import inspect import os import re import traceback from copy import deepcopy from dataclasses import dataclass from enum import Enum from functools import partial from typing import Any, Optional, Union import PIL from bs4 import BeautifulSoup from mistune import HTMLRenderer, create_markdown from pptx.enum.text import PP_ALIGN from pptx.oxml import parse_xml from pptx.shapes.base import BaseShape from pptx.shapes.graphfrm import GraphicFrame as PPTXGraphicFrame from pptx.text.text import _Run from pptx.util import Pt from pptagent.document import Document from pptagent.presentation import Closure, ClosureType, Picture, ShapeElement, SlidePage from pptagent.utils import get_logger, runs_merge logger = get_logger(__name__) TABLE_REGEX = re.compile(r".*table_[0-9a-fA-F]{4}\.png$") class SlideRenderer(HTMLRenderer): """ A renderer that does not render lists. """ def list(self, text: str, ordered: bool, **attrs: Any) -> str: return text def list_item(self, text: str) -> str: return text markdown = create_markdown(renderer=SlideRenderer(), plugins=["strikethrough"]) class SlideEditError(Exception): """ Exception raised when an edit operation fails. """ @dataclass class HistoryMark: """ Mark the execution status of the API call, comment and a line of code. """ API_CALL_ERROR = "api_call_error" API_CALL_CORRECT = "api_call_correct" COMMENT_CORRECT = "comment_correct" COMMENT_ERROR = "comment_error" CODE_RUN_ERROR = "code_run_error" CODE_RUN_CORRECT = "code_run_correct" class CodeExecutor: """ Execute code actions and manage API call history, and providing error feedback. """ def __init__(self, retry_times: int): """ Initialize the CodeExecutor. Args: retry_times (int): The number of times to retry failed actions. """ self.api_history = [] self.command_history = [] self.code_history = [] self.retry_times = retry_times self.registered_functions = API_TYPES.all_funcs() self.function_regex = re.compile(r"^[a-z]+_[a-z_]+\(.+\)") @classmethod def get_apis_docs( cls, funcs: list[callable], show_doc: bool = True, show_return: bool = True, ignore_keys: Optional[list[str]] = None, ) -> str: """ Get the documentation for a list of API functions. Args: funcs (list[callable]): A list of functions to document. show_example (bool): Whether to show examples in the documentation. Returns: str: The formatted API documentation. """ if ignore_keys is None: ignore_keys = {"slide", "self", "doc"} api_doc = [] for func in funcs: sig = inspect.signature(func) params = [] for name, param in sig.parameters.items(): if name in ignore_keys: continue param_str = name if param.annotation != inspect.Parameter.empty: param_str += f": {param.annotation.__name__}" if param.default != inspect.Parameter.empty: param_str += f" = {repr(param.default)}" params.append(param_str) signature = f"def {func.__name__}({', '.join(params)})" if show_return and sig.return_annotation != inspect.Parameter.empty: signature += f" -> {sig.return_annotation.__name__}" if show_doc and inspect.getdoc(func) is not None: doc = "\t" + inspect.getdoc(func) else: doc = "" signature += f"\n{doc}" api_doc.append(signature) return "\n".join(api_doc) def execute_actions( self, actions: str, edit_slide: SlidePage, doc: Document, found_code: bool = False, ) -> Union[tuple[str, str], None]: """ Execute a series of actions on a slide. Args: actions (str): The actions to execute. edit_slide (SlidePage): The slide to edit. found_code (bool): Whether code was found in the actions. Returns: tuple: The API lines and traceback if an error occurs. None: If no error occurs. """ api_calls = actions.strip().split("\n") self.api_history.append( [HistoryMark.API_CALL_ERROR, edit_slide.slide_idx, actions] ) for line_idx, line in enumerate(api_calls): try: if line_idx == len(api_calls) - 1 and not found_code: raise SlideEditError( "No code block found in the output, please output the api calls without any prefix." ) if line.startswith("def"): raise SlideEditError("The function definition were not allowed.") if line.startswith("#"): if len(self.command_history) != 0: self.command_history[-1][0] = HistoryMark.COMMENT_CORRECT self.command_history.append([HistoryMark.COMMENT_ERROR, line, None]) continue if not self.function_regex.match(line): continue found_code = True func = line.split("(")[0] if func not in self.registered_functions: raise SlideEditError(f"The function {func} is not defined.") # only one of clone and del can be used in a row if func.startswith("clone") or func.startswith("del"): tag = func.split("_")[0] if ( self.command_history[-1][-1] is None or self.command_history[-1][-1] == tag ): self.command_history[-1][-1] = tag else: raise SlideEditError( "Invalid command: Both 'clone_paragraph' and 'del_paragraph'/'del_image' are used within a single command. " "Each command must only perform one of these operations based on the quantity_change." ) self.code_history.append([HistoryMark.CODE_RUN_ERROR, line, None]) partial_func = partial(self.registered_functions[func], edit_slide) if func == "replace_image": partial_func = partial(partial_func, doc) eval(line, {}, {func: partial_func}) self.code_history[-1][0] = HistoryMark.CODE_RUN_CORRECT except Exception as e: if not isinstance(e, SlideEditError): logger.warning(f"Encountered unknown error: {e}") trace_msg = traceback.format_exc() if len(self.code_history) != 0: self.code_history[-1][-1] = trace_msg api_lines = ( "\n".join(api_calls[: line_idx - 1]) + f"\n--> Error Line: {line}\n" + "\n".join(api_calls[line_idx + 1 :]) ) return api_lines, trace_msg if len(self.command_history) != 0: self.command_history[-1][0] = HistoryMark.COMMENT_CORRECT self.api_history[-1][0] = HistoryMark.API_CALL_CORRECT def __add__(self, other): self.api_history.extend(other.api_history) self.command_history.extend(other.command_history) self.code_history.extend(other.code_history) return self # supporting functions def element_index(slide: SlidePage, element_id: int) -> ShapeElement: """ Find the an element in a slide. Args: slide (SlidePage): The slide element_id (int): The ID of the element. Returns: ShapeElement: The shape corresponding to the element ID. Raises: SlideEditError: If the element is not found. """ for shape in slide: if shape.shape_idx == element_id: return shape raise SlideEditError( f"Cannot find element {element_id}, is it deleted or not exist?" ) @dataclass class TextBlock: text: str bold: bool = False italic: bool = False code: bool = False strikethrough: bool = False href: str = None def build_run(self, run: _Run): if self.bold: run.font.bold = True if self.italic: run.font.italic = True if self.code: run.font.name = "Consolas" if self.strikethrough: run.font.strikethrough = True if self.href is not None: run.hyperlink.address = self.href run.text = self.text MARKDOWN_STYLES = { "strong": "bold", "em": "italic", "code": "code", "del": "strikethrough", } def process_element(element, styles=None) -> list[TextBlock]: if styles is None: styles = {} result = [] if isinstance(element, str): result.append(TextBlock(element, **styles)) else: if element.name == "a": new_styles = styles.copy() for child in element.children: blocks = process_element(child, new_styles) for block in blocks: block.href = element.get("href") result.extend(blocks) elif MARKDOWN_STYLES.get(element.name): new_styles = styles.copy() new_styles[MARKDOWN_STYLES[element.name]] = True for child in element.children: result.extend(process_element(child, new_styles)) else: for child in element.children: result.extend(process_element(child, styles)) return result def replace_para(paragraph_id: int, new_text: str, shape: BaseShape): """ Replace the text of a paragraph in a shape. """ para = shape.text_frame.paragraphs[paragraph_id] html = markdown(new_text).strip() soup = BeautifulSoup(html, "html.parser") blocks = process_element(soup) empty_run = runs_merge(para) empty_run.text = "" for _ in range(len(blocks) - 1): empty_run._r.addnext(parse_xml(empty_run._r.xml)) for block, run in zip(blocks, para.runs): block.build_run(run) def clone_para(paragraph_id: int, shape: BaseShape): """ Clone a paragraph in a shape. """ para = shape.text_frame.paragraphs[paragraph_id] shape.text_frame.paragraphs[-1]._element.addnext(parse_xml(para._element.xml)) def del_para(paragraph_id: int, shape: BaseShape): """ Delete a paragraph from a shape. """ para = shape.text_frame.paragraphs[paragraph_id] para._element.getparent().remove(para._element) def add_table(table_data: list[list[str]], table: PPTXGraphicFrame): rows = len(table_data) cols = len(table_data[0]) max_lengths = [max(len(row[j]) for row in table_data) for j in range(cols)] total_length = sum(max_lengths) for j in range(cols): col_width = int((max_lengths[j] / total_length) * table.width) table.table.columns[j].width = col_width for i in range(rows): for j in range(cols): table.table.cell(i, j).text = table_data[i][j] def merge_cells(merge_area: list[tuple[int, int, int, int]], table: PPTXGraphicFrame): if merge_area is None or len(merge_area) == 0: return for y1, x1, y2, x2 in merge_area: try: table.table.cell(x1, y1).merge(table.table.cell(x2, y2)) for x, y in zip(range(x1, x2 + 1), range(y1, y2 + 1)): tf = table.table.cell(x, y).text_frame for p in tf.paragraphs: p.alignment = PP_ALIGN.CENTER except Exception as e: logger.warning(f"Failed to merge cells: {e}") # api functions def del_paragraph(slide: SlidePage, div_id: int, paragraph_id: int): """ Delete a paragraph from a slide. Args: slide (SlidePage): The slide containing the paragraph. div_id (int): The ID of the division containing the paragraph. paragraph_id (int): The ID of the paragraph to delete. Raises: SlideEditError: If the paragraph is not found. """ shape = element_index(slide, div_id) if not shape.text_frame.is_textframe: raise SlideEditError( f"The element {shape.shape_idx} of slide {slide.slide_idx} does not have a text frame, please check the element id and type of element." ) for para in shape.text_frame.paragraphs: if para.idx == paragraph_id: shape.text_frame.paragraphs.remove(para) shape._closures[ClosureType.DELETE].append( Closure(partial(del_para, para.real_idx), para.real_idx) ) return else: raise SlideEditError( f"Cannot find the paragraph {paragraph_id} of the element {div_id}," "may refer to a non-existed paragraph or you haven't cloned enough paragraphs beforehand." ) def del_image(slide: SlidePage, figure_id: int): """ Delete an image from a slide. Args: slide (SlidePage): The slide containing the image. figure_id (int): The ID of the image to delete. """ shape = element_index(slide, figure_id) if not isinstance(shape, Picture): raise SlideEditError( f"The element {shape.shape_idx} of slide {slide.slide_idx} is not a Picture." ) slide.shapes.remove(shape) def replace_paragraph(slide: SlidePage, div_id: int, paragraph_id: int, text: str): """ Replace the text of a paragraph in a slide. Args: slide (SlidePage): The slide containing the paragraph. div_id (int): The ID of the division containing the paragraph. paragraph_id (int): The ID of the paragraph to replace. text (str): The new text to replace with. Raises: SlideEditError: If the paragraph is not found. """ shape = element_index(slide, div_id) if not shape.text_frame.is_textframe: raise SlideEditError( f"The element {shape.shape_idx} of slide {slide.slide_idx} does not have a text frame, please check the element id and type of element." ) for para in shape.text_frame.paragraphs: if para.idx == paragraph_id: para.text = text shape._closures[ClosureType.REPLACE].append( Closure( partial(replace_para, para.real_idx, text), para.real_idx, ) ) return else: raise SlideEditError( f"Cannot find the paragraph {paragraph_id} of the element {div_id}," "Please: " "1. check if you refer to a non-existed paragraph." "2. check if you already deleted it." ) def replace_image(slide: SlidePage, doc: Document, img_id: int, image_path: str): """ Replace an image in a slide. Args: slide (SlidePage): The slide containing the image. img_id (int): The ID of the image to replace. image_path (str): The path to the new image. Raises: SlideEditError: If the image path does not exist. """ if not os.path.exists(image_path): raise SlideEditError( f"The image {image_path} does not exist, consider use del_image if image_path in the given command is faked" ) shape = element_index(slide, img_id) if not isinstance(shape, Picture): raise SlideEditError( f"The element {shape.shape_idx} of slide {slide.slide_idx} is not a Picture." ) try: if TABLE_REGEX.match(image_path): return replace_image_with_table(shape, doc, image_path) except Exception as e: logger.warning( f"Failed to replace image with table element: {e}, fallback to use image directly." ) img_size = PIL.Image.open(image_path).size r = min(shape.width / img_size[0], shape.height / img_size[1]) new_width = img_size[0] * r new_height = img_size[1] * r shape.top = Pt(shape.top + (shape.height - new_height) / 2) shape.width = Pt(new_width) shape.height = Pt(new_height) shape.img_path = image_path def clone_paragraph(slide: SlidePage, div_id: int, paragraph_id: int): """ Clone a paragraph in a slide. Args: slide (SlidePage): The slide containing the paragraph. div_id (int): The ID of the division containing the paragraph. paragraph_id (int): The ID of the paragraph to clone. Raises: SlideEditError: If the paragraph is not found. Mention: the cloned paragraph will have a paragraph_id one greater than the current maximum in the parent element. """ shape = element_index(slide, div_id) if not shape.text_frame.is_textframe: raise SlideEditError( f"The element {shape.shape_idx} of slide {slide.slide_idx} does not have a text frame, please check the element id and type of element." ) max_idx = max([para.idx for para in shape.text_frame.paragraphs]) for para in shape.text_frame.paragraphs: if para.idx != paragraph_id: continue shape.text_frame.paragraphs.append(deepcopy(para)) shape.text_frame.paragraphs[-1].idx = max_idx + 1 shape.text_frame.paragraphs[-1].real_idx = len(shape.text_frame.paragraphs) - 1 shape._closures[ClosureType.CLONE].append( Closure( partial(clone_para, para.real_idx), para.real_idx, ) ) return raise SlideEditError( f"Cannot find the paragraph {paragraph_id} of the element {div_id}, may refer to a non-existed paragraph." ) def replace_image_with_table(shape: Picture, doc: Document, image_path: str): table = doc.get_table(image_path) shape.is_table = True shape.grid = (len(table.cells), len(table.cells[0])) shape._closures[ClosureType.REPLACE].append( Closure(partial(add_table, table.cells)) ) shape._closures[ClosureType.MERGE].append( Closure(partial(merge_cells, table.merge_area)) ) return class API_TYPES(Enum): Agent = [ replace_image, del_image, clone_paragraph, replace_paragraph, del_paragraph, ] @classmethod def all_funcs(cls) -> dict[str, callable]: funcs = {} for attr in dir(cls): if attr.startswith("__"): continue funcs |= {func.__name__: func for func in getattr(cls, attr).value} return funcs