autodocgenerator 0.8.9.6

This Project helps you to create docs for your projects

Executive Navigation Tree

• 📂 Setup & Configuration

  • Install Workflow Setup
  • Configuration Loading And Validation
  • Manager Configuration
  • Projectsettings Prompt Builder
  • Command Line Invocation Logic

• ⚙️ Documentation Generation

  • Documentation Pipeline Trigger
  • Execution Flow Summary
  • Documentation Generation Workflow
  • Autodocfile Parameters
  • Docfactory Orchestration
  • Global Introduction Generation
  • Intro With Links Generation
  • Custom Description Generation
  • Anchor‑Ordering‑Cleanup
  • Anchor‑Chunk‑Splitting
  • Semantic‑Ordering
  • Html Link Extraction
  • Factory‑Doc‑Assembly
  • Doc‑Parts‑Generation
  • Custommodule Intro Modules

• 🤖 Model Orchestration

  • Asyncgptmodel Implementation
  • Gptmodel Synchronous Flow
  • Parentmodel Setup And Rotation
  • Synchronous‑Part‑Doc‑Generator
  • Asynchronous‑Part‑Doc‑Generator
  • Synchronous‑Multi‑Part‑Orchestrator
  • Asynchronous‑Multi‑Part‑Orchestrator

• 🔀 Data Splitting & Repository

  • Spliter Entry Point
  • Data‑Splitting Loop
  • Repository‑Mix Builder
  • Code‑Mix Generation

• 🗄️ Caching & Compression

  • Cache‑File Access
  • Compress Function
  • Batch Compression Sync
  • Batch Compression Async

• 📊 Logging & Progress

  • Singleton‑Logger‑Implementation
  • Log‑Message‑Hierarchy
  • Progress‑Abstraction
  • Rich‑Implementation
  • Console‑Task‑Helper
  • Fallback‑Console‑Progress

To set up the installation workflow, run the PowerShell script on Windows using:
irm raw.githubusercontent.com/Drag-GameStudio/ADG/main/install.ps1 | iex

On Linux‑based systems, execute the shell script with:
curl -sSL raw.githubusercontent.com/Drag-GameStudio/ADG/main/install.sh | bash

Additionally, add a secret variable named GROCK_API_KEY to your GitHub Actions configuration, containing the API key obtained from the Grock documentation site (grockdocs.com).

Configuration Loading and Validation

• load_config() reads the global settings file, merges environment overrides, and returns a structured config object.
• Immediate checks ensure required keys (e.g., source_path, output_dir, templates) exist; missing keys raise ConfigurationError.
• The validated config is the sole input to the pipeline, guaranteeing deterministic behavior.

Manager Class – Configuration & State

Manager orchestrates the end‑to‑end documentation pipeline. It receives the project root, a Config object, optional LLM model instances (Model / AsyncModel), and a progress UI (BaseProgress). During construction it:

• Stores configuration and progress objects.
• Instantiates a file‑based logger that writes to report.txt inside a hidden cache folder (.auto_doc_cache).
• Guarantees the cache directory exists, creating it if necessary. ProjectSettings Prompt Builder
  ProjectSettings.prompt concatenates the global BASE_SETTINGS_PROMPT with the project name and any key/value pairs added via add_info. The resulting string feeds the system role of every compression request.

Command‑Line Invocation Logic

The if __name__ == "__main__": block parses CLI arguments (if any), invokes load_config() from autodocgenerator.engine.config.config, validates the returned dictionary, and passes it to run_documentation_pipeline() which orchestrates the end‑to‑end doc generation.

Documentation Pipeline Trigger

run_documentation_pipeline(config) sequentially executes:

1. Source Discovery – walks config["source_path"] to collect parsable modules.
2. Parsing Engine – feeds each file to the parser subsystem, producing intermediate ASTs.
3. Renderer – transforms ASTs into Markdown/HTML using the selected template set.
4. Writer – writes rendered files into config["output_dir"], optionally cleaning stale artifacts.

Side effects include filesystem writes, optional logging to config["log_file"], and temporary cache creation.

Execution Flow Summary

1. CLI start → load & validate config.
2. Valid config → run_documentation_pipeline.
3. Pipeline → generated documentation persisted on disk.

All exceptions propagate to the top level, where a concise error message is printed and the process exits with a non‑zero status.

Documentation Generation Workflow

Responsibility
autodocgenerator/auto_runner/run_file.py orchestrates the end‑to‑end generation of project documentation. It loads the user configuration, instantiates synchronous and asynchronous GPT models, creates a Manager that drives file parsing, doc‑part creation, factory‑based rendering, ordering, and cache cleanup, finally returning the assembled markdown.

Interactions

• Config Reader – imports read_config from config_reader.py to parse autodocconfig.yml.
• Model Layer – creates GPTModel (blocking) and AsyncGPTModel (non‑blocking) using the API key from engine/config/config.py.
• Manager – autodocgenerator.manage.Manager receives the project path, configuration, models, and a progress bar (ConsoleGtiHubProgress).
• Factories – DocFactory builds doc sections from custom modules (CustomModule) and from the built‑in IntroLinks.
• UI – progress feedback is sent to GitHub‑style console via ConsoleGtiHubProgress.

Technical Details

def gen_doc(project_path: str, config: Config, custom_modules):
    sync_model = GPTModel(API_KEY, use_random=False)
    async_model = AsyncGPTModel(API_KEY)

    manager = Manager(
        project_path,
        config=config,
        sync_model=sync_model,
        async_model=async_model,
        progress_bar=ConsoleGtiHubProgress(),
    )
    # Core pipeline
    manager.generate_code_file()
    manager.generete_doc_parts(max_symbols=7000)
    manager.factory_generate_doc(DocFactory(*custom_modules))
    manager.order_doc()
    manager.factory_generate_doc(DocFactory(IntroLinks()))
    manager.clear_cache()
    return manager.read_file_by_file_key("output_doc")

• generate_code_file() scans the repository, respecting Config.ignore_files.
• generete_doc_parts() chunks source code (≤ 7000 symbols) and queries the GPT models.
• Two factory_generate_doc calls render custom user‑supplied modules first, then prepend standard intro links.
• order_doc() ensures a logical section order before writing the final file.

Data Flow

1. Input – project_path (root of the repo) and a fully populated Config object (language, ignore patterns, project info).
2. Processing – Files → code extraction → GPT prompts → text fragments → module factories → ordered markdown.
3. Outputs – Cached files under .auto_doc_cache/, and the final document string returned by read_file_by_file_key("output_doc").
4. Side Effects – Writes intermediate cache, updates progress UI, and may raise exceptions from model calls or file I/O.

This module acts as the command‑line entry point (if __name__ == "__main__":) that ties configuration loading to the documentation pipeline. The file is a YAML document that defines the behavior of the documentation generator.
Key sections and available options:

• project_name – a short title for the project.
• language – language code for the generated text, default “en”.
• project_options – a map of boolean and numeric controls:
  • save_logs – true to keep generation logs, false otherwise.
  • log_level – integer indicating the verbosity of log output.
• project_additional_info – free‑form key/value pairs that are inserted into the generated material (e.g., a global idea or description).
• ignore_files – list of glob patterns for files that should be skipped during analysis.
• custom_descriptions – an array of strings; each string is interpreted as a custom instruction for the generator, allowing you to request specific sections or explanations.

When writing the file, list each option under its heading using proper YAML indentation. Only the sections you need are required; omitted sections will use the defaults (language “en”, empty project name, default project_options).

DocFactory Orchestration

DocFactory receives an ordered list of BaseModule subclasses. generate_doc creates a progress sub‑task, invokes each module’s generate(info, model), concatenates their outputs, and logs module completion. It returns the assembled documentation string. Global Introduction Generation
get_introdaction (note spelling) builds a similar prompt using BASE_INTRO_CREATE and the whole documentation body, then forwards it to the model. The raw intro string is returned unchanged.

Introduction‑With‑Links Generation
get_links_intro receives the link list and a Model implementation. It composes a three‑message prompt (system language directive, BASE_INTRODACTION_CREATE_TEXT, and the link payload) and calls model.get_answer_without_history. Logging surrounds the call, and the generated introduction text is returned.

Custom Description Generation
generete_custom_discription iterates over split document chunks, sending each to the model with a strict system prompt that forces a single‑anchor output (<a name="…"></a>). It aborts on the first non‑error response, otherwise returns an empty string. The rules prevent filename, extension, or external URL leakage.

Anchor‑Based Ordering & Cache Cleanup

order_doc extracts anchor‑segmented sections via split_text_by_anchors, asks the LLM (self.sync_model) to compute the correct order (get_order), and rewrites the file.

clear_cache removes report.txt unless config.pcs.save_logs is true, ensuring optional log retention.

Data Flow Summary – Input files → CodeMix → raw mix → gen_doc_parts → partial doc → DocFactory → enriched doc → split_text_by_anchors/get_order → final ordered doc. Side effects include file writes, logger entries, and progress UI updates.

Anchor‑Based Chunk Splitting
split_text_by_anchors uses a look‑ahead regex to split a full doc into sections that start with a valid anchor (<a name="…"></a>). It validates that each chunk yields a corresponding link via extract_links_from_start; mismatches return None. The result is a dict mapping “#anchor” keys to their text blocks.

Semantic Ordering of Documentation Sections
get_order receives the chunk dict and a Model. It logs the incoming keys, prompts the model to return a comma‑separated, semantically sorted list of titles (preserving the leading “#”). The function reassembles the final ordered document by concatenating the chunks in the returned order, logging each addition.

HTML Link Extraction Logic
get_all_html_links scans a documentation string for anchor tags (<a name=…></a>). It logs start/end messages via BaseLogger, builds a regex pattern, iterates with re.finditer, and appends up‑to‑five links prefixed with “#”. The function returns the collected list, providing the first‑stage data for downstream ordering.

Factory‑Driven Documentation Assembly

factory_generate_doc loads the current output and code mix, builds an info dict (language, full_data, code_mix), and logs the module chain of the supplied DocFactory. It then calls doc_factory.generate_doc(info, self.sync_model, self.progress_bar). The factory‑produced fragment is prepended to the existing doc and persisted.

Synchronous Documentation Chunking

generete_doc_parts (typo retained) reads the code‑mix, then invokes gen_doc_parts with:

• raw code mix,
• max_symbols limit (default 5 000),
• the synchronous LLM (self.sync_model),
• target language (self.config.language),
• the progress UI.

The resulting Markdown is written to output_doc.md and progress updated.

CustomModule & Intro Modules

• CustomModule injects a user‑provided description into a generated custom intro by preprocessing code via split_data and delegating to generete_custom_discription.
• IntroLinks extracts HTML links from info["full_data"] and builds a linked introduction using get_links_intro.
• IntroText produces a plain introduction via get_introdaction.

All modules depend on a Model instance for LLM calls and output plain‑text Markdown/HTML fragments.

AsyncGPTModel Implementation

AsyncGPTModel extends AsyncModel to call Groq’s async client. It builds a shuffled list of candidate model names (regen_models_name) from the global config, logs each step via BaseLogger, and retries on failure, cycling through the list. Input: optional prompt or full history; output: generated answer string. Side‑effects: async HTTP request, log entries, possible ModelExhaustedException if no models remain.

GPTModel Synchronous Flow

GPTModel mirrors AsyncGPTModel but uses the synchronous Groq client. It follows the same retry loop, logs progress, and returns the answer. It also respects the with_history flag to select either the stored conversation (self.history.history) or a raw prompt.

ParentModel Setup & Model Rotation

ParentModel (base for both sync/async) stores the API key, a History instance, and prepares regen_models_name—a shuffled copy of MODELS_NAME when use_random=True. It tracks current_model_index to rotate through candidates after each failure. Synchronous Part Documentation Generator (write_docs_by_parts)
Builds a two‑message system prompt (language + part‑ID, then BASE_PART_COMPLITE_TEXT). If a previous part’s summary exists, it is appended as an additional system message. The user message contains the raw source fragment. The Model’s get_answer_without_history call returns a markdown block; surrounding triple‑backticks are stripped before the final string is returned. Logs start, length, and raw answer (level 2).

Asynchronous Part Documentation Generator (async_write_docs_by_parts)
Mirrors the synchronous flow but runs inside an async with semaphore to cap concurrency (default 4). Uses an AsyncModel for non‑blocking await get_answer_without_history. An optional update_progress callback is invoked after each answer. Logging mirrors the sync variant.

Synchronous Multi‑Part Documentation Orchestrator (gen_doc_parts)

1. Calls split_data to obtain splited_data.
2. Creates a progress sub‑task.
3. Sequentially invokes write_docs_by_parts for each chunk, concatenating results into all_result.
4. Keeps a 3000‑character tail of the previous answer to provide context for the next part.
5. Updates the progress bar after each iteration and logs final output length.

Asynchronous Multi‑Part Documentation Orchestrator (async_gen_doc_parts)
Splits the input, creates a semaphore (max 4), and dispatches async_write_docs_by_parts for every chunk via asyncio.gather. Progress updates are wired through a lambda calling progress_bar.update_task(). Results are concatenated in order, progress sub‑task is removed, and the assembled documentation is logged.

Spliter Entry Point (split_data) – Partial
split_data(data, max_symbols) prepares to split a large document into sub‑strings respecting a symbol limit. The implementation continues beyond the shown snippet, but its purpose is to feed the compressor pipeline with appropriately sized chunks.

Data Splitting Loop (split_data)
Iteratively refines a list of file‑derived chunks (splited_by_files) so that no element exceeds 1.5 × max_symbols. Oversized entries are bisected at the halfway point, inserted back, and the process repeats until stability. The second phase packs these normalized fragments into split_objects, inserting line breaks and respecting a 1.25 × max_symbols buffer. Returns a list of size‑constrained text blocks ready for downstream processing.

Repository Content Aggregation (CodeMix)
CodeMix walks a repository rooted at root_dir, respecting ignore_patterns. should_ignore checks each path against glob patterns, file basenames, and any path component. build_repo_content writes a hierarchical tree view to repomix-output.txt, then appends each non‑ignored file’s relative path (<file path="…">) followed by its raw content. Progress is logged throughout.

Data Flow:

• Source files → CodeMix → mixed text file → split_text_by_anchors → chunk dict → get_order → ordered doc → optional get_links_intro/get_introdaction/generete_custom_discription → final documentation output.
• Side effects: file creation (repomix-output.txt), logger entries, and UI progress updates.

Code Mix Generation Workflow

generate_code_file creates a CodeMix instance (respecting config.ignore_files) and calls build_repo_content to serialize the entire repository into code_mix.txt. Logging marks start/end, and the progress bar task is advanced.

Cached File Access Helpers

• get_file_path(file_key) builds an absolute path inside the cache using the static FILE_NAMES map.
• read_file_by_file_key(file_key) opens the derived path, reads UTF‑8 content and returns it. These utilities centralise path handling for all subsequent steps. Compress Function Logic
  compress(data, project_settings, model, compress_power) builds a three‑message prompt: the project‑specific system prompt, a compression directive from get_BASE_COMPRESS_TEXT, and the raw data. It forwards this prompt to model.get_answer_without_history and returns the model’s summary.
  Input: plain text, ProjectSettings instance, Model, integer power.
  Output: compressed string.
  Side‑effects: none.

Synchronous Batch Compression (compress_and_compare)
Partitions a list of strings into groups of compress_power. For each element it calls compress, concatenates results per group, and updates a BaseProgress sub‑task. Returns a list where each entry aggregates the compressed texts of one group.

Asynchronous Batch Compression (async_compress_and_compare)
Creates a semaphore (max 4 concurrent calls) and launches async_compress for every element. Each coroutine builds the same three‑message prompt, awaits model.get_answer_without_history, and updates progress. After asyncio.gather, groups results into chunks of size compress_power and returns the aggregated list.

Singleton Logger Implementation (BaseLogger)
BaseLogger.__new__ ensures a single shared instance. Clients set a concrete BaseLoggerTemplate (e.g., FileLoggerTemplate) via set_logger. Calls to log forward to the template’s global_log, which respects the configured log‑level filter.

Log Message Hierarchy (BaseLog & subclasses)
BaseLog stores a message and level; subclasses (InfoLog, WarningLog, ErrorLog) override format() to prepend a timestamp and severity tag. The hierarchy enables uniform, level‑aware console or file output across the documentation pipeline.

Progress Abstraction (BaseProgress)
Defines the minimal interface for creating, updating, and removing sub‑tasks. Concrete classes implement these hooks so the documentation pipeline can switch between rich‑based UI or plain console output without code changes.

Rich‑Library Implementation (LibProgress)
Wraps rich.Progress:

• __init__ registers a base task (General progress) with a configurable total (default 4).
• create_new_subtask(name, total_len) adds a child task and stores its ID.
• update_task() advances the current sub‑task if present, otherwise the base task.
• remove_subtask() clears the reference, allowing the next chunk to start fresh.
  All calls forward to rich.Progress.update, guaranteeing thread‑safe visual feedback.

Console Task Helper (ConsoleTask)
Utility that prints a simple progress line.

• start_task() emits the start banner.
• progress() increments an internal counter, computes a percentage, and prints it.
  Used by the fallback console progress class.

Fallback Console Progress (ConsoleGtiHubProgress)
Implements the same API as BaseProgress for environments lacking Rich:

• Holds a persistent “General Progress” ConsoleTask.
• create_new_subtask spawns a dedicated ConsoleTask.
• update_task delegates to the active task or the general one.
• remove_subtask discards the current sub‑task.

Data Flow – Caller (e.g., gen_doc_parts) invokes create_new_subtask → progress updates via update_task → optional remove_subtask. No side effects beyond console/rich output. The abstraction keeps the rest of the system agnostic to the UI backend.