autodocgenerator 0.9.2.7

This Project helps you to create docs for your projects

Executive Navigation Tree

• 📘 Introduction
  • Basic Introduction
  • Intro Text
  • Intro Links
  • Intro With Links
  • CONTENT DESCRIPTION
• ⚙️ Setup & Configuration
  • Install Workflow Setup
  • Autodoc Yml Structure
  • Read Config
• 🛠️ Core Functions
  • _Print Welcome Function
  • Logger Initialization
  • Logging Infrastructure
  • Base Progress
  • Lib Progress
  • Console Github Progress
• 📦 Manager
  • Manager Class
  • Manager Class Usage
• 🧩 Modules
  • Base Module Abstract
  • Custom Module
  • Custom Module Without Context
  • Docfactory Orchestrator
• 🤖 Models
  • Parent Model
  • Gpt Model
  • Async Gpt Model
  • Semantic Ordering Llm
• 📡 Extraction & Processing
  • Html Link Extraction
  • Anchor Extraction Chunk Splitting
  • Regex Pattern
• 📝 Documentation Generation
  • Gen Doc
  • Gen Doc Parts
  • Write Docs By Parts
  • Custom Description Template
  • Custom Description Iterative
• 📚 Repository & Mixing
  • Repository Mixer
• 📦 Compression
  • Compressor Core Function
  • Batch Compression Compare
  • Iterative Full Compression
• 🔄 Data Loop
  • Split Data Loop

Basic Introduction Generation (get_introdaction)

Entity	Type	Role	Notes
global_data	str	Input	Full documentation content.
model	Model	Input	LLM interface.
language	str	Input (default "en")	Language hint.
intro	str	Output	Generated introduction.

Builds a prompt with BASE_INTRO_CREATE and global_data, then returns the LLM answer.

IntroText – Global Data Intro Builder

Responsibility – Generates a textual introduction from global_data.
Visible Interactions – Calls get_introdaction.
Logic Flow –

1. generate(info, model) →
   • intro = get_introdaction(info.get("global_data"), model, info.get("language")).
   • Return intro.

Warning: All modules assume the presence of specific keys (code_mix, full_data, global_data, language). Absence results in None being passed to downstream functions, which may raise runtime errors if those functions lack internal checks.

IntroLinks – HTML Link Intro Builder

Responsibility – Extracts HTML links from full_data and builds an introductory paragraph.
Visible Interactions – Calls get_all_html_links, get_links_intro.
Logic Flow –

1. generate(info, model) →
   • links = get_all_html_links(info.get("full_data")).
   • intro_links = get_links_intro(links, model, info.get("language")).
   • Return intro_links.

Introduction Generation with Links (get_links_intro)

Entity	Type	Role	Notes
links	list[str]	Input	Anchor list from get_all_html_links.
model	Model	Input	Provides get_answer_without_history.
language	str	Input (default "en")	Language hint for the LLM.
intro_links	str	Output	Generated introductory text.
logger	BaseLogger	Side‑effect	Logs start, completion, and result.

Creates a prompt containing the language directive, BASE_INTRODACTION_CREATE_LINKS, and the stringified links. Calls model.get_answer_without_history and returns the LLM’s response. ` tag with strict naming constraints, then returns the LLM’s response.

Warning – All functions assume the provided Model instance implements get_answer_without_history. Missing methods will raise AttributeError. The logger must be operational; otherwise logging calls will fail.

PowerShell installation (Windows)
Execute the following command in an elevated PowerShell window to fetch and execute the installation script directly from the repository:

irm raw.githubusercontent.com/Drag-GameStudio/ADG/main/install.ps1 | iex

• irm (Invoke‑WebRequest) downloads the script content.
• The pipeline to iex (Invoke‑Expression) runs the script in the current session.

Shell installation (Linux/macOS)
Run this one‑liner in a terminal to retrieve and execute the Linux installer:

curl -sSL raw.githubusercontent.com/Drag-GameStudio/ADG/main/install.sh | bash

• curl -sSL silently follows redirects and outputs the script.
• The output is piped to bash for immediate execution.

GitHub Actions secret configuration
To enable the workflow to interact with the Grock service, add a secret named GROCK_API_KEY to the repository’s GitHub Actions secrets:

1. Navigate to the repository’s Settings → Secrets and variables → Actions.
2. Click New repository secret.
3. Enter Name: GROCK_API_KEY.
4. Paste the API key obtained from the Grock documentation.
5. Save the secret.

The workflow will automatically read GROCK_API_KEY from the environment, allowing authenticated calls to the Grock API during execution. The configuration file is a YAML document that defines several top‑level sections:

Project metadata

• project_name: a string that specifies the name of the project.
• language: the language code (e.g., "en").

Files to ignore

• ignore_files: a list of directory or file names that should be excluded from documentation generation (e.g., "dist").

Build settings (build_settings block)

• save_logs: boolean, when true the generation process stores logs.
• log_level: integer controlling verbosity (e.g., 2).

Structure settings (structure_settings block)

• include_intro_links: boolean, adds introductory links if true.
• include_order: boolean, keeps the original order of sections when true.
• max_doc_part_size: integer, maximum size (in characters) for each documentation part.

Additional project information (project_additional_info block)

• Custom key/value pairs, such as global idea, provide free‑form description of the project’s purpose.

Custom descriptions (custom_descriptions block)

• A list of strings that can contain special instructions or explanatory text. These strings are processed by the generator to enrich the final documentation (e.g., how to install the workflow, how to write the configuration file, how to use the Manager class).

When creating the file, follow standard YAML syntax: use indentation (two spaces) for nested sections, place lists under a dash (-), and ensure string values are quoted if they contain special characters. This structure enables the documentation generator to read all required options and produce the desired output.

read_config – YAML Configuration Parsing

Entity	Type	Role	Notes
file_data	str	Raw YAML content	Passed by caller
data	dict	Parsed YAML tree	Result of yaml.safe_load
config	Config	Central project configuration	Instantiated locally
ignore_files	list[str]	Files/patterns to skip	Defaults from YAML or empty
language	str	Documentation language	Defaults to "en"
project_name	str	Identifier for the project	May be None
project_additional_info	dict	Arbitrary key‑value pairs	Empty dict if omitted
pcs	ProjectBuildConfig	Build‑time settings container	Populated via load_settings
custom_modules	`list[CustomModule	CustomModuleWithOutContext]`	User‑defined processing hooks
structure_settings_object	StructureSettings	Runtime doc‑structure flags	Loaded from structure_settings

Logic Flow

1. Parse file_data with yaml.safe_load.
2. Instantiate a fresh Config.
3. Extract top‑level keys (ignore_files, language, project_name, project_additional_info, build_settings).
4. Load build settings into a ProjectBuildConfig and attach to config via set_pcs.
5. Populate ignore patterns and additional info on config.
6. Build custom_modules list, choosing CustomModuleWithOutContext when a description starts with "%".
7. Initialise StructureSettings, apply any overrides from YAML.
8. Return tuple (config, custom_modules, structure_settings_object).

Assumption – All expected YAML sections exist; missing sections yield defaults as coded.

_print_welcome Function Execution

Entity	Type	Role	Notes
BLUE, BOLD, CYAN, RESET	str	ANSI colour codes	Used only within this function
ascii_logo	str	Formatted ASCII art	Combines colour codes for visual header
print (built‑in)	callable	Output side‑effect	Sends logo and status lines to stdout

Purpose – Render a coloured ASCII banner and a one‑line status message when the package is imported.

Logic Flow

1. Define colour constants (BLUE, BOLD, CYAN, RESET).
2. Build ascii_logo using an f‑string that inserts the colour constants.
3. Print the logo, then a line showing “ADG Library | Status: Ready to work” with cyan text.
4. Print a separator line ('—' * 35).

Warning – This function executes at import time, causing I/O side‑effects even if the consumer only needs the library’s API.

Logger Initialization

Entity	Type	Role	Notes
BaseLogger	class	Core logger object	Instantiated as logger
BaseLoggerTemplate	class	Formatting/template provider	Passed to logger.set_logger
InfoLog, ErrorLog, WarningLog	classes	Log‑level helpers (imported but not used here)	Exported for downstream modules

Purpose – Prepare a singleton‑style logger ready for use throughout the package.

Logic Flow

1. Import logger classes from .ui.logging.
2. Create logger = BaseLogger().
3. Apply a default template via logger.set_logger(BaseLoggerTemplate()).

Interactions – Subsequent modules import logger from this package; they rely on the pre‑configured template for consistent output formatting. No external configuration files are consulted in this fragment.

---

All information is derived exclusively from the provided autodocgenerator/__init__.py source.

Logging Infrastructure – BaseLog Hierarchy & Singleton BaseLogger

Entity	Type	Role	Notes
BaseLog	class	Abstract	Stores message and level; provides _log_prefix.
ErrorLog, WarningLog, InfoLog	subclasses	Concrete	Override format() with severity tag.
BaseLoggerTemplate	class	Abstract logger	Holds log_level; routes via global_log.
FileLoggerTemplate	subclass	File‑output logger	Writes formatted logs to a file.
BaseLogger	singleton class	Proxy	Delegates log() to assigned logger_template.

Logic Flow

1. BaseLogger.__new__ ensures a single instance.
2. set_logger() injects a concrete BaseLoggerTemplate.
3. log() forwards the BaseLog object to logger_template.global_log, which respects log_level.

Assumption – A logger template is set before any BaseLogger.log call; otherwise an AttributeError would occur.

BaseProgress – Abstract Progress Interface

Entity	Type	Role	Notes
BaseProgress	class	Abstract base	Defines the progress API used by the documentation generator.
create_new_subtask	method(name: str, total_len: int) → None	Abstract	Intended to start a child progress track.
update_task	method() → None	Abstract	Advances either the current sub‑task or the base task.
remove_subtask	method() → None	Abstract	Clears the active sub‑task reference.

Component Responsibility
Provides a minimal contract that concrete progress reporters must implement, allowing the generator to switch between rich‑terminal, console, or other visual back‑ends without code changes.

Visible Interactions
LibProgress and ConsoleGtiHubProgress inherit from it and are passed to higher‑level functions (e.g., gen_doc_parts) as the progress_bar argument.

Technical Logic Flow

1. Instantiation does nothing (__init__ is empty).
2. Sub‑classes override the three abstract methods to manipulate their own state.

Data Contract

Entity	Type	Role	Notes
name	str	Input to create_new_subtask	Human‑readable identifier of the sub‑task.
total_len	int	Input to create_new_subtask	Expected number of update_task calls.
current progress	internal	Side‑effect	Updated via update_task; no external return.

Assumption – The base class is never used directly; calling its methods without a concrete implementation would raise NotImplementedError.

LibProgress – Rich‑Library Progress Implementation

Entity	Type	Role	Notes
LibProgress	class	Concrete BaseProgress	Wraps rich.progress.Progress.
progress	Progress	Dependency	Provided by the rich library.
_base_task	int	Internal	ID of the top‑level “General progress” task.
_cur_sub_task	`int	None`	Internal

Component Responsibility
Renders hierarchical progress bars in terminals that support ANSI graphics, tracking overall and per‑chunk documentation generation.

Visible Interactions

• create_new_subtask(name, total_len) registers a new rich task and stores its ID.
• update_task() advances either _cur_sub_task or _base_task.
• remove_subtask() discards the current sub‑task reference, leaving the base task alive.

Technical Logic Flow

1. Constructor receives a Progress instance and creates the base task (total defaults to 4).
2. create_new_subtask calls self.progress.add_task(name, total=total_len) and caches the ID.
3. update_task checks _cur_sub_task; if set, updates it, otherwise updates the base task.
4. remove_subtask simply null‑ifies _cur_sub_task.

Data Contract – same as abstract plus the internal task IDs.

ConsoleGtiHubProgress – Simple Console Fallback

Entity	Type	Role	Notes
ConsoleGtiHubProgress	class	Concrete BaseProgress	Prints textual progress to stdout.
gen_task	ConsoleTask	Internal	Represents the overall “General Progress”.
curr_task	`ConsoleTask	None`	Internal

Component Responsibility
Provides a lightweight, environment‑agnostic progress indicator when the rich library cannot be used (e.g., CI logs).

Visible Interactions

• Mirrors the abstract API: create_new_subtask, update_task, remove_subtask.
• Delegates actual printing to ConsoleTask.progress().

Technical Logic Flow

1. Instantiation creates a ConsoleTask named “General Progress” with 4 steps.
2. create_new_subtask replaces curr_task with a new ConsoleTask.
3. update_task calls curr_task.progress() if a sub‑task exists; otherwise updates gen_task.
4. remove_subtask clears curr_task, causing subsequent updates to affect the general task again.

Data Contract – identical to the abstract contract; progress is emitted via print.

Warning – This implementation does not persist state beyond the current process; restarting the script will reset all counters.

Manager – Orchestrator for Documentation Pipeline

Responsibility – Coordinates preprocessing, LLM‑driven generation, post‑processing and caching for a single project directory.

Visible Interactions –

• Imports from preprocessor, postprocessor, engine.models, ui, factory.
• Uses BaseProgress/LibProgress for task updates, BaseLogger for log emission, DocFactory for modular doc assembly.

Logic Flow –

1. __init__ stores paths, config, models, logger and ensures a cache folder exists.
2. read_file_by_file_key → open cached file → return its contents.
3. get_file_path → compose absolute path under CACHE_FOLDER_NAME.
4. generate_code_file → instantiate CodeMix, build repository content into code_mix.txt, log start/finish, update progress.
5. generate_global_info → read code_mix.txt, split via split_data, compress with compress_to_one, write global_info.md, update progress.
6. generete_doc_parts → optionally read global_info.md, call gen_doc_parts with sync model & settings, write output_doc.md, update progress.
7. factory_generate_doc → read current doc & code mix, build info dict, log module list & key sizes, invoke doc_factory.generate_doc, prepend result to existing doc, update progress.
8. order_doc → split current doc by anchors, reorder sections via get_order, overwrite output_doc.md.
9. clear_cache → delete log file if save_logs is false.

Data Contract

Entity	Type	Role	Notes
project_directory	str	Input	Root of the target code base.
config	Config	Input	Supplies ignore_files, language, pbc.log_level, pbc.save_logs, get_project_settings().
sync_model / async_model	Model / AsyncModel	Input	LLM interface passed to preprocessors/post‑processors.
progress_bar	BaseProgress	Input/Side‑effect	Must implement update_task().
self.logger	BaseLogger	Side‑effect	Writes to report.txt via FileLoggerTemplate.
Cached files (code_mix.txt, global_info.md, output_doc.md, report.txt)	on‑disk	Persistent storage	Created/overwritten by manager methods.
info (in factory_generate_doc)	dict	Input to DocFactory.generate_doc	Keys: language, full_data, code_mix.
Return values	None (except read_file_by_file_key)	Side‑effects	Methods write files or update progress; only read_file_by_file_key returns file contents.

Warning – The manager assumes all configured keys exist; missing files raise FileNotFoundError, and absent config attributes cause AttributeError. !noinfo

BaseModule Abstract Interface

Responsibility – Defines the contract for all documentation generation modules.
Visible Interactions – Sub‑classes inherit from this ABC and are invoked by DocFactory.
Logic Flow –

1. Constructor (__init__) does nothing (pass).
2. Declares abstract generate(info: dict, model: Model) method.

CustomModule – Context‑Aware Description Generator

Responsibility – Generates a custom description using the supplied discription and a code split.
Visible Interactions – Calls generete_custom_discription, split_data.
Logic Flow –

1. Store discription at init.
2. generate(info, model) →
   • Retrieve code_mix via info.get("code_mix").
   • Split with split_data(..., max_symbols=5000).
   • Invoke generete_custom_discription(split_result, model, self.discription, info.get("language")).
   • Return the resulting string.

CustomModuleWithOutContext – Description‑Only Generator

Responsibility – Produces a description without processing source code.
Visible Interactions – Calls generete_custom_discription_without.
Logic Flow –

1. Store discription.
2. generate(info, model) →
   • Invoke generete_custom_discription_without(model, self.discription, info.get("language")).
   • Return the result.

DocFactory Orchestrator

Responsibility – Executes a sequence of BaseModule instances, aggregates their outputs, and logs progress.
Visible Interactions – Receives a list of module instances, a Model, a BaseProgress reporter, and uses BaseLogger.
Logic Flow –

1. Store modules tuple as list; instantiate BaseLogger.
2. generate_doc(info, model, progress)
   • Initialise output = "".
   • progress.create_new_subtask("Generate parts", len(self.modules)).
   • Iterate module in self.modules:
     • Call module.generate(info, model) → module_result.
     • Append module_result + "\n\n" to output.
     • Log informational messages with InfoLog.
     • progress.update_task().
   • End loop → progress.remove_subtask().
   • Return aggregated output.

Data Contract

Entity	Type	Role	Notes
info	dict	Input data (code snippets, language, etc.)	Keys accessed via info.get(...) in modules.
model	Model	LLM interface used by modules	Passed unchanged to post‑processor calls.
progress	BaseProgress	Progress tracking object	Must implement create_new_subtask, update_task, remove_subtask.
module_result	str	Partial documentation fragment	Concatenated into final output.
self.logger	BaseLogger	Logging sink	Emits InfoLog entries.

Note: DocFactory does not perform any validation of info keys; missing keys may cause None to be passed downstream.

ParentModel – Shared configuration & model rotation

Entity	Type	Role	Notes
api_key	str	API credential	Defaults to API_KEY from autodocgenerator.engine.config.config
history	History	Message buffer	Injected or created on‑demand
use_random	bool	Randomise model order	True enables random.shuffle
regen_models_name	list[str]	Rotation list	Copied from MODELS_NAME then shuffled if use_random
current_model_index	int	Cursor in regen_models_name	Starts at 0

Assumption – MODELS_NAME and API_KEY are defined in the imported config module.

The class stores credentials, a shared History, and prepares a shuffled list of model identifiers for fail‑over usage.

---

GPTModel – Synchronous Groq client wrapper

Entity	Type	Role	Notes
client	Groq	Sends sync chat requests	Instantiated with self.api_key
logger	BaseLogger	Emits runtime logs	Same log types as async variant
generate_answer	def	Returns LLM reply	Same signature as async version

Logic Flow mirrors AsyncGPTModel but uses the synchronous self.client.chat.completions.create call and returns the extracted result directly.

Both wrappers expose the same public contract, differing only in async vs sync execution.

AsyncGPTModel – Asynchronous Groq client wrapper

Entity	Type	Role	Notes
client	AsyncGroq	Sends async chat requests	Instantiated with self.api_key
logger	BaseLogger	Emits runtime logs	Uses InfoLog, WarningLog, ErrorLog
generate_answer	async def	Returns LLM reply	Parameters: with_history: bool = True, prompt: str = None

Logic Flow

1. Log start of generation.
2. Select messages from self.history.history or the supplied prompt.
3. Loop until a model succeeds:
   • If regen_models_name empty → log error & raise ModelExhaustedException.
   • Pick model_name from regen_models_name[current_model_index].
   • Call await self.client.chat.completions.create(messages=messages, model=model_name).
   • On exception → log warning, advance current_model_index (wrap‑around).
4. Extract result = chat_completion.choices[0].message.content.
5. Log the selected model and the answer (level 2).
6. Return result.

Interactions – Relies on AsyncGroq for network I/O, History for context, and the shared logger. No file I/O.

---

Semantic Title Ordering (LLM Interaction)

Entity	Type	Role	Notes
model	Model	Input	Must expose get_answer_without_history.
chanks	dict[str,str]	Input	Mapping from anchor links to text chunks.
logger	BaseLogger	Side‑effect	Emits InfoLog entries at various levels.
prompt	list[dict]	Local	Single user‑role message requesting a comma‑separated ordering.
result	str	Output (raw LLM reply)	Expected list of #anchor titles.
new_result	list[str]	Local	Stripped titles from result.
order_output	str	Return	Concatenated chunk texts in LLM‑provided order.

Logic Flow

1. Initialise BaseLogger; log start and input keys/values.
2. Build prompt containing the titles (list(chanks.keys())).
3. Call model.get_answer_without_history(prompt); capture raw string.
4. Split on commas, strip whitespace → new_result.
5. Iterate new_result; for each anchor retrieve its chunk via chanks.get(el) and append to order_output (newline‑separated).
6. Log each addition; return the assembled order_output.

Warning – If the model returns a malformed list (missing titles or extra commas), mismatches may lead to None values in the final output.

HTML Link Extraction (get_all_html_links)

Entity	Type	Role	Notes
data	str	Input	Raw documentation text.
links	list[str]	Output	Anchor links prefixed with #.
logger	BaseLogger	Side‑effect	Logs progress at INFO level.
pattern	str	Internal	Regex \<a name=["']?(.*?)["']?\>\</a\> .

The function scans data for <a name=…></a> anchors, captures the name when longer than five characters, prefixes it with #, logs the count and the list, then returns links. No file I/O occurs.

Anchor Extraction & Chunk Splitting

Entity	Type	Role	Notes
chunks	list[str]	Input (raw HTML fragments)	Passed to extract_links_from_start.
pattern (in extract_links_from_start)	str	Constant	r'^<a name=["\']?(.*?)["\']?></a>' matches a leading anchor.
anchor_name	str	Local	Captured name; processed only if len > 5.
links	list[str]	Output (prefixed anchors)	Each stored as # + name.
text	str	Input (full doc)	Split by split_text_by_anchors.
result_chanks	list[str]	Output	Trimmed, non‑empty chunks after re.split.
all_links	list[str]	Output	Result of extract_links_from_start on the chunks.
result	dict[str,str]	Output	Mapping #anchor → chunk. Returns None on size mismatch.

Logic Flow

1. extract_links_from_start iterates chunks; regex anchors at start are captured.
2. If an anchor’s name exceeds five characters, "#"+name is appended to links.
3. split_text_by_anchors builds a look‑ahead pattern ((?=<a name=…>)) to split the full text into chunks.
4. Whitespace‑trimmed chunks become result_chanks.
5. extract_links_from_start processes these chunks; if the count of anchors ≠ chunks, the function aborts (None).
6. Otherwise a dict result maps each #anchor to its associated chunk.

Assumption – No file I/O occurs; the functions only transform in‑memory strings.

gen_doc – End‑to‑End Documentation Generation Pipeline

Entity	Type	Role	Notes
project_path	str	Root of the source project	Supplied by caller
config	Config	Project‑wide settings	From read_config
custom_modules	list[...]	Extension hooks for doc factories	Passed to DocFactory
structure_settings	StructureSettings	Controls ordering & intro links	Determines size & optional steps
sync_model / async_model	GPTModel / AsyncGPTModel	Language model back‑ends	Instantiated with global API_KEY
manager	Manager	Orchestrates parsing, generation, caching	Core engine object
Return value	str	Final assembled documentation	Read from manager’s output cache

Logic Flow

1. Create GPTModel and AsyncGPTModel using the shared API_KEY.
2. Initialise Manager with project_path, config, both models, and a console progress bar.
3. Invoke manager.generate_code_file() → extracts source code.
4. Call manager.generete_doc_parts(...) → splits docs per structure_settings.max_doc_part_size.
5. Run manager.factory_generate_doc(DocFactory(*custom_modules)) → applies user‑defined modules.
6. If include_order is true, manager.order_doc() reorders sections.
7. If include_intro_links is true, generate intro links via IntroLinks() factory.
8. Clear temporary caches with manager.clear_cache().
9. Return the assembled document read by manager.read_file_by_file_key("output_doc").

Interactions – gen_doc ties together configuration (Config), custom processing (CustomModule*), the GPT backend, and the Manager orchestration. No external I/O occurs beyond the initial YAML read (handled elsewhere).

Warning – The function assumes API_KEY is defined in autodocgenerator.engine.config.config; absence raises an import error.

gen_doc_parts – Orchestrated Multi‑Chunk Documentation

Entity	Type	Role	Notes
full_code_mix	str	Input	Complete source to be documented.
max_symbols	int	Input	Symbol limit passed to split_data.
model	Model	Input	Same contract as above.
project_settings	ProjectSettings	Input	Supplies global prompts.
language	str	Input	Target language for docs.
progress_bar	BaseProgress	Side‑effect	Tracks per‑chunk progress.
global_info	str / None	Optional	Extra project context.
all_result	str	Output	Concatenated documentation.
logger	BaseLogger	Side‑effect	Logs orchestration milestones.

Logic Flow

1. Split source via split_data.
2. Create a sub‑task on progress_bar for the number of chunks.
3. For each chunk el:
   • Call write_docs_by_parts passing the previous result as prev_info.
   • Append returned text and two newlines to all_result.
   • Retain only the last 3000 characters of result for the next iteration (context window).
   • Update progress bar.
4. Remove sub‑task, log final length, and return all_result.

write_docs_by_parts – Per‑Chunk Documentation Generation

Entity	Type	Role	Notes
part	str	Input	Text fragment to document.
model	Model	Input	Must implement get_answer_without_history(prompt).
project_settings	ProjectSettings	Input	Provides prompt for system context.
prev_info	str / None	Optional	Previously generated doc fragment.
language	str	Input	Language code (default "en").
global_info	str / None	Optional	Additional project‑wide context.
answer	str	Output	Raw LLM response, possibly wrapped in markdown fences.
logger	BaseLogger	Side‑effect	Logs generation steps.

Logic Flow

1. Initialise logger and base system prompts (language, global project info, static BASE_PART_COMPLITE_TEXT).
2. Append optional global_info and prev_info prompts.
3. Append user prompt containing part.
4. Call model.get_answer_without_history(prompt=prompt).
5. Strip leading/trailing ````` fences if present and return cleaned text.

Warning – The function assumes the LLM returns a string; non‑string answers would raise a type error.

Template‑Based Custom Description Generation (generete_custom_discription_without)

Entity	Type	Role	Notes
model	Model	Input	LLM interface.
custom_description	str	Input	Desired description task.
language	str	Input (default "en")	Language hint.
result	str	Output	LLM answer respecting strict tag rules.

Constructs a prompt that enforces a single opening `

Iterative Custom Description Generation (generete_custom_discription)

Entity	Type	Role	Notes
splited_data	str (iterable)	Input	Segments of documentation.
model	Model	Input	LLM interface.
custom_description	str	Input	Task description for the LLM.
language	str	Input (default "en")	Language hint.
result	str	Output	First non‑error description.

Iterates over splited_data, sending a prompt that includes the segment, BASE_CUSTOM_DISCRIPTIONS, and the task. Breaks when the LLM response does not contain "!noinfo" or "No information found" early in the text; otherwise continues. Returns the final result.

Repository Structure Packing (CodeMix)

Entity	Type	Role	Notes
root_dir	Path	Input	Base directory for scanning.
ignore_patterns	list[str]	Input	Glob patterns to exclude files/dirs.
logger	BaseLogger	Side‑effect	Logs ignored paths.
should_ignore	method	Local	Returns True if a path matches any ignore pattern.
output_file	str	Input	Destination text file for the mix.
out	file handle	Side‑effect	Receives repository tree and file contents.

Logic Flow

1. Resolve root_dir; instantiate BaseLogger.
2. should_ignore converts a Path to a relative string, then checks against each glob pattern (full path, basename, and any path part).
3. build_repo_content opens output_file for writing.
4. First pass: iterate sorted rglob("*"); for each non‑ignored path write an indented tree line (dir/ or file).
5. Write a separator line ("="*20).
6. Second pass: iterate again; for each non‑ignored file, write a <file path="…"> tag, then the file’s text (read with UTF‑8, errors ignored). Errors are captured and written as plain text.

Critical – No network or external service calls; all operations are local filesystem reads/writes.

compress – Core LLM‑driven Compression

Entity	Type	Role	Notes
data	str	Input	Raw source text to be compressed.
project_settings	ProjectSettings	Input	Supplies system prompt via its prompt property.
model	Model	Input	Must implement get_answer_without_history(prompt).
compress_power	int	Input	Determines compression intensity; passed to helper text.
prompt	list[dict]	Local	System‑user message stack fed to the LLM.
answer	str	Output	LLM’s compressed response.

Logic Flow

1. Assemble a three‑message prompt: project‑level system prompt, a dynamically built “base compress” text (get_BASE_COMPRESS_TEXT), and the user‑provided data.
2. Invoke model.get_answer_without_history(prompt=prompt).
3. Return the LLM’s reply unchanged.

Assumption – No file I/O; transformation occurs entirely in‑memory.

compress_and_compare – Batch Compression with Progress

Entity	Type	Role	Notes
data	list[str]	Input	Collection of text fragments.
model	Model	Input	Same requirements as above.
project_settings	ProjectSettings	Input	Shared prompt source.
compress_power	int	Input (default 4)	Max fragments per batch.
progress_bar	BaseProgress	Side‑effect	Visual sub‑task tracking.
compress_and_compare_data	list[str]	Local	Holds concatenated compressed batches.

Logic Flow

1. Initialise compress_and_compare_data sized to ceil(len(data)/compress_power).
2. Create a sub‑task on progress_bar for the full length of data.
3. Iterate data with index i; compute batch index curr_index = i // compress_power.
4. Call compress on each element, appending the result plus a newline to the appropriate batch slot.
5. Update the progress bar after each element; finally remove the sub‑task.
6. Return the list of batched compressed strings.

compress_to_one – Recursive Full‑Document Collapse

Entity	Type	Role	Notes
data	list[str]	Input/Output	Shrinks each iteration until a single string remains.
model	Model	Input	Same LLM contract.
project_settings	ProjectSettings	Input	Provides prompt context.
compress_power	int	Input (default 4)	Base batch size; may be reduced to 2 for small tails.
progress_bar	BaseProgress	Side‑effect	Propagated to inner calls.
count_of_iter	int	Local	Iteration counter (for potential logging).

Logic Flow

1. Loop while len(data) > 1.
2. Adjust new_compress_power: if remaining items are fewer than compress_power + 1, set to 2 to avoid a final singleton batch.
3. Replace data with the result of compress_and_compare(data, …, new_compress_power, progress_bar).
4. Increment count_of_iter.
5. Once reduced to a single element, return data[0].

Warning – The function assumes compress_and_compare always returns a non‑empty list; an empty return would raise an IndexError.

split_data – Symbol‑Based Chunking Loop

Entity	Type	Role	Notes
splited_by_files	list[str]	Local	Initial list of file‑level fragments (source not shown).
max_symbols	int	Input	Upper bound for symbols per chunk.
split_objects	list[str]	Local	Accumulates final chunks respecting max_symbols.
logger	BaseLogger	Side‑effect	Emits progress messages.

Logic Flow

1. Log start.
2. Refinement loop – while any element exceeds 1.5 × max_symbols, split it at max_symbols/2 and insert the second half right after the original.
3. Packing loop – iterate splited_by_files; create a new split_objects entry when adding the element would breach 1.25 × max_symbols. Otherwise, concatenate with a newline.
4. Log final chunk count and return split_objects.

Assumption – splited_by_files is pre‑populated; no I/O occurs here.