RobotCode Files

Bug Fixes

• robot: Strip trailing = from variable names in resource builder (e11ca60)
• robot: Do not override the source on a model if there is already a source field (7e74c2d)

Newer Robot Framework versions already set the source field on File AST blocks, so overriding it here is unnecessary.

• robot: Resolve relative library paths in namespace cache validation (45d8f9f)

Namespace cache validation failed for files importing libraries via relative paths (e.g. ../../resources/libraries/common.py), causing unnecessary full rebuilds on every warm start.

Two issues fixed:

• validate_namespace_meta now passes the source file's directory as base_dir when falling back to get_library_meta/get_variables_meta, so relative paths resolve correctly

• get_library_meta and get_variables_meta initialize import_name before the try block to avoid UnboundLocalError when find_library or find_variables fails

• robot: Correct check for init.py file in _is_valid_file function (ffad219)

• robot: Add lock for _workspace_languages to prevent race condition (9bf0ee6)

WeakKeyDictionary is not thread-safe for concurrent read/write. Protect all access to _workspace_languages with a dedicated RLock.

Documentation

• news: Add supplementary release note for v2.5.0 (a8c39a3)
• Update release notes for v2.5.0 with (9978a88)
• Add news section to documentation site (7f14925)

Add a dedicated news section for release announcements with auto-generated sidebar (sorted newest-first), content loader, and redirect from /news/ to the latest article.

Features

• analyze: Add cache management CLI commands (1421107)

Add robotcode analyze cache subcommand group with five commands:

• path: print resolved cache directory
• info: show cache statistics with per-section breakdown
• list: list cached entries with glob pattern filtering (-p)
• clear: remove cached entries by section
• prune: delete entire .robotcode_cache directory

Output adapts to the global --format flag (text/json/toml) and --no-color mode. Text output uses rich-rendered markdown tables when color is enabled, plain aligned tables otherwise.

• analyze: Enable namespace caching by default (d896213)

Namespace caching speeds up startup for large projects by skipping re-analysis of unchanged files. It is now mature enough to be enabled by default.

• analyze: Add --cache-namespaces/--no-cache-namespaces CLI option (b050c32)

Add option to enable/disable caching of fully analyzed namespace data to disk. Disabled by default until the cache is fully optimized and validated. When enabled, skips re-analysis of unchanged files on startup, which can significantly speed up large projects.

Available as CLI flag (--cache-namespaces/--no-cache-namespaces), robot.toml setting (tool.robotcode-analyze.cache.cache_namespaces), and VS Code setting (robotcode.analysis.cache.cacheNamespaces).

• analyze: Add collection of unused keywords and variables on the CLI (5734690)

Add support for collecting unused keyword and variable diagnostics when running robotcode analyze code from the command line.

The feature can be enabled in robot.toml:

toml [tool.robotcode-analyze.code] collect-unused = true

or controlled directly per invocation:

bash robotcode analyze code --collect-unused robotcode analyze code --no-collect-unused

• langserver: Add code completion for Literal type hint values (42ab3a0)

Keywords with Literal["fast", "slow", "auto"] type hints now show their allowed values in the completion list, making it easier to discover and select valid argument values without checking the keyword documentation.

Also supports Union types containing Literal, e.g. Union[Literal["x", "y"], int].

• language_server: Add json to watched file extensions (74e92b0)
• plugin: Add has_rich property to detect rich availability (4581cc6)

Fall back to plain-text output in cache CLI commands when the rich package is not installed, instead of dumping raw unformatted markdown.

• robot: Add ROBOTCODE_CACHE_DIR env var for cache path override (a778cd5)

Enable redirecting the analysis cache to a custom directory via the ROBOTCODE_CACHE_DIR environment variable. This allows all RobotCode tools — CLI commands, analyze runs, and the language server — to share the same cache regardless of where it is stored.

The VS Code extension automatically sets this variable in the integrated terminal, so CLI commands find the correct cache without any manual setup.

Cache clearing now operates on the database directly instead of removing the directory, preventing errors when multiple processes access the cache concurrently.

• robot: Add advisory file locking to SQLite cache (b9351fc)

Add shared/exclusive advisory file locking (fcntl.flock) to prevent prune from deleting a cache database while another process (language server, analyze run) is using it.

• robot: Add SQLite cache backend for library and namespace data (edaea47)

Replace file-based pickle caching with a single SQLite database per workspace, consolidating all cache entries into one file instead of many individual .pkl files scattered across directories.

• robot: Add namespace disk cache for cold-start acceleration (abf3387)

Speed up first-time file analysis by caching fully resolved namespace results to disk. On subsequent IDE starts, cached namespaces are reused instead of re-analyzing every file from scratch, significantly reducing the time until diagnostics, code completion, and navigation become available.

The cache is automatically invalidated when source files, library dependencies, environment variables, command-line variables, language configuration, or the RobotCode version change.

• robot: Add ProjectIndex for O(1) workspace-wide reference lookups (6ef90f1)

Add an incrementally maintained inverse reference index (ProjectIndex) scoped per WorkspaceFolder. On each file build, references are updated in-place instead of scanning all files. All six reference types are supported: keywords, variables, namespace entries, keyword tags, testcase tags, and metadata.

• robot: Track tag and metadata references in namespace analysis (6a372fe)

Add structured reference tracking for tags and metadata during analysis. Tags are split into keyword_tag_references and testcase_tag_references to reflect their different semantics in Robot Framework (keyword visibility vs test selection/reporting). Metadata references track settings-level metadata entries by key. All reference keys are normalized for consistent lookups. Remove unused TagDefinition class.

• settings: Remove deprecated robocop configuration options (fa2217e)
• vscode: Show "What's New" notification after extension update (7fb2071)

Display an info notification when the extension version changes, linking to the news page on robotcode.io via the built-in Simple Browser. Also adds a public "RobotCode: Show What's New" command to the palette.

Performance

• cache: Defer data blob loading in CacheEntry until first access (795c420)
• diagnostics: Remove redundant exists() check in get_resource_meta (5696868)
• diagnostics: Consolidate to single AST model and remove data_only parameter (e2e4cbd)

This eliminates the second cached AST model per document, saving ~500KB/doc in CLI and ~200KB/doc in the Language Server. Releasing the model reference after analysis frees an additional ~100-500KB/doc in CLI mode.

• imports: Cache get_module_spec results in ImportsManager (fa83b30)
• robot: Patch RF's variable_not_found to skip slow RecommendationFinder (78a26c3)

Robot Framework's variable_not_found() calls RecommendationFinder for fuzzy "Did you mean...?" suggestions on every unresolved variable. This is O(n*m) over all variable candidates and extremely slow for large projects (see [#587]: 30+ min for 375 files).

RobotCode never uses the recommendation text — all VariableError catch sites either ignore the error or use it for unrelated diagnostics.

The monkey-patch replaces variable_not_found in all 5 RF modules that import it (notfound, finders, evaluation, store, init) with a fast version that raises VariableError with just the base message.

• robot: Speed up namespace cache loading with source hints and top-level-only imports (1d4ba6b)

Add resolved_resource_sources to NamespaceData that maps import hint keys to resolved file paths. During from_data() re-resolution, these hints skip expensive find_resource() filesystem searches when the cached path still exists on disk.

Also limit cached imports to top-level only (recursive imports are re-discovered during resolution), replace Path-based normalization with string-only os.path calls in imports_manager, and add a hint_key property to Import for consistent key computation.

• robot: Fix ArgumentSpec.resolve() caching bug for RobotArgumentSpec (05142c8)

The RobotArgumentSpec was recreated on every call (100K times) because:

• hasattr() always returned False with @dataclass(slots=True)
• Assignment went to a local variable instead of self

RobotArgumentSpec.init calls: 100K → 1.3K (-98.7%) resolve cumtime: 3.93s → 3.29s (-0.64s) Warm no-NS total: 30.78s → 29.81s (-0.97s, -3.2%)

• robot: Replace pathlib with os.path in file resolution (f3118bd)

Replace pathlib.Path operations with os.path string functions in robot_path.py for significantly faster file resolution:

• Use os.path.join/abspath/isfile/isdir instead of Path objects
• Cache validated sys.path entries lazily (_get_valid_sys_path)
• Separate basedir check from sys.path iteration
• Introduce _PathLike type alias for cleaner signatures
• Fix is_fifo() bug: use os.path.isfile() for init.py check

Warm namespace-cache scenario: 19.30s -> 13.84s (-28.3%). Pathlib calls reduced by 78-100%, posix.stat calls by 35%.

• robot: Cache KeywordMatcher and add dict-index to KeywordStore (5d21baf)

• Cache KeywordMatcher on KeywordDoc via lazy-init _matcher slot, eliminating 7.5M redundant instantiations per analysis run

• Add dict-index (_index + _embedded) to KeywordStore for O(1) keyword lookup by normalized name; linear scan only for embedded keywords
• Add getstate/setstate to KeywordDoc and KeywordStore to exclude transient fields (_matcher, _index, _embedded, parent, _hash_value, _stable_id) from pickle serialization
• Restore parent references in LibraryDoc.setstate via _update_keywords after deserialization
• Remove unused nosave metadata from argument_definitions, parent, and keyword_doc fields
• Fix Application.keyboard_interrupt to use self.exit() for consistent shutdown behavior

Measured improvement (cProfile, 1065 Robot files):

• Warm no-NS: 37.46s -> 29.25s (-22%)
• Cold no-NS: 42.94s -> 33.39s (-22%)
• Keyword matching: ~9.0s -> ~0.5s (-94%)

• Function calls: 118M -> 81M (-31%)

• robot: Optimize dataclass identity for ProjectIndex caching (deedd7b)

Prepare VariableDefinition, KeywordDoc, and LibraryDoc for efficient use as dictionary keys in the upcoming ProjectIndex disk cache and cross-process worker communication.

Refactor

• cache: Optimize SqliteDataCache with lazy deserialization and per-section tables (eeec613)

Replace single cache_entries table with per-section tables (libdoc, variables, resource, namespace) for better query performance.

Introduce CacheEntry with lazy meta/data deserialization — data blobs are only unpickled on cache hit, avoiding expensive deserialization on meta mismatch.

Move version checking from per-entry meta_version field to DB-level app_version parameter. On version mismatch all tables are dropped and recreated automatically.

Simplify cache API from 3 methods (cache_data_exists, read_cache_data, save_cache_data) to 2 (read_entry, save_entry), reducing DB round-trips from 3 to 1 for reads and 2 to 1 for writes.

Remove filepath_base property (adler32 hash-based keys) from all metadata classes — cache keys now use source paths or library names directly.

• diagnostics: Extract _run_in_subprocess and _save_import_cache helpers (0c347fd)

• Unify duplicated ProcessPoolExecutor lifecycle from _get_library_libdoc and _get_variables_libdoc into _run_in_subprocess

• Unify duplicated cache save logic into _save_import_cache

• diagnostics: Deduplicate and optimize metadata classes in imports_manager (a0cb4c6)

• Extract _collect_library_mtimes() to unify duplicated mtime collection from get_library_meta and get_variables_meta, using os.walk instead of Path.rglob for better performance

• Extract _matches_any_pattern() to unify triplicated pattern matching
• Add slots=True to LibraryMetaData, RobotFileMeta, and NamespaceMetaData
• Move mtime collection into dataclass constructors to avoid post-init mutation

• Remove unused itertools import

• jsonrpc2: Harden resource lifecycle without del (5cb7bb7)

• remove del-based resource cleanup from server, debugger client, and protocol

• move shutdown to explicit close/close_async paths
• make JsonRPCServer.close wait for server shutdown when safe
• add best-effort weakref finalizers for forgotten cleanup without noisy warnings

• unify runtime and finalizer bookkeeping into shared state objects

• robot: Remove unnecessary type casting in CacheEntry (c4eb160)

• robot: Optimize RF_VERSION checks with module-level dispatch (eec1627)

• Remove redundant _ROBOT_VERSION alias in semantic_tokens.py

• Remove always-true RF >= 5.0 guards (minimum supported version)
• Use conditional class-level property/method definitions for is_private and is_reserved in library_doc.py
• Add _RF7_PLUS module-level bool for BDD keyword search hot path

• Move ExceptHeader/WhileHeader imports to top-level (always available)

• robot: Replace Namespace getter methods with read-only properties (127ad65)

• robot: Extract Namespace into DTO with event-driven invalidation (4f37419)

Split the monolithic Namespace class into focused modules to reduce complexity and improve maintainability. The Namespace is now a pure data container built by NamespaceBuilder, with dedicated modules for import resolution, AST analysis, variable scoping, and scope trees.

• Introduce RF_VERSION constant and remove RF < 5.0 dead code (02cf495)

• Add module-level RF_VERSION constant to robotcode.robot.utils, resolved once at import time instead of repeated lru_cache lookups

• Remove dead code paths for Robot Framework < 5.0 (no longer supported), including the internal tidy-based formatter and obsolete version branches
• Replace all ~134 get_robot_version() call sites across 27 files with the RF_VERSION constant

Testing

• namespace: Ensure mtime differs on coarse resolution filesystems (c2b6a79)
• Add unit tests for SqliteDataCache backend functionality (f315426)
• Improve resource file modification simulation for validation (a3107fd)
• Improve mtime simulation for file modification in namespace validation (9a58278)