Repository navigation
(See description!) Pull request for repo [flow] branch [ipc-143_perf-dive-major] to base-branch [main]. - #117
Merged
Merged
Conversation
…ixes (mirror of internal feature branch).
…comparisons in new unit tests; Doxygen cross-namespace link.
…iberate-leak death test); add Conan dependency caching to pipeline.
…ify a filter string when triggering the pipeline, so that only some jobs run.
… test binaries and to the dependent Flow-IPC pipeline.
…/build trees before saving. Documentation: _LOCKED log macros: note the C++17 structured-bindings restriction.
…p Conan cache sizes down by pruning dep source/build trees before saving.
…: remove vendored-Boost-recipe kludge, as Conan-2 conan-center has boost/1.87.0 (the Conan-1 remote was frozen before 1.87 landed, making the kludge mandatory until now). Details: - Conan (+ PyYAML) now installed, at requirements.in-specified versions, into a Python venv via pip-tools (new requirements.in); each Conan-wrangling step sources the venv. - Custom sanitizer settings: mutate-settings.yml approach replaced by the Conan-2 settings_user.yml overlay; matrix defs now supply the full overlay dict. - Conan profile remains a workspace file, now passed explicitly via -pr:a -- to conan install and (newly needed) conan build, which in Conan 2 recomputes the dep-graph; profile thus still participates in the dep-cache key via hashFiles(). - Dep-cache steps: path ~/.conan/data -> ~/.conan2/p; key prefixes conan-data- -> conan2-data- so stale Conan-1 caches cannot restore (per the TODO planted there). - Post-install source/build-tree prune: conan remove --builds --src --force -> conan cache clean '*' --source --build. - Option scoping: flow: -> flow/*:. Dropped Conan-1-only tools.env.virtualenv:auto_use conf. - Install step: make install DESTDIR -> cmake --install --prefix (same final tree); install-root-dir env var retired. - conanfile.py: boost requires-string loses the kludge "@" and its comment block; also removed an accidental duplicate layout() def. - conan/: vendored boost recipe deleted; README.txt kept (documents the general borrowed-recipe convention; the dir reaches its preferred-empty state).
…unterpart: shared explanations now live in that file, with this one referring to them (previously the split was ad-hoc); consistent reference wording; a few comments that were missing on one side filled in. Also ported from that counterpart: flag-snapshot guard making the dep-cache restore-keys fallback safe; job-level timeout-minutes bound; OS-release info print in the apt step.
…ifiers in doc prose).
…...} form; straggling 0-for-null instances to nullptr.
…::Node::set_options() and ctor, net_flow::Peer_socket::set_options(), if used in non-throwing form, did not set *err_code to falsy on success. / Minor bug fix: NetFlow options sanity-validator, due to a typo, skipped several intended checks of custom settings as optionally set by a user. Splash radius of both: tiny, as this is a demo/test module. / Comment and/or doc changes.
…rt-worthy header is not actually exported by the build; doc-only headers (e.g., concept docs) belong in the new optional DOC_ONLY_HDRS list. / Added public utility APIs util::construct_at() (mimicking the C++20 STL facility) and util::default_init_at() (reminiscent of C++23 std::start_lifetime_as()). / Comment and/or doc changes.
…et with (), not {}, avoiding initializer-list-ctor pitfalls in forwarding code.
…instream jobs, which test against the min-supported versions of key dependencies (for Flow: Boost), a few jobs test against the newest conan-center-available versions of the same, so upstream changes are noticed before such a version becomes the min/default. Also: 2 comments updated for boost.asio 1.90+ (default associated executor).
…ow. The conan-center boost/1.91.0 recipe fails its own library-consistency check with our option set ("These libraries were built, but were not used in any boost module: boost_regex", though b2 reports regex as not building); bisecting.
… for now (conan-center boost/1.91.0 recipe consistency-check failure with our option set).
…ost.Regex no longer excluded in conanfile.py: the conan-center boost/1.91.0 recipe built boost_regex anyway (some module we build now needs it) and then failed its own library-consistency check ("built, but were not used in any boost module: boost_regex"); same class of problem, and same disposition, as the 1.87 `context` case already noted in conanfile.py.
…no longer excluded in conanfile.py (comment explains); comment tweaks.
…eadSanitizer build (for skipping tests TSAN cannot follow).
…rams; at the end of the job generate readable stack traces, available as artifacts near the existing logs. Plus comment and/or doc changes.
…peeding up various operations including making it possible to avoid all copying of `Key`s. Also general speed improvement by leveraging `unordered_flat_...` (open-addressed) lookups internally instead of the vanilla `unordered_` lookups. Added pointer-stability unit test for `Linked_hash_map/set`. `util::stat` now uses lazy evaluation (new `Stat_name`) to avoid computing field names and prefixes when the op has nothing to do with pretty-printing names.
…o not rely on transitive including). For Linked_hash_map/set added tests of move-only keys/values. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Kwu1pnHVN84wqQRauxuCfg
…II characters to ASCII).
…_flat_map perf sweep (log, struc::Channel, SHMJ Thread_cache); test-util pool-name sanitize; doc/to-do updates.
…ools, arena-create errors, Shm_session teardown race + liveness via new session::Session_cfg, fancy-pointer fast <=/>=/difference + signed-division fix); docs/tests.
…with STL-compliant containers to avoid zeroings of things like PoDs or `array<>`s when used within other containers. / Comment and/or doc changes.
…eased PR #114: sync_rcvd => syn_rcvd. CHANGELOG: Added v3.0.0.
Contributor
Author
|
This fixed #76 also |
Contributor
Author
|
This fixed #119 also |
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
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.
fixes #115, fixes #116
related to #82
flow::util::statsub-namespace: facilities for working with simple stat-holdingstructs ("stat-sets"). Each data member ("stat-member") is classified as an accumulator (counter), gauge, or high-water mark of a particular gauge.Histogram_counter: usable as accumulator stat-member and standalone.flow::util:Thread_local_ptr<T>(fastboost::thread_specific_ptr<T>equivalent).Thread_token,this_thread_unique_token()(actually-unique thread IDs).Thread_local_obj_deinit_safe<T, Tag>.flow::util::Action_registry,Locked_proxy<Target, Mutex>,Default_init_allocator,construct_at(),default_init_at(), and pointer-math/alignment free functions.thread_localobject -- could hit undefined behavior, because certain thread-local flow.log facilities may by then be deinitialized. flow.log: Logging near thread exit (e.g., from a thread_local object's destructor) may hit undefined behavior. #116flow::log::Loggers.flow::log::Log_context_mt,flow::util::Thread_local_state_registry,Polled_shared_state: please disregard their API descriptions in (See description!) Pull request for repo [flow] branch [ipc-143_perf-dive-nov] to base-branch [main]. #111 (not released); their final forms are described in API notes below.)flow::util::Linked_hash_mapandLinked_hash_set: perf improvements. (Adds to (See description!) Pull request for repo [flow] branch [ipc-143_perf-dive-nov] to base-branch [main]. #111, which improved them quite a bit: no key copying/key need not be copyable.)flow::net_flow): minor bug fixes (~no splash radius).API notes
flow::util:statsub-namespace:structof stats with adeclare_stats()member):stats_to_ostream(),stats_reset(),stats_assign(),stats_aggregate(),stats_aggregate_one(),stats_aggregate_shards(),stats_reset_shard_aggregate(),stats_sum(),stats_field_names(),stats_mark_reset_state(),stats_since_reset_state().atomicmembers:load(),store(),fetch_add(),fetch_sub(),update_hi_wmark().Histogram_counter;Stat_set_list;Global_stats;Stat_name.Thread_local_ptr<T>: semantically identical toboost::thread_specific_ptr<T>but with significantly faster.get().Thread_local_obj_deinit_safe<T, Tag>: equivalent to a nakedthread_local T, but for non-trivially-destructibleTguards access to the object, yielding null if the object has been destroyed near thread exit.this_thread_unique_token()(and typeThread_token): similar tothis_thread::get_id()but guaranteed never to recycle values.Thread_local_state_registry<T>(supersedes its description in (See description!) Pull request for repo [flow] branch [ipc-143_perf-dive-nov] to base-branch [main]. #111): maintains a thread-localTper thread, created on-demand viathis_thread_state()and deleted when its thread exits or the registry is destroyed, whichever happens first. To access all extant threads'Ts, callwhile_locked(task);taskreceives the registry's map of extantTs, valid only duringtask().Poll_flag: a single boolean flag, armed viaarm_next_poll()(typically from another thread) and both checked and disarmed viapoll_armed(), cheaply enough for the hottest of paths.Polled_shared_state<Shared_state>(supersedes its description in (See description!) Pull request for repo [flow] branch [ipc-143_perf-dive-nov] to base-branch [main]. #111): optional companion toThread_local_state_registry, for the pattern wherein one embeds aPoll_flagin each thread'sTand arms them all to signal a rare cross-thread request, which each thread notices cheaply on its fast-path.Shared_statecarries any state about that request, accessed viawhile_locked().Action_registry: a simple singleton template (with a tag parameter separating registries); code sites (particularly singleton initializers) can register actions (functors); other code can execute all actions in registration order, passing args to each.Locked_proxy<Target, Mutex>: proxy-pointer class template for the execute-around-pointer idiom.Default_init_allocator: allocator for STL-compliant containers that avoids zeroing things like PoDs orarray<>s.construct_at(), mimicking the C++20 STL facility of the same name;default_init_at(), reminiscent of C++23std::start_lifetime_as().is_power_of_two(),max_align_sz(),aligned_sz_of(),aligned_prefix_before(),after_aligned_prefix().String_view::starts_with()andends_with()are nowconstexpr.setup_auto_cleanup()takes its functor via move semantics.flow::cfg:value_set_member_id_to_opt_name(): new optional argsep_replacement.flow::log:Log_context_mt(supersedes its description in (See description!) Pull request for repo [flow] branch [ipc-143_perf-dive-nov] to base-branch [main]. #111): variant ofLog_contextwhoseLogger*can be changed thread-safely viaset_logger(); the mutex also protects copy/move andswap().FLOW_LOG_..._LOCKED()(lock, log, unlock); methodlog_while_locked()for a custom operation. New macroFLOW_LOG_SET_LOCKED_CONTEXT()is toFLOW_LOG_SET_CONTEXT()whatFLOW_LOG_..._LOCKED()is toFLOW_LOG_...().Log_context, it has noget_logger()orget_log_component()(an unlocked accessor would defeat the purpose).flow::util::Basic_blob<>::Allocator_rawalias renamed toAllocator.flow::log::Msg_metadata::m_call_thread_nicknameis now of typelog::Fixed_string(new alias forboost::static_string<128>) instead ofstd::string; thread nicknames are truncated to 128 characters. (Affects mainly customLoggerimplementations. Needed so that logging works even duringthread_localdeinit.)flow::util::bind_nsnamespace alias.#includeee is not exported. Doc-only headers (such as concept docs) must be listed in new listDOC_ONLY_HDRS.Impl notes
thread_specific_ptrreplaced by C++ built-inthread_local; all are exercised at every log call site that passes the verbosity filter.Logger*uses a faster data structure.Config::output_component_to_ostream()(used by ~allLoggers when writing the message) uses a faster data structure for component lookups.Thread_local_state_registryuses the new, fasterThread_local_ptrinternally.Linked_hash_mapandLinked_hash_setinternally use open-addressedboost::unordered_flat_setlookups instead ofunordered_set.Uniq_id_holder: slightly faster ID generation (relaxed memory-order); localstaticto avoid any init-misordering issue.util::statfield-name computation is lazy (no cost when not pretty-printing).net_flow::Event_set::events_wanted(),net_flow::Node::set_options()and ctor,net_flow::Peer_socket::set_options(): in non-throwing form, did not set*err_codeto falsy on success (left it alone).src/sanitize, so they apply to all test binaries and to the dependent Flow-IPC pipeline.To code reviewer
This has been reviewed elsewhere, including with: 🤖 Claude Code