import secrets import subprocess from collections.abc import Iterable from functools import lru_cache from os import sep as pathsep from pathlib import Path from typing import Annotated import git import typer import yaml from github import Github from pydantic_ai import Agent from rich import print non_translated_sections = ( f"reference{pathsep}", "release-notes.md", "fastapi-people.md", "external-links.md", "newsletter.md", "management-tasks.md", "management.md", "contributing.md", ) general_prompt = """ ### About literal text in this prompt 1) In the following instructions (after I say: `The above rules are in effect now`) the two characters `«` and `»` will be used to surround LITERAL TEXT, which is text or characters you shall interpret literally. The `«` and the `»` are not part of the literal text, they are the meta characters denoting it. 2) Furthermore, text surrounded by `«««` and `»»»` is a BLOCK OF LITERAL TEXT which spans multiple lines. To get its content, dedent all lines of the block until the `«««` and `»»»` are at column zero, then remove the newline (`\n`) after the `«««` and the newline before the `»»»`. The `«««` and the `»»»` are not part of the literal text block, they are the meta characters denoting it. 3) If you see backticks or any other quotes inside literal text – inside `«` and `»` – or inside blocks of literal text – inside `«««` and `»»»` – then interpret them as literal characters, do NOT interpret them as meta characters. The above rules are in effect now. ### Definitions of terms used in this prompt "backtick" The character «`» Unicode U+0060 (GRAVE ACCENT) "single backtick" A single backtick – «`» "triple backticks" Three backticks in a row – «```» "neutral double quote" The character «"» Unicode U+0022 (QUOTATION MARK) "neutral single quote" The character «'» Unicode U+0027 (APOSTROPHE) "English double typographic quotes" The characters «“» and «”» Unicode U+201C (LEFT DOUBLE QUOTATION MARK) and Unicode U+201D (RIGHT DOUBLE QUOTATION MARK) "English single typographic quotes" The characters «‘» and «’» Unicode U+2018 (LEFT SINGLE QUOTATION MARK) and Unicode U+2019 (RIGHT SINGLE QUOTATION MARK) "code snippet" Also called "inline code". Text in a Markdown document which is surrounded by single backticks. A paragraph in a Markdown document can have a more than one code snippet. Example: ««« `i am a code snippet` »»» Example: ««« `first code snippet` `second code snippet` `third code snippet` »»» "code block" Text in a Markdown document which is surrounded by triple backticks. Spreads multiple lines. Example: ««« ``` Hello World ``` »»» Example: ««« ```python print("hello World") ``` »»» "HTML element" a HTML opening tag – e.g. «

» – and a HTML closing tag – e.g. «

» – surrounding text or other HTML elements. ### Your task Translate an English text – the original content – to a target language. The original content is written in Markdown, write the translation in Markdown as well. The original content will be surrounded by triple percentage signs («%%%»). Do not include the triple percentage signs in the translation. ### Technical terms in English For technical terms in English that don't have a common translation term, use the original term in English. ### Content of code snippets Do not translate the content of code snippets, keep the original in English. For example, «`list`», «`dict`», keep them as is. ### Content of code blocks Do not translate the content of code blocks, except for comments in the language which the code block uses. Examples: Source (English) – The code block is a bash code example with one comment: ««« ```bash # Print greeting echo "Hello, World!" ``` »»» Result (German): ««« ```bash # Gruß ausgeben echo "Hello, World!" ``` »»» Source (English) – The code block is a console example containing HTML tags. No comments, so nothing to change here: ««« ```console $ fastapi run main.py FastAPI Starting server Searching for package file structure ``` »»» Result (German): ««« ```console $ fastapi run main.py FastAPI Starting server Searching for package file structure ``` »»» Source (English) – The code block is a console example containing 5 comments: ««« ```console // Go to the home directory $ cd // Create a directory for all your code projects $ mkdir code // Enter into that code directory $ cd code // Create a directory for this project $ mkdir awesome-project // Enter into that project directory $ cd awesome-project ``` »»» Result (German): ««« ```console // Gehe zum Home-Verzeichnis $ cd // Erstelle ein Verzeichnis für alle Ihre Code-Projekte $ mkdir code // Gehe in dieses Code-Verzeichnis $ cd code // Erstelle ein Verzeichnis für dieses Projekt $ mkdir awesome-project // Gehe in dieses Projektverzeichnis $ cd awesome-project ``` »»» If there is an existing translation and its Mermaid diagram is in sync with the Mermaid diagram in the English source, except a few translated words, then use the Mermaid diagram of the existing translation. The human editor of the translation translated these words in the Mermaid diagram. Keep these translations, do not revert them back to the English source. Example: Source (English): ««« ```mermaid flowchart LR subgraph global[global env] harry-1[harry v1] end subgraph stone-project[philosophers-stone project] stone(philosophers-stone) -->|requires| harry-1 end ``` »»» Existing translation (German) – has three translations: ««« ```mermaid flowchart LR subgraph global[globale Umgebung] harry-1[harry v1] end subgraph stone-project[philosophers-stone-Projekt] stone(philosophers-stone) -->|benötigt| harry-1 end ``` »»» Result (German) – you change nothing: ««« ```mermaid flowchart LR subgraph global[globale Umgebung] harry-1[harry v1] end subgraph stone-project[philosophers-stone-Projekt] stone(philosophers-stone) -->|benötigt| harry-1 end ``` »»» ### Special blocks There are special blocks of notes, tips and others that look like: ««« /// note »»» To translate it, keep the same line and add the translation after a vertical bar. For example, if you were translating to Spanish, you would write: ««« /// note | Nota »»» Some examples in Spanish: Source: ««« /// tip »»» Result: ««« /// tip | Consejo »»» Source: ««« /// details | Preview »»» Result: ««« /// details | Vista previa »»» ### Tab blocks There are special blocks surrounded by four slashes («////»). They mark text, which will be rendered as part of a tab in the final document. The scheme is: //// tab | {tab title} {tab content, may span many lines} //// Keep everything before the vertical bar («|») as is, including the vertical bar. Translate the tab title. Translate the tab content, applying the rules you know. Keep the four block closing slashes as is. Examples: Source (English): ««« //// tab | Python 3.8+ non-Annotated Hello //// »»» Result (German): ««« //// tab | Python 3.8+ nicht annotiert Hallo //// »»» Source (English) – Here there is nothing to translate in the tab title: ««« //// tab | Linux, macOS, Windows Bash Hello again //// »»» Result (German): ««« //// tab | Linux, macOS, Windows Bash Hallo wieder //// »»» ### Headings Every Markdown heading in the English text (all levels) ends with a part inside curly brackets. This part denotes the hash of this heading, which is used in links to this heading. In translations, translate the heading, but do not translate this hash part, so that links do not break. Examples of how to translate a heading: Source (English): ««« ## Alternative API docs { #alternative-api-docs } »»» Result (Spanish): ««« ## Documentación de la API alternativa { #alternative-api-docs } »»» Source (English): ««« ### Example { #example } »»» Result (German): ««« ### Beispiel { #example } »»» ### Links Use the following rules for links (apply both to Markdown-style links ([text](url)) and to HTML-style tags): 1) For relative URLs, only translate link text. Do not translate the URL or its parts Example: Source (English): ««« [One of the fastest Python frameworks available](#performance) »»» Result (German): ««« [Eines der schnellsten verfügbaren Python-Frameworks](#performance) »»» 2) For absolute URLs which DO NOT start EXACTLY with «https://fastapi.tiangolo.com», only translate link text and leave the URL unchanged. Example: Source (English): ««« SQLModel docs »»» Result (German): ««« SQLModel-Dokumentation »»» 3) For absolute URLs which DO start EXACTLY with «https://fastapi.tiangolo.com», only translate link text and change the URL by adding language code («https://fastapi.tiangolo.com/{language_code}[rest part of the url]»). Example: Source (English): ««« Documentation »»» Result (Spanish): ««« Documentación »»» 3.1) Do not add language codes for URLs that point to static assets (e.g., images, CSS, JavaScript). Example: Source (English): ««« Something »»» Result (Spanish): ««« Algo »»» 4) For internal links, only translate link text. Example: Source (English): ««« [Create Pull Requests](help-fastapi.md#create-a-pull-request){.internal-link target=_blank} »»» Result (German): ««« [Pull Requests erzeugen](help-fastapi.md#create-a-pull-request){.internal-link target=_blank} »»» 5) Do not translate anchor fragments in links (the part after «#»), as they must remain the same to work correctly. 5.1) If an existing translation has a link with an anchor fragment different to the anchor fragment in the English source, then this is an error. Fix this by using the anchor fragment of the English source. Example: Source (English): ««« [Body - Multiple Parameters: Singular values in body](body-multiple-params.md#singular-values-in-body){.internal-link target=_blank} »»» Existing wrong translation (German) – notice the wrongly translated anchor fragment: ««« [Body – Mehrere Parameter: Einfache Werte im Body](body-multiple-params.md#einzelne-werte-im-body){.internal-link target=_blank}. »»» Result (German) – you fix the anchor fragment: ««« [Body – Mehrere Parameter: Einfache Werte im Body](body-multiple-params.md#singular-values-in-body){.internal-link target=_blank}. »»» 5.2) Do not add anchor fragments at will, even if this makes sense. If the English source has no anchor, don't add one. Example: Source (English): ««« Create a [virtual environment](../virtual-environments.md){.internal-link target=_blank} »»» Wrong translation (German) – Anchor added to the URL. ««« Erstelle eine [virtuelle Umgebung](../virtual-environments.md#create-a-virtual-environment){.internal-link target=_blank} »»» Good translation (German) – URL stays like in the English source. ««« Erstelle eine [Virtuelle Umgebung](../virtual-environments.md){.internal-link target=_blank} »»» ### HTML abbr elements Translate HTML abbr elements («text») as follows: 1) If the text surrounded by the abbr element is an abbreviation (the text may be surrounded by further HTML or Markdown markup or quotes, for example «text» or «`text`» or «"text"», ignore that further markup when deciding if the text is an abbreviation), and if the description (the text inside the title attribute) contains the full phrase for this abbreviation, then append a dash («–») to the full phrase, followed by the translation of the full phrase. Conversion scheme: Source (English): {abbreviation} Result: {abbreviation} Examples: Source (English): ««« IoT CPU TL;DR: »»» Result (German): ««« IoT CPU TL;DR: »»» 1.1) If the language to which you translate mostly uses the letters of the ASCII char set (for example Spanish, French, German, but not Russian, Chinese) and if the translation of the full phrase is identical to, or starts with the same letters as the original full phrase, then only give the translation of the full phrase. Conversion scheme: Source (English): {abbreviation} Result: {abbreviation} Examples: Source (English): ««« JWT Enum ASGI »»» Result (German): ««« JWT Enum ASGI »»» 2) If the description is not a full phrase for an abbreviation which the abbr element surrounds, but some other information, then just translate the description. Conversion scheme: Source (English): {text} Result: {translation of text} Examples: Source (English): ««« path linter parsing 0.95.0 at the time of writing this »»» Result (German): ««« Pfad Linter Parsen 0.95.0 zum Zeitpunkt als das hier geschrieben wurde »»» 3) If the text surrounded by the abbr element is an abbreviation and the description contains both the full phrase for that abbreviation, and other information, separated by a colon («:»), then append a dash («–») and the translation of the full phrase to the original full phrase and translate the other information. Conversion scheme: Source (English): {abbreviation} Result: {abbreviation} Examples: Source (English): ««« I/O CDN IDE »»» Result (German): ««« I/O CDN IDE »»» 3.1) Like in rule 2.1, you can leave the original full phrase away, if the translated full phrase is identical or starts with the same letters as the original full phrase. Conversion scheme: Source (English): {abbreviation} Result: {abbreviation} Example: Source (English): ««« ORM »»» Result (German): ««« ORM »»» 4) If there is an existing translation, and it has ADDITIONAL abbr elements in a sentence, and these additional abbr elements do not exist in the related sentence in the English text, then KEEP those additional abbr elements in the translation. Do not remove them. Except when you remove the whole sentence from the translation, because the whole sentence was removed from the English text, then also remove the abbr element. The reasoning for this rule is, that such additional abbr elements are manually added by the human editor of the translation, in order to translate or explain an English word to the human readers of the translation. These additional abbr elements would not make sense in the English text, but they do make sense in the translation. So keep them in the translation, even though they are not part of the English text. This rule only applies to abbr elements. 5) Apply above rules also when there is an existing translation! Make sure that all title attributes in abbr elements get properly translated or updated, using the schemes given above. However, leave the ADDITIONAL abbr's from rule 4 alone. Do not change their formatting or content. """ app = typer.Typer() @lru_cache def get_langs() -> dict[str, str]: return yaml.safe_load(Path("docs/language_names.yml").read_text(encoding="utf-8")) def generate_lang_path(*, lang: str, path: Path) -> Path: en_docs_path = Path("docs/en/docs") assert str(path).startswith(str(en_docs_path)), ( f"Path must be inside {en_docs_path}" ) lang_docs_path = Path(f"docs/{lang}/docs") out_path = Path(str(path).replace(str(en_docs_path), str(lang_docs_path))) return out_path def generate_en_path(*, lang: str, path: Path) -> Path: en_docs_path = Path("docs/en/docs") assert not str(path).startswith(str(en_docs_path)), ( f"Path must not be inside {en_docs_path}" ) lang_docs_path = Path(f"docs/{lang}/docs") out_path = Path(str(path).replace(str(lang_docs_path), str(en_docs_path))) return out_path @app.command() def translate_page( *, language: Annotated[str, typer.Option(envvar="LANGUAGE")], en_path: Annotated[Path, typer.Option(envvar="EN_PATH")], ) -> None: assert language != "en", ( "`en` is the source language, choose another language as translation target" ) langs = get_langs() language_name = langs[language] lang_path = Path(f"docs/{language}") lang_path.mkdir(exist_ok=True) lang_prompt_path = lang_path / "llm-prompt.md" assert lang_prompt_path.exists(), f"Prompt file not found: {lang_prompt_path}" lang_prompt_content = lang_prompt_path.read_text(encoding="utf-8") en_docs_path = Path("docs/en/docs") assert str(en_path).startswith(str(en_docs_path)), ( f"Path must be inside {en_docs_path}" ) out_path = generate_lang_path(lang=language, path=en_path) out_path.parent.mkdir(parents=True, exist_ok=True) original_content = en_path.read_text(encoding="utf-8") old_translation: str | None = None if out_path.exists(): print(f"Found existing translation: {out_path}") old_translation = out_path.read_text(encoding="utf-8") print(f"Translating {en_path} to {language} ({language_name})") agent = Agent("openai:gpt-5") prompt_segments = [ general_prompt, lang_prompt_content, ] if old_translation: prompt_segments.extend( [ "There is an existing previous translation for the original English content, that may be outdated.", "Update the translation only where necessary:", "- If the original English content has added parts, also add these parts to the translation.", "- If the original English content has removed parts, also remove them from the translation, unless you were instructed earlier to not do that in specific cases.", "- If parts of the original English content have changed, also change those parts in the translation.", "- If the previous translation violates current instructions, update it.", "- Otherwise, preserve the original translation LINE-BY-LINE, AS-IS.", "Do not:", "- rephrase or rewrite correct lines just to improve the style.", "- add or remove line breaks, unless the original English content changed.", "- change formatting or whitespace unless absolutely required.", "Only change what must be changed. The goal is to minimize diffs for easier human review.", "UNLESS you were instructed earlier to behave different, there MUST NOT be whole sentences or partial sentences in the updated translation, which are not in the original English content, and there MUST NOT be whole sentences or partial sentences in the original English content, which are not in the updated translation. Remember: the updated translation shall be IN SYNC with the original English content.", "Previous translation:", f"%%%\n{old_translation}%%%", ] ) prompt_segments.extend( [ f"Translate to {language} ({language_name}).", "Original content:", f"%%%\n{original_content}%%%", ] ) prompt = "\n\n".join(prompt_segments) print(f"Running agent for {out_path}") result = agent.run_sync(prompt) out_content = f"{result.output.strip()}\n" print(f"Saving translation to {out_path}") out_path.write_text(out_content, encoding="utf-8", newline="\n") def iter_all_en_paths() -> Iterable[Path]: """ Iterate on the markdown files to translate in order of priority. """ first_dirs = [ Path("docs/en/docs/learn"), Path("docs/en/docs/tutorial"), Path("docs/en/docs/advanced"), Path("docs/en/docs/about"), Path("docs/en/docs/how-to"), ] first_parent = Path("docs/en/docs") yield from first_parent.glob("*.md") for dir_path in first_dirs: yield from dir_path.rglob("*.md") first_dirs_str = tuple(str(d) for d in first_dirs) for path in Path("docs/en/docs").rglob("*.md"): if str(path).startswith(first_dirs_str): continue if path.parent == first_parent: continue yield path def iter_en_paths_to_translate() -> Iterable[Path]: en_docs_root = Path("docs/en/docs/") for path in iter_all_en_paths(): relpath = path.relative_to(en_docs_root) if not str(relpath).startswith(non_translated_sections): yield path @app.command() def translate_lang(language: Annotated[str, typer.Option(envvar="LANGUAGE")]) -> None: paths_to_process = list(iter_en_paths_to_translate()) print("Original paths:") for p in paths_to_process: print(f" - {p}") print(f"Total original paths: {len(paths_to_process)}") missing_paths: list[Path] = [] skipped_paths: list[Path] = [] for p in paths_to_process: lang_path = generate_lang_path(lang=language, path=p) if lang_path.exists(): skipped_paths.append(p) continue missing_paths.append(p) print("Paths to skip:") for p in skipped_paths: print(f" - {p}") print(f"Total paths to skip: {len(skipped_paths)}") print("Paths to process:") for p in missing_paths: print(f" - {p}") print(f"Total paths to process: {len(missing_paths)}") for p in missing_paths: print(f"Translating: {p}") translate_page(language="es", en_path=p) print(f"Done translating: {p}") @app.command() def list_removable(language: str) -> list[Path]: removable_paths: list[Path] = [] lang_paths = Path(f"docs/{language}").rglob("*.md") for path in lang_paths: en_path = generate_en_path(lang=language, path=path) if not en_path.exists(): removable_paths.append(path) print(removable_paths) return removable_paths @app.command() def list_all_removable() -> list[Path]: all_removable_paths: list[Path] = [] langs = get_langs() for lang in langs: if lang == "en": continue removable_paths = list_removable(lang) all_removable_paths.extend(removable_paths) print(all_removable_paths) return all_removable_paths @app.command() def remove_removable(language: str) -> None: removable_paths = list_removable(language) for path in removable_paths: path.unlink() print(f"Removed: {path}") print("Done removing all removable paths") @app.command() def remove_all_removable() -> None: all_removable = list_all_removable() for removable_path in all_removable: removable_path.unlink() print(f"Removed: {removable_path}") print("Done removing all removable paths") @app.command() def list_missing(language: str) -> list[Path]: missing_paths: list[Path] = [] en_lang_paths = list(iter_en_paths_to_translate()) for path in en_lang_paths: lang_path = generate_lang_path(lang=language, path=path) if not lang_path.exists(): missing_paths.append(path) print(missing_paths) return missing_paths @app.command() def list_outdated(language: str) -> list[Path]: dir_path = Path(__file__).absolute().parent.parent repo = git.Repo(dir_path) outdated_paths: list[Path] = [] en_lang_paths = list(iter_en_paths_to_translate()) for path in en_lang_paths: lang_path = generate_lang_path(lang=language, path=path) if not lang_path.exists(): continue en_commit_datetime = list(repo.iter_commits(paths=path, max_count=1))[ 0 ].committed_datetime lang_commit_datetime = list(repo.iter_commits(paths=lang_path, max_count=1))[ 0 ].committed_datetime if lang_commit_datetime < en_commit_datetime: outdated_paths.append(path) print(outdated_paths) return outdated_paths @app.command() def update_outdated(language: Annotated[str, typer.Option(envvar="LANGUAGE")]) -> None: outdated_paths = list_outdated(language) for path in outdated_paths: print(f"Updating lang: {language} path: {path}") translate_page(language=language, en_path=path) print(f"Done updating: {path}") print("Done updating all outdated paths") @app.command() def add_missing(language: Annotated[str, typer.Option(envvar="LANGUAGE")]) -> None: missing_paths = list_missing(language) for path in missing_paths: print(f"Adding lang: {language} path: {path}") translate_page(language=language, en_path=path) print(f"Done adding: {path}") print("Done adding all missing paths") @app.command() def update_and_add(language: Annotated[str, typer.Option(envvar="LANGUAGE")]) -> None: print(f"Updating outdated translations for {language}") update_outdated(language=language) print(f"Adding missing translations for {language}") add_missing(language=language) print(f"Done updating and adding for {language}") @app.command() def make_pr( *, language: Annotated[str | None, typer.Option(envvar="LANGUAGE")] = None, github_token: Annotated[str, typer.Option(envvar="GITHUB_TOKEN")], github_repository: Annotated[str, typer.Option(envvar="GITHUB_REPOSITORY")], ) -> None: print("Setting up GitHub Actions git user") repo = git.Repo(Path(__file__).absolute().parent.parent) if not repo.is_dirty(untracked_files=True): print("Repository is clean, no changes to commit") return subprocess.run(["git", "config", "user.name", "github-actions"], check=True) subprocess.run( ["git", "config", "user.email", "github-actions@github.com"], check=True ) branch_name = "translate" if language: branch_name += f"-{language}" branch_name += f"-{secrets.token_hex(4)}" print(f"Creating a new branch {branch_name}") subprocess.run(["git", "checkout", "-b", branch_name], check=True) print("Adding updated files") git_path = Path("docs") subprocess.run(["git", "add", str(git_path)], check=True) print("Committing updated file") message = "🌐 Update translations" if language: message += f" for {language}" subprocess.run(["git", "commit", "-m", message], check=True) print("Pushing branch") subprocess.run(["git", "push", "origin", branch_name], check=True) print("Creating PR") g = Github(github_token) gh_repo = g.get_repo(github_repository) pr = gh_repo.create_pull( title=message, body=message, base="master", head=branch_name ) print(f"Created PR: {pr.number}") print("Finished") if __name__ == "__main__": app()