Conversation
Introduces a CMake option `MECH_CONFIG_USE_FMT` (default OFF) to use the
{fmt} library instead of `std::format`. A shim header (`format_compat.hpp`)
aliases fmt or std into `mc_fmt` so call sites are unchanged. Packaging
conditionally installs fmt when built from source and emits
`find_dependency(fmt)` when found via the system. fmt headers are exposed
with PUBLIC linkage since they appear in the public `format_compat.hpp`
header. CI tests both `MECH_CONFIG_USE_FMT=OFF` and `ON` via a matrix
dimension across Mac, Ubuntu, and Windows workflows.
Adds a `MECH_CONFIG_COMPILE_WARNING_AS_ERROR` option (default OFF) that
passes `/WX` on MSVC and `-Werror` elsewhere. Enabled in all CI workflows
and Dockerfiles. Fixes the clang-cl warnings this exposed: removes unused
`status` variables in `v0/parser.cpp` and `v1/mechanism_parsers.cpp`,
drops unused `phase_optional` return values in the development model
validators, and removes redundant `std::move()` on temporaries returned
by `GetComments()` (pessimizing-move).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: GitHub Actions <actions@github.com>
Setting BUILD_SHARED_LIBS globally with FORCE causes all subsequent FetchContent dependencies (e.g. TUV-x) to also build as shared, which breaks their builds if their objects lack -fPIC. Instead, set POSITION_INDEPENDENT_CODE on the fmt target directly after fetch so fmt's static library is PIC-safe when linked into a shared mechanism_configuration, without side effects on other dependencies. Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
…rivate headers) and support in-code mechanisms (#276) * refactoring v1 * v0 refactor * development * Accept `name` or `species name` for reaction components (v1 alias) Add an `exactly_one_of` group to ValidateSchema so a schema can require exactly one of a set of keys, then use it to let reaction components reference their species with either the canonical `name` or the legacy `species name` alias (used by v1 config files): exactly one, error if both or neither. - ValidateSchema: defaulted `exactly_one_of` param. None present -> RequiredKeyNotFound; more than one -> MutuallyExclusiveOption. Group members count as allowed keys (never flagged invalid). - validation.hpp: add `species_name = "species name"`. - ValidateReactantsOrProducts requires exactly one of {name, species name}. - New GetReactionComponentName() helper; route the component-name reads across all reaction validators and the parser through it. - Tests + fixtures: AcceptsSpeciesNameAlias, RejectsBothNameAndSpeciesName. Prep for consolidating development -> v1 (issue #269): keeps community v1 files (which use `species name`) parseable by the development engine. Also adds planning docs: PLAN-consolidate-v1.md, TODO-name-species_name-alias.md. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Rename development engine to v1; delete the old combined v1 parser Consolidate onto a single v1 engine (issue #269). The two-phase Validate()/Parse() "development" engine becomes the canonical v1 implementation; the old combined v1 parser is removed. - Delete old src/v1 combined parser (parser.cpp, mechanism_parsers.cpp, utils.cpp, reactions/*_parser.cpp) and its headers (parser.hpp, mechanism_parsers.hpp, reaction_parsers.hpp, validation.hpp, utils.hpp). - git mv src/development/* -> src/v1/* and include/.../development/* -> include/.../v1/* (sources, headers, reactions/{parsers,validators} tree, CMakeLists), preserving history. - Sweep namespace development -> v1 and mechanism_configuration/development/ includes -> v1/. Old v1::validation (with its own species_name) is gone; the engine now resolves unqualified validation:: to the shared top-level namespace. - Drop add_subdirectory(development) from src/CMakeLists.txt; remove the emptied development dirs and a dead, unreferenced reaction_types.hpp stub. Library builds. Tests are not yet migrated (Stage 4): test/unit/v1 still calls the old API and test/unit/development references the gone namespace. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Wire top-level parse() dispatch (file + string) Replace the throwing stub with a real version dispatch returning std::expected<Mechanism, Errors>: - directory input or a missing version field -> v0 parser - major 0 -> v0, major 1 -> v1 engine (FileToYaml -> Validate -> Parse) - unsupported versions -> InvalidVersion Add parse(const std::string&) for in-memory v1+ documents. Remove the dead f() and commented-out dispatch; re-enable the v0 include; fix parse.hpp (drop the duplicated declaration, add the string overload). validate(const Mechanism&) (the in-code path, Option 2) and test migration are still pending. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Stage 4: unify v1 tests on the engine; add file-list config support Engine / parse(): - Engine now accepts major version 1 (MAJOR_VERSION 1, not 2), since the renamed engine IS the v1 parser. - Add v1::Parser::ResolveFileConfig(): ports the multi-file "{ files: [...] }" logic from PR #256 to the unified engine. species/phases/reactions sections may be split across files (v1.1+); they are merged into one inline node, then validated/parsed normally. Inline sections pass through unchanged. Structural checks (missing 'files' key, non-array/object section, minor-version >= 1) and file-load errors are reported with the config path prefix. - parse() v1 case now goes ResolveFileConfig -> Validate -> Parse. - Drop the parse(const std::string&) content overload: it hijacked parse(std::string path). parse(filesystem::path) accepts strings implicitly. Tests: - development fixtures renumbered 2.0.0 -> 1.0.0; development test sources swept development:: -> v1::, includes development/ -> v1/. - v1 community tests migrated from v1::Parser::Parse(path) to top-level parse(). Status: file-list mechanism verified (its structural sub-tests pass). Remaining 14 failures are a single separate gap: community v1 lists phase species as bare strings ([A, B, C]) but the engine expects objects ([{name: A}]). Tracked next. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Accept bare-string phase species in the v1 engine (restore v1 behavior) The old v1 phase parser accepted a phase's species as either bare strings (`species: [A, B]`) or objects (`species: [{name: A, ...}]`). The development engine only handled the object form and threw on scalars. Restore the dual handling in both ParsePhases (parse) and ValidatePhases (validate): a scalar is shorthand for the species name; a map carries name + optional properties. Valid v1 reaction/phase configs now parse. Remaining v1 test failures are a separate matter (stale error-count expectations + a few validator key-set gaps such as FIRST_ORDER_LOSS dropping `products`). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Restore v1 leniency in the engine: optional/scalar components, FOL products, surface condensed_phase Reconcile the engine's per-reaction validators/parsers with v1 community files: - Reaction components may be bare strings (e.g. `gas-phase species: A`), not just objects. GetReactionComponentName handles scalars; ParseReactionComponents only reads coefficient/comments on the object form (scalar -> name only). - first_order_loss: restore `products` (optional key + parse it, guarded since optional). The type already had the field; old v1 allowed it. - surface: `condensed_phase` is optional, not required (community surface files omit it). Validator only verifies the phase when present; parser reads it when present. No more segfaults. Remaining v1 failures are stale test expectations (the engine reports richer errors than old v1), to be updated next per decision. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Drop the 6 CAMP-only reactions from v1; delete development tests These reactions are supported by CAMP (v0) and aren't used by the ODE solver, so remove them from the v1 engine (re-addable later atop the canonical types): aqueous_equilibrium, condensed_phase_arrhenius, condensed_phase_photolysis, henrys_law, simpol_phase_transfer, wet_deposition. - Delete their parser + validator sources and CMake entries. - Remove their IReactionParser classes + registry entries from reaction_parsers.hpp. - Remove their type structs + Reactions members from types.hpp (species-level henrys_law_constant_* properties are unrelated and kept). - Delete all development tests (they had been swept onto the v1 engine during the rename and were the only coverage for these now-removed reactions). Unused validation:: keys for the dropped reactions are left in place (harmless). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Update v1 test expectations to the engine's error reporting; fix file-list version-check ordering - v1 community tests: update error-count expectations to the engine's richer reporting (unknown species -> ReactionRequiresUnknownSpecies + RequestedSpeciesNotRegisteredInPhase; duplicates report one error per occurrence). Drop branched's stale third bad-component assertion. - ResolveFileConfig: check the file-list minor-version requirement BEFORE loading referenced files and return early, instead of also emitting file-not-found errors for a config already known to be invalid (matches old v1). Suite now 25/26. The lone remaining failure (file_configs TwoPhasesSets) is a semantic-policy question: a gas-phase reaction produces a species defined only in the aqueous phase, which the engine's phase-membership check rejects. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Exempt reaction products from the phase-membership check A reaction's products may legitimately reference species defined in another phase (e.g. a gas-phase reaction producing an aqueous species), so only reactants are required to belong to the reaction's phase. Reactants stay phase-checked; products are still required to be known species (unknown-species check unchanged). - All reaction validators build a reactants-only list for CheckSpeciesPresenceInPhase; products go only into the unknown-species check. emission (products only) drops the phase check entirely; first_order_loss already checked reactants only. - Realign v1 test error-count expectations accordingly. Full suite green: 26/26. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Add version-agnostic validate(const Mechanism&) Semantic validation belongs to the canonical Mechanism, not to a specific parser version. Add a top-level validate(const Mechanism&) -> Errors that checks the mechanism's semantic invariants on the struct, independent of how it was produced (parsed from any version, or built in code): - unique species names; unique phase names; unique species within a phase - phase species exist in the species list - reaction species exist; reactants are registered in the reaction's phase (products may reference any phase) This is the in-code validation path from issue #269's "ideas" (in-code mechanism -> validate). New unit tests cover an in-code mechanism directly. Note: additive. parse() still uses v1's YAML-level (line-numbered) validation; routing parse() through validate(Mechanism) and removing v1's duplicated semantic checks is a follow-up. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Refactor validate() onto a single semantics checker over a located intermediate Introduce semantics::Input (species/phases/reactions as name + optional ErrorLocation) and ValidateSemantics(Input) as the single place the semantic rules live. validate(const Mechanism&) now builds a location-free Input and calls it. Errors carry line:col whenever a source location is supplied — which lets the parse path reuse the same checker with locations next, instead of duplicating the rules in v1. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Add v1 BuildSemanticInput (YAML -> located semantics::Input) Extracts species/phase/reaction species references from a fully-resolved v1 YAML node, each carrying its source ErrorLocation, into the version-neutral semantics::Input. This lets the parse path feed the single ValidateSemantics checker with line:col (the YAML counterpart to validate(Mechanism)'s location-free build). Not yet wired into Parser::Validate — that wiring + removing v1's own semantic checks is the next step. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Route parse() semantic validation through ValidateSemantics; strip v1's copy v1's parsers now own only deserialization + structural (schema/cardinality/ mutually-exclusive) validation. All semantic checks (unknown species, phase membership, duplicates, unknown phase) are performed once by the version-neutral ValidateSemantics: - Parser::Validate runs the structural validators, then calls ValidateSemantics(BuildSemanticInput(object)) so parse-time semantic errors keep line:col. - The 12 reaction validators drop their species/phase/duplicate checks (keeping ValidateSchema, ValidateReactantsOrProducts, cardinality, Ea-vs-C). - ValidateSpecies / ValidatePhases reduced to structural schema checks. Semantic rules now exist in exactly one place and apply to the canonical Mechanism (parsed or in-code). The old per-component semantic helpers in v1/utils are now unused (left in place; can be removed in a follow-up). Suite: 27/27. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Modernize integration tests to the parse()/validate() API Rewrite the integration tests onto the current public API: - test_parser: use top-level parse() for version dispatch (v0 fallback, v1, missing file, unsupported version, v0 directory). Drop the v2/development case. - test_v0_parser: v0::Parser::Parse still returns expected; update to the canonical Mechanism + .error(). - test_v1_parser: files via parse(); in-memory YAML/JSON via the v1 engine (Validate + Parse(node)), since v1::Parser no longer has Parse(path)/ ParseFromString/ParseFromNode. - test_v1_read_from_file_configs: multi-file configs via parse(). - Drop the stale development_parser CMake entry (source no longer exists). Known: examples/v1/full_configuration.{json,yaml} is semantically invalid under the engine's checks (undeclared foo/bar/baz; third-body M used outside the gas phase), so the two full-configuration tests fail pending an example-data decision. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Fix examples/v1/full_configuration to be semantically valid The full-configuration example tripped the engine's semantic checks: - third-body species M was used as a TROE reactant but absent from the gas phase -> add M to the gas phase species list - the TERNARY_CHEMICAL_ACTIVATION reaction referenced undeclared species foo/bar/baz -> use declared species (A/B/C) Species count is unchanged (M was already declared), so the integration test assertions are unaffected. Full suite green: 31/31. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Remove dead semantic helpers from v1/utils After semantic validation moved to the version-neutral ValidateSemantics, these v1/utils helpers had no remaining callers: FindUnknownObjectsByName, CheckSpeciesPresenceInPhase, CheckPhaseExists, ReportUnknownSpecies, FindDuplicateObjectsByName, GetSpeciesNames, and the NodeInfo/DuplicateEntryInfo structs. Remove them. v1/utils now holds only the parser-support helpers still in use: AsSequence, AppendFilePath, GetComments, GetReactionComponentName. Full suite green: 31/31. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Tidy v1 type_validators: structural-only docs, drop dead ValidateParticles - Rewrite the docstrings to reflect that these are structural (schema) checks only; semantic invariants moved to ValidateSemantics. - Remove ValidateParticles (header + definition): its only callers were the Henry's-law/SIMPOL validators, which were dropped with the CAMP-only reactions. Suite green: 31/31. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Rename validation.hpp -> validation_keys.hpp The header holds the schema key-name constants (the `validation::` namespace), not validation logic. Renaming disambiguates the three similarly-named headers: - validation_keys.hpp : the key strings - validate_schema.hpp : structural/schema checking - validate.hpp : semantic validation (ValidateSemantics / validate) The `validation` namespace name is unchanged; only the file + its include sites (7) are updated. v0/validation.hpp is unrelated and untouched. Suite green: 31/31. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs: add README usage example (parse a file or validate an in-code mechanism) A single, compile-checked example showing the two entry points around the canonical Mechanism: parse() a YAML/JSON file (version auto-dispatched, returning std::expected<Mechanism, Errors>), and build a Mechanism in code then validate() it with the same semantic checks the parser uses. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * private headers * Flatten v1 detail headers to nested namespace; fix ::v typo Collapse the v1 private headers to the joined `mechanism_configuration::v1` namespace form and normalize include ordering. Fixes a typo in type_validators.hpp (`::v`) that broke compilation. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs: document lambda_rate_constant, add it to examples; drop planning notes - Add docs/source/v1/reactions/lambda_rate_constant.rst and wire it into the v1 reactions index/toctree (it was the only v1 engine reaction undocumented). - Add a LAMBDA_RATE_CONSTANT reaction to examples/v1/full_configuration.{yaml,json} so the example exercises every v1 reaction type (both parse cleanly). - Remove the scratch planning docs PLAN-consolidate-v1.md and TODO-name-species_name-alias.md from the repo root. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Release 2.0.0: bump version and add docs changelog Bump project version 1.1.2 -> 2.0.0 to reflect the breaking C++ API changes on this branch (std::expected parse()/validate(), canonical Mechanism, the public/private header split with a single umbrella header, and removed v1 reactions). The v0/v1 config file formats are unchanged. Add docs/source/changelog.rst (modeled on music-box's changelog) covering the API changes, header reorganization, removed reactions, and a migration example, and wire it into the docs toctree. Update the docs |project_version|. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ci: fix Ubuntu workflow matrix exclude indentation `exclude` was nested under `strategy` instead of `strategy.matrix`, which is an invalid workflow definition — GitHub Actions rejected it at startup, so the Ubuntu job failed in 0s on every push. Re-indent `exclude` under `matrix`. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Fix Linux clang and MSVC CI builds Two separate stdlib/include issues that GCC papered over via transitive includes but Linux clang and MSVC did not: - MSVC: errors.hpp defines PrintTo(..., std::ostream*) and streams a string into it, but never included <ostream>, leaving std::basic_ostream an incomplete type (C2027 in <string_view>'s operator<<). Add <ostream>. - Linux clang: clang-18 paired with libstdc++ (gcc-13/14) does not expose std::expected at any -std flag (reproduced in ubuntu:24.04). Build the Ubuntu clang job against libc++ (-stdlib=libc++ + libc++-dev), which provides std::expected and std::format. Verified: full repo builds clean with clang-18 + libc++ for both fmt OFF and ON under -Werror, 31/31 tests. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Rename structural checks Validate -> CheckSchema; reserve validate() for semantics The word "validate" was overloaded: the public semantic validate(const Mechanism&) / ValidateSemantics shared the name with the v1 structural YAML checks, making it hard to see what the real validation is. Reserve "validate" for semantics and name the structural layer "CheckSchema": - Parser::Validate -> Parser::CheckSchema - IReactionParser::Validate -> CheckSchema (+ all 12 reaction overrides) - ValidateSpecies/Phases/Reactions/ReactantsOrProducts -> CheckSpeciesSchema/CheckPhasesSchema/ CheckReactionsSchema/CheckReactantsOrProductsSchema - ValidateSchema primitive -> CheckSchema - detail/validate_schema.{hpp,cpp} -> detail/check_schema.{hpp,cpp} (+ src/check_schema.cpp) In-member primitive calls are qualified (mechanism_configuration::CheckSchema) to avoid the member name hiding the free function. Fix the stale validator docstrings that claimed to "ensure all referenced species and phases exist" (that moved to ValidateSemantics) and mark the now-unused existing_species/existing_phases params. validate()/ValidateSemantics and the validators/ + type_validators.* names are unchanged. Build clean (gcc-14 -Werror), 31/31 tests pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Tidy v1 schema checkers: drop dead is_valid, rename validation_errors The reaction validators carried an is_valid flag that was either never read (arrhenius) or only gated a trailing `return errors;` that the function reached anyway. Where it did guard a later check (first_order_loss/surface/photolysis cardinality, and CheckReactionsSchema's per-reaction loop), it was equivalent to `!errors.empty()` at that point, so the flag is replaced by that check. Also rename the reused local `validation_errors` -> `schema_errors`, matching the CheckSchema naming and avoiding the "validation" overload (reserved for the semantic validate()/ValidateSemantics). No behavior change. Build clean, 31/31 tests pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Rename type_validators -> type_schema, reactions/validators/ -> reactions/schema/ These hold structural schema checks (CheckSpeciesSchema, XxxParser::CheckSchema, etc.), not semantic validation, so "validators" was misleading. Rename to match the CheckSchema naming and pair cleanly with the sibling reactions/parsers/. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Flatten reactions/{parsers,schema}/ into one file per reaction Each reaction was one IReactionParser subclass (e.g. ArrheniusParser) whose two methods, CheckSchema and Parse, were split across reactions/schema/X.cpp and reactions/parsers/X.cpp with near-identical include blocks. Splitting one class across two files/dirs cost cohesion for no real benefit. Merge each pair into a single src/v1/reactions/X.cpp holding the whole parser (CheckSchema then Parse), drop the parsers/ and schema/ subdirectories, and list the 12 sources directly in reactions/CMakeLists.txt. (taylor_series needed <detail/constants.hpp>, previously only pulled in on the parser side.) No behavior change. Build clean (gcc-14 -Werror), 31/31 tests pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * v1::Parser::Parse owns validation; return std::expected, hide CheckSchema Parse(node) now runs CheckSchema (structure + semantics) itself and returns std::expected<Mechanism, Errors>, so callers can't build from an unvalidated node or forget to validate. CheckSchema and the construction step (Build) are now private. The top-level parse() v1 branch collapses to ResolveFileConfig -> Parse, and the string-parsing test helper to a single Parse call. No behavior change. Build clean (gcc-14 -Werror), 31/31 tests pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * v1::Parser::Parse: add path/string/node overloads; mirror v0 in parse() Parsing a v1 file no longer requires calling ResolveFileConfig by hand. Parse is now overloaded: - Parse(filesystem::path) -> resolve file-list sections, validate, build - Parse(std::string) -> treat as a YAML/JSON document, validate, build - Parse(YAML::Node, bool) -> validate + build a loaded node ResolveFileConfig, CheckSchema, Build, and the config-path setters are now private; the dead FileToYaml is removed. The top-level parse() v1 branch collapses to `return v1::Parser{}.Parse(config_path);`, matching v0. Exception safety moved into Parse(node) (the shared path), so value-conversion throws surface as Errors rather than escaping. No behavior change. Build clean (gcc-14 -Werror), 31/31 tests pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs: use a static Apache-2.0 license badge; fix LICENSE link to main The github/license shields.io badge is dynamic (hits the GitHub API) and intermittently fails with "Unable to select next GitHub token from pool", rendering broken. Switch to a static Apache-2.0 badge that never calls the API, and point the link at blob/main/LICENSE (the default branch) instead of master. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Drop read_from_config_file flag from v1::Parser The boolean only told CheckSchema whether to reset config_path_ (the file-path prefix for error messages) — a flag argument that callers had to reason about. Each Parse overload already knows its context, so it now sets config_path_ itself (ResolveFileConfig for files; empty for string/node) and delegates to a private flag-free ValidateAndBuild(node) that runs CheckSchema + Build. CheckSchema and Parse(node) lose the bool. No behavior change. Build clean, 31/31 tests pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * test: fix stale version_mismatch comment The comment claimed version "1.3.0" / "minor != 1", but the config is "1.0.0" and the gate is `minor < 1` (file-list requires v1.1+). Correct the comment to match the config and logic. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Hoist semantic validation out of CheckSchema into ValidateAndBuild CheckSchema also ran ValidateSemantics, so its name undersold what it did and re-overloaded "validate" in the structural layer. Move the semantic pass up into ValidateAndBuild, which now reads as an explicit three-phase pipeline: structural (CheckSchema) -> semantic (ValidateSemantics, gated on clean structure) -> Build. CheckSchema is now schema-only, matching its name. No behavior change. Build clean, 31/31 tests pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Rename validation key-vocabulary to keys The validation:: namespaces were just dictionaries of YAML key-name strings, used by parsing and construction as much as by checking (e.g. Build reads object[validation::gas_phase]) — so "validation" misdescribed them. Rename mechanism_configuration::validation -> keys and v0::validation -> v0::keys (validation_keys.hpp -> keys.hpp, v0/validation.hpp -> v0/keys.hpp), and validation:: -> keys:: throughout (~600 refs). "validation" is now reserved for actual semantic validation. No behavior change. Build clean (gcc-14 -Werror), 31/31 tests pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Move the v1 key vocabulary under v1 (keys -> v1::keys) detail/keys.hpp held the v1 keys but lived at the detail root, asymmetric with v0's detail/v0/keys.hpp. Its only cross-version use was parse()'s version dispatch reading the "version" field. Move it to detail/v1/keys.hpp as mechanism_configuration::v1::keys (v1 call sites are inside namespace v1, so keys:: still resolves with no change), and have the version-agnostic dispatcher in parse() read the literal "version" instead of depending on a version's keys. Now each version owns its key vocabulary: v0::keys and v1::keys. No behavior change. Build clean (gcc-14 -Werror), 31/31 tests pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Prefix the unsupported-version error with the config file path The "Unsupported version number" error was the one parse() error without a file path. We only reach that branch after reading the version out of the file, so it always names a real document — prefix it (path: message) like the other errors, so it points at the offending file. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Include line:col in the unsupported-version error GetVersion now also returns the source location of the `version` field (absent for directories / version-less docs). parse()'s default branch uses it to format the error as path:line:col error: ... , matching the located v1 errors, instead of just the file path. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * adding readme example parsing * update to camelcase * remove old aqueous and henrys law configs * update the copyright header * rearrage the types * replace phase location with location --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: Jiwon Gim <jiwongim@ucar.edu>
Co-authored-by: GitHub Actions <actions@github.com>
* clean up the config file for cloud chemistry * distinguish between micm arrhenius and miam arrhneius * add miam struct and keys * code cleanup * Update test/integration/integration_configs/cam_clound_chemistry.json Co-authored-by: Kyle Shores <kyle.shores44@gmail.com> --------- Co-authored-by: Kyle Shores <kyle.shores44@gmail.com>
Co-authored-by: GitHub Actions <actions@github.com>
* Fix code coverage to exclude third-party and system code
Coverage was reporting ~30% because the lcov capture only excluded
test/*, so the totals included the FetchContent dependencies built
under build/_deps/ (yaml-cpp, googletest, optionally fmt) plus inlined
system/libstdc++ headers. The denominator was dominated by untested
third-party code, masking the project's real coverage.
- Add _deps/* and /usr/* to the lcov EXCLUDE list in CMakeLists.txt
(BASE_DIRECTORY only sets path display, it does not restrict capture).
- Add codecov.yml with an ignore list as a backstop for Codecov's own
path matching.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Instrument the library for coverage (apply flags before add_subdirectory)
The real cause of the low (~30%) coverage: append_coverage_compiler_flags()
was called after add_subdirectory(src), so CMAKE_CXX_FLAGS only picked up the
coverage flags for the test targets defined later. The src/ library was built
without instrumentation, so it produced no .gcno/.gcda data and the report
covered only a handful of header lines inlined into the test TUs (51 lines).
Move the include + append_coverage_compiler_flags() ahead of
add_subdirectory(src) so the library itself is instrumented. The
setup_target_for_coverage_lcov() target definition stays in the test block.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Drop Fortran-only flags from coverage compiler flags
COVERAGE_COMPILER_FLAGS included -fcheck=bounds,do,pointer and
-ffpe-trap=zero,overflow,invalid, which are gfortran-only. This is a
CXX-only project, so cc1plus rejects them ("valid for Fortran but not
for C++"). It went unnoticed because the flags were previously only
applied to the gtest test targets (no -Werror); now that the library
is instrumented (and built with -Werror), the warnings became errors.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* add parser accepting string config * adding yaml test * changing default gas phase name to gas * using correct reaction array for each type * correcting tests * make fmt private
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
2 participants
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.
No description provided.