mercure support
ok
test callbacks::input::tests::go_read_cookies_with_a_single_empty_cookie_value_returns_null ... ok
test callbacks::input::tests::go_apache_request_headers_with_no_context_returns_null_and_zero ... ok
test callbacks::input::tests::go_read_cookies_with_no_context_returns_null ... ok
test callbacks::input::tests::go_read_cookies_with_no_cookie_header_returns_null ... ok
test callbacks::input::tests::go_apache_request_headers_layout_is_pair_count_with_2n_entries_and_joins_repeats ... ok
test callbacks::input::tests::go_read_cookies_with_no_request_returns_null ... ok
test callbacks::input::tests::go_read_cookies_strips_embedded_nul_bytes ... ok
test callbacks::input::tests::go_apache_request_headers_with_no_sink_returns_null_and_zero ... ok
test callbacks::input::tests::go_read_post_fills_the_buffer_completely_from_a_one_byte_source ... ok
test callbacks::input::tests::go_read_post_with_no_context_returns_zero ... ok
test callbacks::input::tests::go_read_post_with_no_sink_returns_zero ... ok
test callbacks::log::tests::disabled_level_does_not_format_the_message_or_build_attrs ... ok
test callbacks::log::tests::escape_for_log_line_disambiguates_distinct_invalid_utf8_bytes ... ok
test callbacks::log::tests::escape_for_log_line_escapes_newlines_so_a_message_cannot_forge_a_second_line ... ok
test callbacks::log::tests::escape_for_log_line_passes_printable_ascii_through_unchanged ... ok
frankenrust[Level(0)]: hello
test callbacks::log::tests::go_log_attaches_the_syslog_level_name ... ok
test callbacks::log::tests::go_log_captures_level_and_arbitrary_bytes ... ok
test callbacks::log::tests::out_of_range_syslog_level_lands_on_info_not_error ... ok
test callbacks::log::tests::go_log_tolerates_a_null_message ... ok
test callbacks::misc::tests::every_callback_in_this_module_tolerates_no_current_request ... ok
test callbacks::log::tests::log_once_does_not_consume_the_once_on_a_disabled_level ... ok
test callbacks::log::tests::syslog_level_mapping_matches_upstreams_table ... ok
test callbacks::misc::tests::go_is_context_done_is_false_for_a_live_context ... ok
test callbacks::misc::tests::go_is_context_done_is_true_once_the_context_is_marked_done ... ok
test callbacks::mainthread::tests::custom_php_ini_uses_libc_memory_and_honors_disable_timeouts ... ok
test callbacks::misc::tests::go_is_context_done_is_true_with_no_context_installed ... ok
test callbacks::misc::tests::go_putenv_returns_true ... ok
test callbacks::misc::tests::go_mercure_publish_returns_the_out_of_scope_sentinel ... ok
test callbacks::misc::tests::go_schedule_opcache_reset_returns_promptly ... ok
test callbacks::misc::tests::putenv_notice_logs_exactly_once_across_repeated_calls ... ok
test callbacks::output::tests::clamp_status_100_is_in_range ... ok
test callbacks::misc::tests::mercure_publish_notice_logs_exactly_once_across_repeated_calls ... ok
test callbacks::misc::tests::opcache_reset_notice_logs_exactly_once_across_repeated_calls ... ok
test callbacks::output::tests::clamp_status_999_is_in_range ... ok
test callbacks::output::tests::go_sapi_flush_calls_the_sinks_flush_and_returns_false_on_success ... ok
test callbacks::output::tests::clamp_status_above_999_clamps_to_500 ... ok
test callbacks::output::tests::go_sapi_flush_returns_false_when_closed_but_already_done ... ok
test callbacks::output::tests::go_sapi_flush_does_not_warn_for_an_ordinary_io_error_and_still_returns_false ... ok
test callbacks::output::tests::go_sapi_flush_with_no_sink_returns_false ... ok
test callbacks::output::tests::go_sapi_flush_returns_true_only_when_closed_and_not_done ... ok
test callbacks::output::tests::go_ub_write_after_close_context_discards_the_write_and_reports_the_cached_snapshot ... ok
test callbacks::output::tests::go_ub_write_delivers_bytes_to_the_sink_and_returns_its_count ... ok
test callbacks::output::tests::go_ub_write_reports_a_short_write ... ok
test callbacks::output::tests::go_ub_write_reports_zero_and_does_not_abort_on_a_write_error ... ok
test callbacks::output::tests::go_ub_write_with_no_context_does_not_abort_and_reports_the_full_length ... ok
test callbacks::output::tests::go_write_headers_forwards_every_header_sets_status_and_clears_after_a_1xx ... ok
test callbacks::output::tests::go_write_headers_returns_false_when_the_request_is_already_done ... ok
test callbacks::output::tests::go_write_headers_with_no_context_returns_true ... ok
test callbacks::output::tests::go_ub_write_with_no_sink_logs_and_reports_the_full_length ... ok
test callbacks::output::tests::go_write_headers_with_no_sink_returns_true_and_does_not_walk_headers ... ok
test callbacks::output::tests::clamp_status_below_100_clamps_to_500 ... ok
test callbacks::output::tests::split_header_only_skips_spaces_not_other_whitespace ... ok
test callbacks::output::tests::split_header_value_containing_a_colon_splits_at_the_first_one_only ... ok
test callbacks::output::tests::split_header_empty_value ... ok
test callbacks::output::tests::split_header_with_leading_colon_is_none ... ok
test callbacks::output::tests::split_header_value_may_be_non_utf8 ... ok
test callbacks::output::tests::split_header_skips_leading_spaces_in_the_value ... ok
test callbacks::output::tests::go_sapi_flush_warns_when_the_sink_is_not_a_flusher_and_still_returns_false ... ok
test callbacks::output::tests::split_header_with_no_colon_is_none ... ok
test callbacks::servervars::tests::an_absent_content_length_header_is_a_null_pointer ... ok
test callbacks::servervars::tests::collection_rejects_a_null_out_pointer ... ok
test callbacks::servervars::tests::authorization_is_returned_when_present ... ok
test callbacks::servervars::tests::collection_reports_nothing_without_a_request ... ok
test callbacks::output::tests::go_sapi_flush_with_no_context_returns_false ... ok
test callbacks::servervars::tests::a_present_but_empty_header_registers_as_a_null_pointer ... ok
test callbacks::servervars::tests::authorization_is_null_for_absent_and_empty_headers ... ok
test callbacks::servervars::tests::header_pointers_survive_reallocation_and_install ... ok
test callbacks::servervars::tests::path_translated_is_null_when_path_info_is_empty ... ok
test callbacks::servervars::tests::collection_reports_nothing_without_a_context ... ok
test callbacks::servervars::tests::content_length_minus_one_passes_through_verbatim ... ok
test callbacks::worker::tests::finish_php_request_with_no_context_does_not_abort ... ok
test callbacks::servervars::tests::phase_one_computes_every_documented_server_var ... ok
test callbacks::servervars::tests::content_type_is_null_for_a_present_but_empty_header ... ok
test callbacks::servervars::tests::path_translated_is_sanitized_join_when_path_info_is_present ... ok
test cgi::tests::clean_path_matches_go_filepath_clean ... ok
test callbacks::worker::tests::go_ub_write_after_finish_php_request_discards_the_write_and_reports_the_cached_snapshot ... ok
test callbacks::servervars::tests::update_request_info_returns_null_without_a_request ... ok
test callbacks::servervars::tests::server_port_defaults_to_80_when_host_carries_no_port ... ok
test callbacks::worker::tests::finish_php_request_closes_the_context_fires_the_signal_once_and_leaves_the_slot_installed ... ok
test cgi::tests::header_value_join_is_byte_oriented ... ok
test cgi::tests::ensure_leading_slash_matches_upstream_table ... ok
test cgi::tests::header_canonicalization_and_common_lookup_match_go ... ok
test cgi::tests::header_mangler_agrees_with_all_101_common_entries ... ok
test callbacks::servervars::tests::the_slot_is_free_once_collection_returns ... ok
test cgi::tests::sanitized_path_join_matches_root_and_non_utf8_cases ... ok
test cgi::tests::sanitized_path_join_preserves_encoded_separators_and_trailing_slashes ... ok
test cgi::tests::sanitized_path_join_rejects_traversal_and_absolute_request_paths ... ok
test cgi::tests::split_cgi_path_distinguishes_absent_empty_and_configured_lists ... ok
test cgi::tests::split_remote_addr_matches_upstream_table ... ok
test context::tests::a_completion_signal_may_consume_what_it_captures ... ok
test cgi::tests::server_name_and_port_match_upstream_host_parsing ... ok
test context::tests::canonical_header_name_leaves_non_token_names_unchanged ... ok
test cgi::tests::split_path_validation_matches_upstream_and_enables_uppercase_config ... ok
test cgi::tests::split_pos_matches_upstream_table ... ok
test cgi::tests::split_pos_security_regression_unicode_bypass ... ok
test cgi::tests::split_pos_unicode_case_folding_length_expansion_regression ... ok
test context::tests::clear_headers_stops_a_1xx_leaking_its_headers_into_the_final_response ... ok
test context::tests::close_context_fires_the_completion_signal_exactly_once ... ok
test context::tests::content_length_is_the_parsed_length_not_the_header ... ok
test context::tests::a_poisoned_slot_is_recovered_rather_than_deadlocking_or_aborting ... ok
test context::tests::arena_pointers_are_writable_not_just_readable ... ok
test context::tests::context_slots_set_replaces_and_drops_previous ... ok
test context::tests::close_context_snapshots_client_has_closed_into_client_had_closed ... ok
test context::tests::drop_releases_arena_but_is_done_does_not ... ok
test context::tests::go_quote_escapes_the_scalars_go_1_26_calls_unassigned ... ok
test context::tests::go_quote_matches_gos_percent_q_over_byte_strings ... ok
test context::tests::headers_get_first_is_the_first_value_not_a_join ... ok
test context::tests::headers_clear_empties_a_map_with_multiple_names_and_values ... ok
test context::tests::context_slots_set_get_clear_round_trip ... ok
test context::tests::fake_response_sink_records_writes_header_adds_status_and_flush ... ok
test context::tests::flush_error_distinguishes_not_a_flusher_from_a_real_io_error ... ok
test context::tests::headers_lookup_is_canonicalisation_insensitive ... ok
test context::tests::headers_present_but_empty_is_some_empty ... ok
test context::tests::invalid_content_length_message_keeps_non_utf8_bytes ... ok
test context::tests::nested_access_to_a_different_slot_panics_instead_of_deadlocking ... ok
test context::tests::request_body_empty_reads_zero_without_a_source_to_block_on ... ok
test context::tests::request_body_fill_stops_when_the_source_is_exhausted_short_of_the_buffer ... ok
test context::tests::arena_pointers_stay_valid_across_spine_reallocation ... ok
test context::tests::request_body_fill_loops_a_one_byte_source_until_the_buffer_is_full ... ok
test context::tests::request_context_validate_delegates_to_the_request ... ok
test context::tests::request_body_fill_never_reports_more_bytes_than_the_buffer_holds ... ok
test context::tests::request_uri_is_empty_without_a_raw_target_or_a_request ... ok
test context::tests::request_context_with_no_sink_is_constructible_and_distinguishable ... ok
test context::tests::split_path_keeps_absent_and_explicitly_empty_distinguishable ... ok
test context::tests::validate_accepts_duplicate_valid_content_length_headers ... ok
test context::tests::validate_accepts_no_content_length_and_a_valid_one ... ok
test context::tests::validate_rejects_negative_content_length ... ok
test context::tests::validate_rejects_non_numeric_content_length ... ok
test state::tests::compare_and_swap_changes_only_a_matching_state ... ok
test context::tests::request_uri_is_the_raw_target_verbatim_never_reconstructed ... ok
test state::tests::every_state_has_the_upstream_discriminant_and_name ... ok
test context::tests::request_uri_takes_raw_target_even_when_it_is_not_the_wire_bytes ... ok
test context::tests::validate_rejects_nul_byte_in_path ... ok
test context::tests::nested_set_from_within_a_closure_panics_even_for_a_new_index ... ok
test context::tests::context_slots_concurrent_access_neither_deadlocks_nor_loses_a_slot ... ok
test state::tests::setting_the_same_value_still_notifies_subscribers ... ok
test state::tests::safe_state_change_rejects_done_and_accepts_ready ... ok
test state::tests::safe_state_change_wakes_for_reserved_then_rejects_it ... ok
test state::tests::overlapping_subscribers_are_compacted_like_upstream ... ok
test state::tests::safe_state_change_waits_until_the_state_is_stable ... ok
test state::tests::wait_for_returns_immediately_for_an_existing_target_state ... ok
test state::tests::one_transition_wakes_every_matching_waiter ... ok
test thread::tests::automatic_max_threads_divides_and_truncates ... ok
test state::tests::wait_for_is_woken_by_set_from_another_thread ... ok
test state::tests::waiting_timestamp_preserves_elapsed_and_future_signs ... ok
test state::tests::two_threads_yield_to_each_other_via_states ... ok
test thread::tests::drain_channel_receiver_survives_the_close_reopen_window ... ok
test thread::tests::automatic_max_threads_doubles_when_total_memory_is_unknown ... ok
test thread::tests::automatic_max_threads_doubles_when_there_is_no_real_memory_limit ... ok
test thread::tests::drain_channel_close_is_one_shot_per_generation ... ok
test thread::tests::explicit_max_threads_passes_through_untouched ... ok
test thread::tests::drain_receivers_are_isolated_by_generation ... ok
test thread::tests::explicit_max_threads_below_num_threads_is_still_raised ... ok
test context::tests::opposite_order_cross_thread_nesting_panics_on_both_threads_rather_than_deadlocking ... ok
test thread::tests::resolve_max_threads_auto_uses_the_automatic_result_when_it_is_higher ... ok
test thread::tests::resolve_max_threads_raises_a_low_automatic_result_to_num_threads ... ok
test thread::tests::total_system_memory_is_nonzero_on_linux ... ok
test thread_regular::tests::directed_try_send_only_succeeds_against_a_parked_receiver ... ok
test thread_regular::tests::shared_channel_fallback_delivers_and_tracks_the_queued_counter ... ok
test state::tests::concurrent_set_and_wait_for_stress_completes_without_deadlock ... ok
test thread::tests::one_shot_real_boot_and_drain_completes_without_aborting ... ok
test state::tests::timed_out_waiter_removes_its_subscriber ... ok
test state::tests::waiter_for_two_states_ignores_other_transitions ... ok
test thread::tests::handler_value_lock_excludes_replacement_during_a_read ... ok
test thread::tests::inactive_slot_claim_is_held_through_handler_assignment ... ok
test thread::tests::per_slot_boot_transition_shutdown_repeats_three_times ... ok
test context::tests::go_is_print_matches_go_1_26_over_every_scalar ... ok
test context::tests::go_quote_matches_go_1_26_over_a_generated_corpus ... ok

test result: ok. 169 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 1.75s

     Running tests/abort_stub.rs (target/debug/deps/abort_stub-477e246262e3e6ca)

running 2 tests
frankenrust[Level(4)]: mercure_publish() was called, but frankenrust is not built with Mercure support
test go_mercure_publish_returns_the_not_built_sentinel_not_a_fabricated_success ... ok
test abort_stub_aborts_the_process_naming_the_symbol ... ok

test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.01s

     Running unittests src/lib.rs (target/debug/deps/frankenrust_server-c0d7b5be71533094)

running 14 tests
test request::tests::percent_decode_leaves_a_malformed_escape_untouched ... ok
test request::tests::percent_decode_leaves_plus_and_decodes_escapes ... ok
test request::tests::build_request_reads_method_path_and_headers ... ok
test response::tests::build_response_carries_status_headers_and_body ... ok
test response::tests::reject_response_carries_the_message_as_the_body ... ok
test response::tests::build_response_falls_back_to_200_for_an_unset_status ... ok
test server::tests::a_genuinely_broken_listener_is_not_retryable ... ok
test server::tests::a_rewritten_path_splits_with_an_empty_path_info ... ok
test server::tests::fd_exhaustion_and_aborted_handshakes_are_retryable_accept_errors ... ok
test server::tests::resolve_directory_index_appends_index_php_to_a_trailing_slash ... ok
test server::tests::resolve_directory_index_leaves_a_non_directory_path_alone ... ok
test server::tests::resolve_directory_index_leaves_a_php_bearing_trailing_slash_path_alone ... ok
test sink::tests::clear_headers_empties_the_map_but_not_the_body ... ok
test sink::tests::writes_headers_and_status_accumulate_and_are_readable_back ... ok

test result: ok. 14 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

     Running unittests src/main.rs (target/debug/deps/frankenrust_server-b78993721e6a852b)

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

     Running tests/concurrency.rs (target/debug/deps/concurrency-51bf13fd7085a631)

running 1 test
test two_hundred_concurrent_requests_across_four_php_threads_all_succeed ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 1.05s

     Running tests/directory_index.rs (target/debug/deps/directory_index-ff96f34f3e3993e7)

running 1 test
test get_root_serves_index_php_instead_of_the_document_root_directory ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.01s

     Running tests/directory_index_cgi_vars.rs (target/debug/deps/directory_index_cgi_vars-d007030177f5a039)

running 1 test
test get_root_and_get_sub_resolve_cgi_path_vars_as_if_index_php_were_requested ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.01s

     Running tests/hello_world.rs (target/debug/deps/hello_world-47d36b225df49e88)

running 1 test
test one_request_end_to_end_serves_hello_world ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.01s

     Running tests/keepalive_shutdown.rs (target/debug/deps/keepalive_shutdown-20ac843812266017)

running 1 test
test shutdown_completes_while_a_client_holds_an_idle_keep_alive_connection_open ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.01s

     Running unittests src/lib.rs (target/debug/deps/frankenrust_sys-9849a88ca91ef0fd)

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

     Running tests/version.rs (target/debug/deps/version-5074a8df99e82ebd)

running 2 tests
test frankenphp_get_version_reports_a_real_php ... ok
test all_26_go_symbols_are_defined_in_the_test_binary ... ok

test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.03s

     Running tests/zval_tags.rs (target/debug/deps/zval_tags-f82d6438eccab667)

running 3 tests
test get_ht_packed_data_reads_the_packed_flag ... ok
test type_tags_are_pairwise_distinct_and_undef_is_the_zero_sentinel ... ok
test zval_long_round_trips_through_the_bound_constant ... ok

test result: ok. 3 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

   Doc-tests frankenrust_core

running 3 tests
test crates/frankenrust-core/src/callbacks/thread.rs - callbacks::thread::go_frankenphp_store_force_kill_slot (line 23) - compile fail ... ok
test crates/frankenrust-core/src/thread.rs - thread::PhpMainThread::state (line 616) - compile fail ... ok
test crates/frankenrust-core/src/thread.rs - thread::PhpThread::state (line 346) - compile fail ... ok

test result: ok. 3 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.03s

   Doc-tests frankenrust_server

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

   Doc-tests frankenrust_sys

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

    PASS test
--- gate: miri
Preparing a sysroot for Miri (target: aarch64-unknown-linux-gnu)... fatal error: failed to build sysroot: sysroot build failed; stderr:
   Compiling compiler_builtins v0.1.160 (/usr/local/rustup/toolchains/nightly-2026-08-14-aarch64-unknown-linux-gnu/lib/rustlib/src/rust/library/compiler-builtins/compiler-builtins)
   Compiling core v0.0.0 (/usr/local/rustup/toolchains/nightly-2026-08-14-aarch64-unknown-linux-gnu/lib/rustlib/src/rust/library/core)
   Compiling libc v0.2.189
   Compiling object v0.39.1
   Compiling std v0.0.0 (/usr/local/rustup/toolchains/nightly-2026-08-14-aarch64-unknown-linux-gnu/lib/rustlib/src/rust/library/std)
   Compiling test v0.0.0 (/usr/local/rustup/toolchains/nightly-2026-08-14-aarch64-unknown-linux-gnu/lib/rustlib/src/rust/library/test)
   Compiling rustc-std-workspace-core v1.99.0 (/usr/local/rustup/toolchains/nightly-2026-08-14-aarch64-unknown-linux-gnu/lib/rustlib/src/rust/library/rustc-std-workspace-core)
   Compiling alloc v0.0.0 (/usr/local/rustup/toolchains/nightly-2026-08-14-aarch64-unknown-linux-gnu/lib/rustlib/src/rust/library/alloc)
   Compiling memchr v2.8.3
   Compiling adler2 v2.0.1
   Compiling rustc-demangle v0.1.28
   Compiling panic_abort v0.0.0 (/usr/local/rustup/toolchains/nightly-2026-08-14-aarch64-unknown-linux-gnu/lib/rustlib/src/rust/library/panic_abort)
   Compiling cfg-if v1.0.4
   Compiling rustc-literal-escaper v0.0.8
   Compiling unwind v0.0.0 (/usr/local/rustup/toolchains/nightly-2026-08-14-aarch64-unknown-linux-gnu/lib/rustlib/src/rust/library/unwind)
   Compiling rustc-std-workspace-alloc v1.99.0 (/usr/local/rustup/toolchains/nightly-2026-08-14-aarch64-unknown-linux-gnu/lib/rustlib/src/rust/library/rustc-std-workspace-alloc)
   Compiling panic_unwind v0.0.0 (/usr/local/rustup/toolchains/nightly-2026-08-14-aarch64-unknown-linux-gnu/lib/rustlib/src/rust/library/panic_unwind)
   Compiling gimli v0.34.0
   Compiling hashbrown v0.17.1
   Compiling miniz_oxide v0.9.1
   Compiling std_detect v0.1.5 (/usr/local/rustup/toolchains/nightly-2026-08-14-aarch64-unknown-linux-gnu/lib/rustlib/src/rust/library/std_detect)
   Compiling addr2line v0.27.1
error: failed to write /tmp/.tmpa1AF1d/target/aarch64-unknown-linux-gnu/custom_sysroot/build/object/a95da32f664de8f1/out/rustcH2uF9B/lib.rmeta: No space left on device (os error 28)

error: could not compile `object` (lib) due to 1 previous error

    FAIL miri
--- gate: conformance
    selftest: goldens self-consistent: ok
    selftest: normalisation length-invariant: ok
    selftest: corpus/golden bijection: ok
    selftest: concurrent isolation: ok
    selftest: frankenrust mismatch fails the run: ok
    selftest: skip_targets requires skip_reason: ok
    selftest: skipped case not counted as compared: ok
    selftest: frankenrust leg enabled with missing image fails: ok
    selftest: frankenrust leg disabled prints reason: ok
    selftest: 9 check(s) passed
--- replaying against upstream (dunglas/frankenphp@sha256:4b0713ddad6ca7eb21eb82ac6bdb7cb41de5192a930b615d89af6e15d74e82f8) on host port 51868
--- frankenrust leg disabled by default -- blocked on #159 (header casing and wire order) and #170 (chunked-encoding framing); see #199. Set FRANKENRUST_CONFORMANCE=1 to force it on locally.

conformance: PASS -- 26 case(s) compared, 0 skipped, 0 mismatches
    PASS conformance
--- gate: bench-report-selftest
ok - 3 servers: build_report does not crash
ok - 3 servers: frankenrust vs frankenphp line present with the right delta
ok - 3 servers: pasir vs frankenphp line present with the right delta
ok - 3 servers: no cross line between the two non-baseline servers
ok - 3 servers: exactly one verdict line per non-baseline server
ok - 2 servers: verdict line is actually produced (test isn't vacuous)
ok - 2 servers: baseline-relative output is byte-identical to the old algorithm
ok - 2 servers: full report does not crash
ok - 1 server: build_report does not crash
ok - 1 server: no verdict line

SELFTEST PASS
    PASS bench-report-selftest

GATE FAIL (default): miri

[melihucar]

Unblocker: requeue, back in the queue.

Diagnosis complete. Evidence assembled, body rewritten, spec verified against current main.

Diagnosis: the environment is wrong

Every failure was environmental. The issue was never actually attempted.

1. All three implementer attempts hit claude's quota wall before doing anything. claude.impl.1/2/3.log are each four lines: a 429 rate-limit rejection, "You've hit your weekly limit · resets Aug 20 at 4pm (Europe/Istanbul)", zero tokens consumed. There is no diff.*.patch for any attempt — no implementation was ever produced. The issue's Agent: claude line routed the implementer to a rate-limited agent.

2. The gate then failed on miri — "No space left on device (os error 28)" during the miri sysroot build in the container's /tmp (gate.1/2/3.log). That is disk exhaustion, not a code defect, and it ran against an empty worktree anyway. Transient: gates for #81, #169, #199, #15 all PASS miri since; the host now has 71Gi free and the dev image is present.

3. The previous unblock pass never touched the issue. claude.unblock.0.log (499KB) is one "Failed to authenticate: OAuth session expired" after another for the whole session. It made no decision.

What changed, so a fresh attempt goes differently

• orchestrator/config.py now routes every role to opencode (ROLE_AGENT all-opencode, landed Aug 17 in 8c982d7/987fb1d). I rewrote core: the prepared environment -- RequestContext.env, registerPreparedEnv, and the $_SERVER merge #207's body from Agent: claude to Agent: opencode — without that, the loop's _work would route the implementer straight back to claude (still rate-limited until Aug 20) and burn three more attempts identically. Gate: default is correct and unchanged; Depends on: was already absent (correct).
• The miri disk-full is gone (transient container /tmp fill); the gate environment is green again.

Why not SPLIT

There is nothing to split. The failure mode here is not "too big with failures clustered in one deliverable" — the implementer never wrote a single line, and the issue is already a single coherent deliverable deliberately carved out of #14. Its parts (the env field, the C add/merge, the $_SERVER merge) are mutually dependent for the end-to-end acceptance test; carved further, the pieces could not each pass the gate. Splitting would add queue noise without addressing the cause.

Why not CLOSE

The work is unlanded and needed: the two stale fc.env (PreparedEnv) is not part of this issue's RequestContext comments still stand in servervars.rs:586 and :639-642, the build.rs allowlist still lacks frankenphp_add_to_prepared_env/frankenphp_merge_with_prepared_env, RequestContext::new still has no env field, and shim.c's go_register_server_variables still omits the merge. Upstream has both functions at the cited lines (frankenphp.c:1270, :1461). #208 (fr:p1, fr:waiting) is gated on it.

Body rewritten (gh issue edit), Agent: claude → Agent: opencode. The loop's recovery flow will stamp Recoveries: 1 and return it to fr:ready on my signal.

RECOVERY: REQUEUE