iroh-tests

Reproductions for issues found while building deCDN on iroh, iroh-blobs, noq and bao-tree.

Start with FEEDBACK.md: 28 items, one per section, each with its test, the upstream source lines and the workaround in deCDN.

Every test pins exact releases: iroh 1.3.0, noq 1.3.0, iroh-blobs 0.103.0, bao-tree 0.16.1. Each file's module doc links the upstream source lines at the release commit and records the result we observed.

Two kinds of test

• tests/bug_*.rs assert the correct behavior. They fail at the pinned versions. A bug_ test that starts passing after a version bump means the issue is fixed upstream. Every operation that can hang runs under a timeout, so a hang shows up as an assertion failure, never as a stuck test run.
• tests/behavior_*.rs assert the current behavior behind a docs or API request. They pass at the pinned versions and fail if upstream changes that behavior.

Run

cargo test --no-fail-fast

cargo test --test bug_endpoint_close_after_runtime_drop -- --nocapture

The full run takes about 40 s, most of it spent in the timeouts of the hanging cases. All networking is local: endpoints bind 127.0.0.1 with relays disabled.

Bugs (fail today)

Test	Crate	What goes wrong
bug_endpoint_close_after_runtime_drop	iroh / noq	Endpoint::close never returns once the runtime that drove a connection is dropped (20 of 20 rounds hang).
bug_second_fsstore_open_deadlocks	iroh-blobs	FsStore::load hangs on a setup error (a second open of the same directory, or a root that is a regular file) instead of returning it.
bug_add_stream_swallows_error	iroh-blobs	add_stream with a source stream that yields Err hangs and never reports the error.
bug_export_bao_stream_ends_without_terminal_item	iroh-blobs	A MemStore shutdown mid-export ends the export_bao stream with no Done or Error; data_to_vec returns Ok with 0 of 8 MiB.
bug_hash_from_str_panics	iroh-blobs	Hash::from_str panics on wrong-length input instead of returning Err.
bug_bao_tree_async_encode_empty_ranges	bao-tree	Empty ChunkRanges panic the fsm encoder and both decoders in debug builds; in release the sync and fsm encoders disagree and the sync output does not decode.
bug_bitfield_total_bytes_underflow	iroh-blobs	Bitfield::total_bytes overflows on a range past size (panics in debug, returns a huge value in release).

Current behavior behind a request (pass today)

Test	Crate	Behavior
behavior_iroh_server_accepts_0rtt_by_default	iroh	A server that never opts into 0-RTT accepts and reads replayable early data (20 of 20), and cannot tell it was early data.
behavior_router_shutdown_unbounded	iroh	Router::shutdown waits forever on a handler whose shutdown never resolves, and the endpoint never closes.
behavior_recv_stream_drop_sends_stop_sending	noq	Dropping an unfinished RecvStream stops the peer's send side with code 0.
behavior_ipv6_port_ephemeral_when_only_v4_set	iroh	With only an IPv4 port configured, the IPv6 socket binds a random port.
behavior_status_partial_size_unknown	iroh-blobs	status() reports Partial { size: None } while observe() already knows the size.
behavior_await_completion_hangs_on_partial	iroh-blobs	await_completion() stays pending on a partial blob until the rest arrives.
behavior_export_bao_absent_range_error_kind	iroh-blobs	export_bao on a range the store lacks fails with LeafHashMismatch or UnexpectedEof, which read as corruption, not "not present".
behavior_mem_observe_unknown_hash_creates_entry	iroh-blobs	MemStore::observe on an unknown hash creates an entry: status turns from NotFound to Partial. FsStore does not.
behavior_add_bytes_creates_named_tag	iroh-blobs	Awaiting add_bytes creates a persistent auto-<timestamp> tag; add_stream does not.

License

Licensed under either of Apache License, Version 2.0 or MIT license at your option.