hydrust 0.5.0__tar.gz

hydrust-0.5.0/.gitignore

@@ -0,0 +1,26 @@
+# Generated by Cargo
+# will have compiled files and executables
+debug
+target
+
+# These are backup files generated by rustfmt
+**/*.rs.bk
+
+# MSVC Windows builds of rustc generate these, which store debugging information
+*.pdb
+
+# Generated by cargo mutants
+# Contains mutation testing data
+**/mutants.out*/
+
+# RustRover
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+dev-docs/
+code_reference/
+# Python
+__pycache__/

hydrust-0.5.0/.roborev.toml

@@ -0,0 +1 @@
+

hydrust-0.5.0/AGENTS.md

@@ -0,0 +1,66 @@
+# Hydrust
+
+## Common commands
+
+```bash
+# Test the project
+cargo test
+
+# Run linting checks
+cargo clippy
+```
+
+## Adding a feature
+
+**A settings key** — parse it in `initialize`, then add it to `CORE_SETTINGS` (or
+to the `feature_toggles!` list for an on/off switch, which registers the key for
+you). Update the counts in [tests/capabilities.rs](tests/capabilities.rs). In the
+extension: declare it in `package.json`, send it from `startServer`, add a
+`SETTING_COMPAT` entry.
+
+**A diagnostic rule** — add it to `diagnostic_rules!` in
+[src/diagnostics.rs](src/diagnostics.rs) and it is advertised automatically. In
+the extension, add a `RULE_COMPAT` entry; for a *rename*, record the old code as
+`previousCode` so the client can rewrite it for older servers, as
+`invalid-target` → `invalid-hydra-parameter` did in v0.3.0.
+
+**A behaviour the client must know about** — only needed when it depends on a
+client capability, or the client has to branch on it. Read the client capability
+in `initialize` and store the flag, add a field to `NegotiatedFeatures` and a
+`(name, gate)` pair to `SUPPORTED_FEATURES`, and gate the behaviour on that same
+flag so the advertised name is never a promise the session will not keep. Cover
+it both ways in [tests/capabilities.rs](tests/capabilities.rs). In the extension,
+add a `FEATURE_COMPAT` entry; names become `hydrust.supports.<name>` context keys.
+
+Anything the server always does and can advertise through a standard LSP
+capability field needs none of this.
+
+### Rules
+
+- Bump `HYDRUST_PROTOCOL_VERSION` only when something already in the block
+  changes meaning: a key that starts doing something different, a repurposed
+  feature name, a field that changes type. Additions never need a bump.
+- Never remove or repurpose a name quietly — a client may still send it.
+- Unknown settings keys stay ignored, never rejected.
+- `features` is always an array, even when empty; clients test membership on it.
+- Bump the crate version and add a CHANGELOG entry. The client's fallback table
+  is keyed on release versions.
+
+### Checking it
+
+`cargo test --test capabilities` covers the block's shape. In the extension repo,
+`npm run test:contract` checks the client against a real running binary, and
+`npm run test:table-audit` re-verifies the fallback table against tagged sources.
+
+## Threading
+
+The server runs its analysis on two `rayon` pools — a latency pool for hover,
+completion and semantic tokens, and a worker pool for diagnostics — alongside a
+single-threaded tokio runtime that handles the protocol itself. The `numThreads`
+setting is the total across all three, and defaults to a size the server picks to
+fit the machine.
+
+[docs/threading-model.md](docs/threading-model.md) is the reference for this:
+where each request handler does its work, why the counts are what they are, and
+which alternatives were tried and rejected. Read it before changing a thread
+count, the concurrency level, or where a handler runs.

hydrust-0.5.0/CHANGELOG.md

@@ -0,0 +1,109 @@
+# Changelog
+
+## [0.5.0]
+
+### Breaking
+
+- **One binary.** The `hydra-lsp` and `hydra-check` binaries are gone. Everything is now `hydrust`, with two subcommands: `hydrust check` and `hydrust server`. An editor that used to launch `hydra-lsp` with no arguments must launch `hydrust server`; the VSCode extension does this from v0.1.7 onwards.
+- **The crate is renamed `hydra-lsp` → `hydrust`**, so the release archives are now `hydrust-<target>.<ext>` and contain a `hydrust` executable. A VSCode extension older than v0.1.7 does not recognise the new asset name and silently stays on v0.4.2. `use hydra_lsp::` becomes `use hydrust::` for anything depending on the library.
+- **Removed `server` cargo feature**, along with `--no-default-features`.
+- `serverInfo.name` in the `initialize` response is now `"hydrust"`.
+- **`hydrust check --disable-rule` rejects an unknown rule** as a usage error (exit 2) instead of warning and continuing.
+- **`hydrust check` without `--workspace` resolves Python modules against the current directory**, even for a single file. `hydra-check` previously used the file's own directory.
+
+### Other changes
+
+- **Published to PyPI as `hydrust`.** e.g. `pip install hydrust`, `uv tool install hydrust`.
+- `hydrust check` now accepts multiple files and directories. Directories are walked recursively for `.yaml` and `.yml` files, honouring `.gitignore`; discovered files that carry no Hydra markers are skipped silently.
+- Added `--output-format github`, emitting GitHub Actions workflow commands so diagnostics render as inline annotations.
+- Renamed `--format` to `--output-format`.
+- JSON output is now a single document covering every checked file, rather than one document per file.
+- The JSON `summary` now counts diagnostics only in `total`, reports `other`, and counts files that could not be read or parsed under a separate `failed_files` field.
+- Reported paths always use `/` separators, so `--output-format github` annotations attach on Windows runners.
+- Directory walks no longer apply the user's global git excludes (`core.excludesFile`), so a local run and a CI run check the same files.
+- An unreadable directory or a file that disappears mid-walk is now logged and skipped, rather than aborting the whole run.
+- Finding no YAML files to check now warns on stderr and exits 0, rather than being a fatal error (exit 2).
+- The JSON `end_column` is now an inclusive 1-based column, matching `endColumn` in the github format; it was previously exclusive, so single-line values are one lower than before. A multi-line range that ends at the start of a line is reported as ending on the previous line, with `end_column` null (and `endColumn` omitted).
+- The workspace root falls back to the canonicalized current directory, so it matches the paths reported for each file.
+
+## [0.4.2]
+
+- Fixed lazy package exports declared under `if TYPE_CHECKING:` (or `if typing.TYPE_CHECKING:`) not resolving when combined with a module-level `__getattr__` and `__all__` (fixes #43)
+
+## [0.4.1]
+
+- Fixed relative re-exports failing to resolve when a package sits under more than one search root, such as a virtual environment stored inside the workspace. The most specific containing root is now used to convert a relative import to an absolute module name.
+
+## [0.4.0]
+
+- Added an incremental cache built on salsa (`HydraDatabase`), so YAML parses, Python target resolutions, and diagnostics are reused between requests instead of being recomputed on every keystroke
+- Cached YAML parsing per document version through the `DocumentInput`/`ParsedYaml` salsa inputs, replacing the previous `DocumentStore`
+- Cached Python definition lookups in `python_cache`, keyed on the target string and the resolved search paths
+- Routed Python source reads through `ruff_db::source_text` so resolved Python definitions participate in salsa's dependency graph
+- Removed the `FileCache` wrapper from `PythonAnalyzer` and `ImportResolver` in favour of a single `read_source` helper backed by salsa
+- Moved analysis off the async runtime onto two `rayon` pools (a latency pool for interactive requests, a worker pool for diagnostics), sized by the new `numThreads` initialization option
+- Answered `textDocument/semanticTokens/full` from a database snapshot on the latency pool instead of inline under the database lock
+- Sized the thread pools to a fixed 2 latency + 3 worker by default, rather than scaling the worker pool with the CPU count
+- Fixed the server hanging instead of exiting when the client's pipe closes while a handler is talking to it
+- Raised `tower-lsp`'s concurrency level from its default of 4 to 8
+- Clamped `numThreads` to 11 rather than to the CPU count. Beyond that the threads are unreachable rather than merely oversubscribed.
+- Registered `workspace/didChangeWatchedFiles` dynamically for `**/*.{py,pyi,pth}`, covering the workspace folders and — where the client supports relative patterns — each out-of-workspace site-packages root
+- Replaced the `PythonConfig::cache_revision` global counter with per-file `ruff_db::files::File::sync_path` invalidation in `did_change_watched_files`, so editing one Python file no longer evicts every cached resolution
+- Added per-directory `PthInventory` salsa input so `.pth` create/delete events invalidate editable-install resolution without flushing every cache entry
+- Added pull diagnostics (`textDocument/diagnostic`), returning unchanged reports via result IDs when nothing has changed, and falling back to push publishing for clients without pull support
+- Sent `workspace/diagnostic/refresh` after watched Python files change, so open configs pick up edits made outside the editor
+- Cleared diagnostics when a document stops being a Hydra file or is closed, instead of leaving stale entries in the client
+- Returned `ServerCancelled` with `retrigger_request` when a pull-diagnostic round is superseded by a newer edit
+- Fixed server panics caused by salsa cancellation escaping into the rayon pools
+- Fixed UTF-16 position handling for non-ASCII content, so positions match the encoding advertised to the client
+- Fixed `.pth` parsing to follow Python's lexical rules, and to only cache `.pth` files on the search path
+- Added `--version` and `--help` flags to the server binary, so a client can identify a binary before launching it; any other invocation still starts the stdio language server
+- Added a self-describing `capabilities.experimental.hydrust` block to the `initialize` response, listing the protocol version, the settings keys and diagnostic rules this build understands, and the coarse features actually switched on for the session after negotiating against the client's capabilities
+
+## [0.3.0]
+
+- Added class name for `__init__` diagnostics
+- Implemented `signature_help` for parameters
+- Added support for `_args_`, `_convert_`, and `_recursive_`
+- Renamed `invalid-target` diagnostic to `invalid-hydra-parameter` and included other checks within that rule
+
+## [0.2.0]
+
+- Refactored `yaml` parsing from `serde_yaml` to `saphyr`
+- Added suppression comments
+- Fixed issue where `missing-argument` diagnostic appeared for `cls` in class methods
+- Fixed issue whereby non-conventionally named first arguments would not be filtered out in instance and class methods
+
+## [0.1.5]
+
+- Added support for `_partial_`
+- Fixed issue of finding parent class docstring and signature when not overridden by child class
+- Fixed module resolution for `classmethod`s and `staticmethod`s
+- Refactored `PythonAnalyzer` to cache file reads
+- Fixed parameter placement issue when parameter's were above `_target_`
+
+## [0.1.4]
+
+- Fixed issue with `goto_definition` filepath for re-exported modules
+- Improved docstring rendering on hover
+
+## [0.1.3]
+
+- Fixed issue with `python` reimports not being found by `PythonAnalyzer`
+- Created `hydra-check` CLI tool
+- Removed dummy `CompletionResponse`s
+- Fixed issue in `YamlParser` whereby parameters weren't being fully recursively searched
+- Fixed issue with incorrect semantic token handling for sequences
+- Fixed issue of `.pth` files not being used for python import resolution
+
+## [0.1.2]
+
+- Refactor of `YamlParser::find_valid_target_key` to simplify implementation
+
+## [0.1.1]
+
+- Fixes to `YamlParser` when encountering commented out `_target_` key or values
+
+## [0.1.0]
+
+- Initial release