test: sweep — consolidate duplication, cover real gaps#51
Merged
Conversation
The config and application suites re-instantiated one shared template's
behavior per protocol by hand-copying near-identical cases. Replace
those with value-parameterized suites over {wol, mdns, ssdp}, so every
protocol stays deliberately covered — robust to the template ever being
split apart — from one test source:
- config duplicate-rule matrix: one TEST_P suite × 3 protocols (the
port-dependent corners stay WoL-only); the per-protocol smoke tests
fold into it. ConfigCount fails loudly on an unhandled protocol.
- application invalid-socket tests: one TEST_P over {protocol × source
/target}; drop the SSDP rollback duplicate (the clear() is one
unconditional path already pinned by the mDNS case).
- udp_socket: the five moved-from/closed guards become one table test.
Dead surface removed: the unused LoopbackReceiver helper.
Two would-be cuts kept as sharper tests instead: the mixed-case MAC
parse now asserts every byte (pins per-nibble, not per-byte, case
folding), and the static-array logger name is retargeted to a
non-NUL-terminated array (the one StaticStringLength branch a literal
can't reach).
The macOS/FreeBSD resolver's pure helpers had no coverage at any layer — only the whole resolver, which needs a real interface. Split the testable logic into detail:: functions (mirroring SelectSourceAddresses) and unit-test them: - MacFromLinkSockaddr (was ReadMac): reads the MAC at the name-length- dependent sdl_data offset; tested for the offset skip, the no-address (alen 0) case, and refusing a MAC that runs past sa_len. - Ipv6SourceFlagsUsable: the SIOCGIFAFLAG_IN6 bitmask decision extracted from IsUsableIpv6 (the ioctl wrapper keeps calling it); tested for a usable address and each unusable flag bit. - CanonicalizeLinkLocalV6: the KAME embedded-scope-id clearing; tested that it zeroes only bytes 2-3 and leaves a canonical address alone. Behavior-preserving — the resolver calls the same code through the new seams. Tests are BSD-lane only (the helpers are non-Linux).
The sweep found live code with no test at any layer. Add narrowly scoped cases, each pinned to a specific production line: - dispatchers: the DrainReadableFd per-poll cap stops one bursty socket from starving the loop, then resumes on the next drain. - SSDP: the session cap is global across groups, a repeat interface change with no capability change is a no-op, a transient bring-up failure on a later group retries, and DIAL still forwards the response Location when the listener mint fails. - DIAL proxy: rediscovery and a spurious accept edge both refresh last_active past the original grace deadline; the closed guard stops minting for a later header once a prior one aborted; an interface change to a different live address drops and re-mints listeners. - raw socket: send fails cleanly when the frame can't be built; RejectsInvalidInterface now runs on both platforms; family-fd reuse on join (RequiresRoot). - config: env tags differing only by case both canonicalize to one name and are rejected. http: a Content-Length that overflows size_t is rejected. - failed-registration rollback: three paths that acquire an fd, fail to register it with a dispatcher, and must release it -- Application's signal-wakeup pipe, DefaultAddressMonitor::Start (Watch fails -> the socket is closed so IsValid() reports false), and EventLoopDispatcher ::Register (kernel-interest programming fails -> the map entry is erased so no fd is stranded; epoll-lane only). None was covered; a leaked fd is invisible to LeakSanitizer. Captured-output assertions match the log level or a short stable token, not full message text. TestCaptureSocket sizes its socketpair buffers so a full drain burst can be staged before the first read.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Test-suite sweep: cut duplication that doesn't earn its keep, add coverage where real gaps existed.
Commits
Consolidate duplicated matrices and drop dead test surface — value-parameterized suites replace hand-copied per-protocol cases (config dedup, application invalid-socket, udp moved-from), so every protocol stays deliberately covered from one source and stays robust if the shared template is ever split apart. Removes the unused
LoopbackReceiver. Two would-be cuts are kept as sharper tests instead (per-byte MAC case-folding; the non-NUL-terminated logger-name branch).Cover the BSD getifaddrs helpers via a
detail::split — the macOS/FreeBSD resolver's pure helpers (MAC read at the name-length offset, the IPv6 source-flag decision, KAME embedded-scope-id clearing) had no coverage at any layer. Split into testabledetail::functions and unit-tested; behavior-preserving.Cover error and edge paths surfaced by the sweep — narrowly scoped cases, each pinned to a specific production line: the drain per-poll cap; SSDP/DIAL interface-change and session-cap behavior; raw-socket build/interface failures; config and http rejections; and three failed-registration fd-rollback paths (the signal-wakeup pipe, the address monitor, and the event-loop dispatcher's kernel-interest programming). Captured-output assertions match the log level or a short token, not full message text.
Verification