pseudocode-i18n-languageserver

Current version: 1.0.2. Release history: pseudocode-i18n-languageserver changelog.

Requirements and dependencies

• Python: 3.11 or newer (>=3.11).
• Pseudocode ecosystem: pseudocode-i18n >= 1.0.2, < 1.1.
• LSP stack: pygls >= 2.1.1, < 2.2 and lsprotocol == 2025.0.0.
• These runtime dependencies are declared by the package and are installed automatically by pip.

pseudocode-i18n-languageserver gives .pseudo and .algo files one reusable intelligence layer for open-source editors and educational IDEs. The server speaks the standard Language Server Protocol (LSP) over stdin/stdout and delegates the language itself to pseudocode-i18n.

Install it once:

python -m pip install 'pseudocode-i18n-languageserver>=1.0.2,<1.1'

On an externally managed Python installation, when you deliberately install into user/system packages:

python -m pip install --break-system-packages 'pseudocode-i18n-languageserver>=1.0.2,<1.1'

Check the command used by editors:

pseudo-lsp --version

You normally do not launch pseudo-lsp yourself. Kate, Neovim, Spyder or another LSP client starts it as a child process and communicates with it through stdin/stdout.

Shared editor intelligence

pseudo-lsp exposes the coordinated editor contract while keeping the language itself in pseudocode-i18n:

• multiple localized diagnostics from the recovering core linter, with exact source ranges; parser-recovery fallback details are localized too, so backend English text never leaks into a diagnostic in another Pseudocode language;
• completion plus retained frontend snippets (including localized XOR expressions, Do/Faire/... loop forms and distinct range/stepped-range/collection-loop snippets), signature help, document symbols, definitions, references, rename and formatting; every structural snippet is linked to its exact shared documentation card instead of being matched only by its first word;
• hover documentation generated by the core for ordinary keywords and structural grammar fragments such as Alors, fois, Jusqu'à and every word of localized multi-word import aliases such as en tant que; complete recognized structures take priority over ambiguous fragments, so French range-loop de, à, pas and Faire show the correct range/stepped-range syntax rather than import help;
• import operands remain symbols rather than documentation keywords, preventing false hover/help on names such as an imported sqrt;
• all 14 source languages receive the same behavior without maintaining language-specific tables in the server;
• project-aware standard/local module resolution and strict type/operation diagnostics are preserved;
• pseudocode/tutor exposes the shared pseudocode-tutor/v2 step-by-step execution model. Its optional fontSize presentation parameter reserializes the already cached flowchart snapshots without replaying the Tutor; the model otherwise includes static flowchart snapshots with stable node/edge IDs, compact incremental output deltas and source-language human labels/values while canonical machine fields remain stable. The response also carries the cached executable state and the exact LSP diagnostic set: error-level diagnostics return a blocked Tutor model without constructing an execution session, while warnings/information accompany a normal executable model; no second lint or parse is performed;
• pseudocode/flowchart exposes the same shared flowchart model for static previews/export as pseudocode-flowchart/v1. An optional fontSize parameter selects the shared renderer-neutral text size while reusing the cached semantic snapshot. Every response carries both canonical sources (sources.mermaid and sources.graphviz), Mermaid remains default_engine, and source follows the requested engine; nodes include conventional process / decision / io / terminal semantic kinds, range Pour uses the shared source-level head → body → exterior-return topology without synthetic init/update nodes, its return edge exposes the implicit increment as presentation annotation metadata, Si/Sinon Si diamonds expose question-form labels, and loop steps expose both taken and natural-exit edge activation when known. Semantic diagnostics do not block a graph when the original AST is structurally available. Symbol scopes carry localized display_label text beside canonical name / kind, and rich Tutor clients can render once and toggle stable IDs locally rather than requesting a new graph per step. Mermaid receives the core-owned validated-loop-attached-edges-v22 ELK profile; Graphviz receives deterministic neato positions with WEST loop-back / EAST exit routing. The server duplicates neither semantics nor geometry;
• each open-document model retains its parsed AST and caches Tutor sessions, symbol/scope lists and flowchart snapshots by document version, so diagnostics and custom requests reuse work already performed for the current source;
• custom Pseudocode requests share one document-parameter normalizer, accepting standard textDocument.uri plus compatibility fallbacks without duplicating parsing in editor plugins.

The server remains a thin LSP layer: source-language knowledge belongs to pseudocode-i18n, so VS Code, Thonny and other clients see the same diagnostics and help.

Language-server capabilities

The server exposes the same language intelligence for all 14 supported natural languages:

• diagnostics from the native pseudocode-i18n parser and semantic linter, including multiple independent errors in one document, explicit type-contract mismatches and incompatible typed operations;
• project-aware module diagnostics: math and random are standard, while existing local .pseudo / .algo modules are resolved from the document/project path;
• i18n code completion and the existing structural snippets, using the normal language-resolution priority;
• completion for document symbols, localized built-ins, constants and types;
• pedagogical hover information for symbols, keywords, types and built-ins, with localized section labels and every accepted syntax synonym exposed;
• signature help;
• document symbols;
• go to definition;
• find references;
• prepare rename and rename;
• whole-document formatting; French repeat-until formatting canonically emits Jusqu'à ... while the longer accepted synonym remains valid;
• .pseudo and .algo as equivalent source extensions.

The server does not maintain a second grammar. pseudocode-i18n remains the source of truth for parsing, language detection, configuration, formatting, diagnostics, snippets and semantic indexing.

Language and snippet resolution

A document uses the same precedence as the core:

1. explicit # language: xx / # lang: xx directive;
2. language: / lang: in pseudocode.config.yml;
3. automatic language detection;
4. fallback_language;
5. French as the final default.

This means typing a condition snippet in a French file proposes Si … Alors, while an English file proposes If … Then, without editor-specific copies of the snippet catalog.

Editor targets

See the Editors section of the documentation for exact configuration and current limitations.

Development

yarn setup
yarn bfc

yarn setup installs the current project and uses a sibling ../pseudocode-i18n checkout when present. yarn bfc is an alias for yarn build:full:check and performs version synchronization, generated integrations, linting, tests, a CLI smoke test, strict MkDocs documentation build and Python package build.

Useful commands:

yarn dev
yarn dev:lan
yarn test
yarn lint
yarn smoke
yarn zip

License

GNU General Public License version 3 or later (GPL-3.0-or-later).