paper2beamer

dataflow_agent/agentroles/paper2any_agents/p2b_pagecontent_to_beamer_agent.py

@@ -0,0 +1,117 @@
+"""
+P2bPagecontentToBeamer agent
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+从 paper2page_content 产出的 pagecontent（结构化大纲）生成 LaTeX Beamer 代码。
+输入：pagecontent (list[dict]: title, layout_description, key_points, asset_ref)
+输出：latex_code，写入 state.beamer_code_path。
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Any, Dict, Optional
+
+from dataflow_agent.state import MainState
+from dataflow_agent.toolkits.tool_manager import ToolManager
+from dataflow_agent.logger import get_logger
+from dataflow_agent.agentroles.cores.base_agent import BaseAgent
+from dataflow_agent.agentroles.cores.registry import register
+from dataflow_agent.toolkits.p2vtool.p2v_tool import extract_beamer_code
+
+log = get_logger(__name__)
+
+
+# ----------------------------------------------------------------------
+# Agent Definition
+# ----------------------------------------------------------------------
+@register("p2b_pagecontent_to_beamer")
+class P2bPagecontentToBeamer(BaseAgent):
+    """从 pagecontent（结构化大纲）生成 Beamer LaTeX 代码"""
+
+    @classmethod
+    def create(cls, tool_manager: Optional[ToolManager] = None, **kwargs):
+        return cls(tool_manager=tool_manager, **kwargs)
+
+    @property
+    def role_name(self) -> str:
+        return "p2b_pagecontent_to_beamer"
+
+    @property
+    def system_prompt_template_name(self) -> str:
+        return "system_prompt_for_p2b_pagecontent_to_beamer"
+
+    @property
+    def task_prompt_template_name(self) -> str:
+        return "task_prompt_for_p2b_pagecontent_to_beamer"
+
+    def get_task_prompt_params(self, pre_tool_results: Dict[str, Any]) -> Dict[str, Any]:
+        return {
+            "pagecontent": pre_tool_results.get("pagecontent", "[]"),
+            "output_language": pre_tool_results.get("output_language", "English"),
+            "pdf_images_working_dir": pre_tool_results.get("pdf_images_working_dir", ""),
+        }
+
+    async def execute_pre_tools(self, state: MainState) -> Dict[str, Any]:
+        """执行前置工具；若 state 上带有 pagecontent（并行时每页的 state），则优先使用，避免用到图节点注册时捕获的全量 pagecontent。"""
+        results = await super().execute_pre_tools(state)
+        pagecontent = getattr(state, "pagecontent", None)
+        if pagecontent is not None and isinstance(pagecontent, list) and len(pagecontent) > 0:
+            results["pagecontent"] = pagecontent
+            log.debug("使用 state.pagecontent 作为本页 pagecontent（共 %s 项）", len(pagecontent))
+        return results
+
+    def get_default_pre_tool_results(self) -> Dict[str, Any]:
+        return {}
+
+    def _get_beamer_code_from_result(self, result: Dict[str, Any]) -> str:
+        """从 result 中取出 Beamer 代码，兼容规范 dict 或解析失败时的 {"raw": content}。"""
+        raw = result.get("latex_code", "") if isinstance(result, dict) else ""
+        if isinstance(raw, str) and raw:
+            code = extract_beamer_code(raw)
+            if code:
+                return code
+        # 解析失败时 result 可能为 {"raw": content}，尝试从原始文本提取
+        raw_content = result.get("raw", "") if isinstance(result, dict) else ""
+        if isinstance(raw_content, str) and raw_content:
+            code = extract_beamer_code(raw_content)
+            if code:
+                return code
+            try:
+                from dataflow_agent.utils import robust_parse_json
+                parsed = robust_parse_json(raw_content)
+                if isinstance(parsed, dict):
+                    raw = parsed.get("latex_code", "")
+                    if isinstance(raw, str) and raw:
+                        code = extract_beamer_code(raw)
+                        if code:
+                            return code
+            except Exception:
+                pass
+        return ""
+
+    def update_state_result(
+        self,
+        state: MainState,
+        result: Dict[str, Any],
+        pre_tool_results: Dict[str, Any],
+    ):
+        beamer_code = self._get_beamer_code_from_result(result)
+        if not beamer_code:
+            log.error("p2b_pagecontent_to_beamer: 未得到有效 Beamer 代码")
+            super().update_state_result(state, result, pre_tool_results)
+            return
+
+        result_path = getattr(state, "result_path", "") or ""
+        if result_path:
+            base = Path(result_path).expanduser().resolve()
+        else:
+            req = getattr(state, "request", None)
+            paper_pdf_path = getattr(req, "paper_pdf_path", "") if req else ""
+            base = Path(paper_pdf_path).expanduser().resolve().parent if paper_pdf_path else Path(".").resolve()
+        output_dir = base / "output"
+        output_dir.mkdir(parents=True, exist_ok=True)
+        beamer_code_path = output_dir / "beamer_code.tex"
+        beamer_code_path.write_text(beamer_code, encoding="utf-8")
+        state.beamer_code_path = str(beamer_code_path)
+        log.info("p2b_pagecontent_to_beamer: Beamer 代码已写入 %s", beamer_code_path)
+        super().update_state_result(state, result, pre_tool_results)

dataflow_agent/agentroles/paper2any_agents/p2v_beamer_code_debug_agent.py

@@ -18,6 +18,7 @@
 from dataflow_agent.logger import get_logger
 from dataflow_agent.agentroles.cores.base_agent import BaseAgent
 from dataflow_agent.agentroles.cores.registry import register
+from dataflow_agent.toolkits.p2vtool.p2v_tool import extract_beamer_code
 
 log = get_logger(__name__)
 
@@ -58,6 +59,40 @@ def get_default_pre_tool_results(self) -> Dict[str, Any]:
         """若调用方未显式传入，返回默认前置工具结果"""
         return {}
 
+    async def execute_pre_tools(self, state: MainState) -> Dict[str, Any]:
+        """先执行父类前置工具；若 state 上有 pre_tool_results（workflow 内注入），则合并进结果，保证 beamer_code/code_debug_result 能进入 prompt。"""
+        results = await super().execute_pre_tools(state)
+        inject = getattr(state, "pre_tool_results", None) or {}
+        for key in ("beamer_code", "code_debug_result"):
+            if key in inject:
+                results[key] = inject[key]
+        return results
+
+    def _get_beamer_code_from_result(self, result: Dict[str, Any]) -> str:
+        """从 result 中取出 Beamer 代码，兼容规范 dict 或解析失败时的 {"raw": content}。"""
+        raw = result.get("latex_code", "") if isinstance(result, dict) else ""
+        if isinstance(raw, str) and raw:
+            code = extract_beamer_code(raw)
+            if code:
+                return code
+        raw_content = result.get("raw", "") if isinstance(result, dict) else ""
+        if isinstance(raw_content, str) and raw_content:
+            code = extract_beamer_code(raw_content)
+            if code:
+                return code
+            try:
+                from dataflow_agent.utils import robust_parse_json
+                parsed = robust_parse_json(raw_content)
+                if isinstance(parsed, dict):
+                    raw = parsed.get("latex_code", "")
+                    if isinstance(raw, str) and raw:
+                        code = extract_beamer_code(raw)
+                        if code:
+                            return code
+            except Exception:
+                pass
+        return ""
+
     # ---------- 结果写回 ----------
     def update_state_result(
         self,
@@ -66,16 +101,19 @@ def update_state_result(
         pre_tool_results: Dict[str, Any],
     ):
         """将推理结果 {latex_code: xxxx} 写回 MainState"""
-        beamer_code = result.get("latex_code", '')
+        beamer_code = self._get_beamer_code_from_result(result)
         beamer_code_path = state.beamer_code_path
         if beamer_code and beamer_code_path:
             from pathlib import Path
 
             tex_path = Path(beamer_code_path)
             tex_path.write_text(beamer_code, encoding='utf-8')
-            # 编译最新的tex代码
+            # 编译最新的 tex 代码并写回 state，便于调用方判断是否仍存在 error/warning
             from dataflow_agent.toolkits.p2vtool.p2v_tool import compile_tex
             is_beamer_wrong, is_beamer_warning, code_debug_result = compile_tex(beamer_code_path)
+            state.is_beamer_wrong = is_beamer_wrong
+            state.is_beamer_warning = is_beamer_warning
+            state.code_debug_result = code_debug_result
             state.ppt_path = beamer_code_path.replace(".tex", ".pdf")
             log.info(f"将更新好的beamer code写回 {beamer_code_path}")
         else:

dataflow_agent/promptstemplates/prompts_repo.py

@@ -1794,23 +1794,85 @@ class Paper2VideoPrompt:
 
 ## Source Content (Markdown)
 {pdf_markdown}
+"""
+
+  system_prompt_for_p2b_pagecontent_to_beamer = """
+You are an expert in LaTeX Beamer. Your task is to convert **one slide's** structured outline (pagecontent) into a **single, structurally complete, compilable** Beamer LaTeX document.
+
+**Context:** You generate slide content **one page at a time**. Each output must be a **full Beamer document** that compiles on its own (with Tectonic or TeX Live). Do not output a bare frame or fragment.
+
+**Required document structure (do not omit any part):**
+1. \\documentclass{{beamer}}
+2. Preamble: **must** use \\usetheme{{Madrid}} (fixed theme).
+3. \\begin{{document}}
+4. **Exactly one** \\begin{{frame}}...\\end{{frame}} containing the slide content.
+5. \\end{{document}}
+
+**CRITICAL:** Ensure every \\begin{{frame}} has a matching \\end{{frame}}, and the document ends with \\end{{document}}. Avoid the error "!File ended while scanning use of \\frame".
+
+**Font and package rules (strict):**
+- **STRICTLY FORBIDDEN:** Times New Roman, Arial, Calibri, TeX Gyre Termes, or any non-standard TeX Live font. Use \\usepackage{{lmodern}} or default LaTeX fonts only.
+- **Do NOT use** \\usepackage{{resizebox}} (invalid/grammar issues).
+- If output_language is Chinese, you **must** include in the preamble: \\usepackage{{fontspec}} and \\usepackage{{ctex}}.
+
+**Syntax rules:**
+- **Do not use & in frame titles** (causes "Misplaced alignment tab character &"). Use "and" or comma instead.
+- **Underscore in plain text:** In LaTeX, underscore `_` is reserved for math subscripts. Any `_` in normal text (e.g. function names like generate_from_input, variable names like user_inputs, system_prompt) **must** be written as \\_ (backslash-underscore). Example: `user_inputs` → `user\\_inputs`, `generate_from_input` → `generate\\_from\\_input`. Otherwise you get "Missing $ inserted" and compilation fails.
+- Use \\alert{{}} for key terms or math symbols when appropriate.
+- For literal percent sign in text use \\% (e.g. 5\\%).
+
+**Content:** Use the given title, layout_description, key_points, and asset_ref. For image paths (e.g. in asset_ref), prepend the absolute base path given by pdf_images_working_dir and use \\includegraphics[width=0.8\\textwidth]{{...}} with \\caption and \\label. For table references (e.g. Table_2) use tabular/booktabs.
+
+**Output:** Return only one JSON object with key "latex_code" containing the **entire** document from \\documentclass to \\end{{document}}, ready to compile.
+"""
+
+  task_prompt_for_p2b_pagecontent_to_beamer = """
+Generate **one** LaTeX Beamer slide as a **complete, compilable document**. The input is a **single slide's** pagecontent (one JSON object). Your output must be a full Beamer file: \\documentclass{{beamer}} + preamble + \\begin{{document}} + **one** \\begin{{frame}}...\\end{{frame}} + \\end{{document}}.
+
+## Output language
+{output_language}
+
+## Images base directory (absolute path prefix for \\includegraphics)
+{pdf_images_working_dir}
+
+## This slide's pagecontent (single object)
+{pagecontent}
+
+## Asset / image rule
+- If **asset_ref** is null or missing: do **not** output any figure, image block, or placeholder (e.g. do not write "配图占位" or "当前页未提供图片资源"). 
+
+## Format requirements
+- **Theme: use \\usetheme{{Madrid}} in the preamble** (fixed; do not use other themes).
+- Font: use \\usepackage{{lmodern}} or default fonts only. **Do not use** Times New Roman, TeX Gyre Termes, resizebox.
+- Chinese: if output language is Chinese, add \\usepackage{{fontspec}} and \\usepackage{{ctex}} in the preamble.
+- No **&** in frame title (use "and" or comma).
+- **Underscores in text:** Write \\_ for every underscore in normal text (e.g. user\\_inputs, generate\\_from\\_input), or you get "Missing $ inserted".
+- Literal percent: use 5\\% not 5%.
+- Every \\begin{{frame}} must have \\end{{frame}}; document must end with \\end{{document}}.
+
+## Output format
+Return a valid JSON object with a single key "latex_code".
+
+{{
+  "latex_code": "FULL_BEAMER_DOCUMENT_WITH_ONE_FRAME_HERE"
+}}
 """
 
   system_prompt_for_p2v_beamer_code_debug = """
-You are an expert in repairing LaTeX beamer code. 
+You are an expert in repairing LaTeX beamer code.
 You must preserve all slide content exactly as written (including text, figures, and layout).
-Your goal is to correct LaTeX compilation errors and return clean, compilable LaTeX code.
+Your goal is to fix LaTeX compilation **errors** and **warnings** (e.g. Overfull box) and return clean, compilable LaTeX code.
 
 Your output must:
 - Be directly compilable using **tectonic** (a simplified TeX Live)
 - Never include explanations, comments, or English/Chinese text outside the LaTeX code
-
 """
 
   task_prompt_for_p2v_beamer_code_debug = """
-(Critical!) Do not modify the file path, ignore the folloing message: "warning: accessing absolute path: "
-You are given a LaTeX beamer code for the slides of a research paper and its error information.
-You should correct these errors but do not change the slide content (e.g., text, figures and layout).
+(Critical!) Do not modify the file path; ignore the following message: "warning: accessing absolute path: "
+
+You are given a LaTeX beamer code for the slides of a research paper and its compilation log (errors and/or warnings).
+Fix the reported issues but do not change the slide content (e.g., text, figures and layout).
 
 ## Content Preservation Rules (Strict)
 - You MUST NOT delete, replace, or reduce the number of figures/images.
@@ -1819,23 +1881,30 @@ class Paper2VideoPrompt:
   ONLY if necessary to fix compilation or layout issues.
 - Keep the slide text content unchanged as much as possible.
 
-## Some instruction
+## Overfull box (warning)
+When the log contains **Overfull \\hbox** or **Overfull \\vbox** (content or font too large), fix by:
+- Reducing font size (e.g. \\small, \\footnotesize in the frame or for specific blocks).
+- Reducing image/figure width or scale (e.g. width=0.7\\textwidth instead of 0.9\\textwidth).
+- Do NOT remove or truncate text or figures; only resize or rescale to fit.
+
+## Other instructions
 **Font Safety**: **MUST** remove or comment out any usage of the `fontspec` package if and only if it causes errors (as it depends on system fonts).
-For instance, if you encounter the error message: Package fontspec Error: The font "Latin Modern Roman" cannot be found, just remove or comment out it and use default TeX Live fonts.
+For instance, if you see: Package fontspec Error: The font "Latin Modern Roman" cannot be found, remove or comment it out and use default TeX Live fonts.
 
-**Image Loading Errors**: 
-If the compiler reports an image loading **error**, such as: "Unable to load picture or PDF file" or "! LaTeX Error: Cannot determine size of graphic", the model **MUST** remove the entire command responsible for loading that specific graphic.
+**Image Loading Errors**:
+If the compiler reports an image loading **error** (e.g. "Unable to load picture or PDF file" or "! LaTeX Error: Cannot determine size of graphic"), **MUST** remove the entire command that loads that graphic.
 
-Output Format:
-- Return a JSON object with a single key "latex_code".
+## Output format
+Return a JSON object with a single key "latex_code".
 {{
   "latex_code": "YOUR_GENERATED_latex_beamer_code_HERE"
 }}
-# Only output latex code which should be ready to compile using tectonic (simple version of TeX Live).
+Output only the JSON; the latex code must be ready to compile with tectonic.
 
-The LateX beamer code is:
+The LaTeX beamer code is:
 {beamer_code}
-The compilation error message is:
+
+The compilation log (errors and/or warnings) is:
 {code_debug_result}
 """
 

dataflow_agent/state.py

@@ -164,6 +164,36 @@ class Paper2VideoState(MainState):
     video_path: str = ""
 
 
+# ==================== Paper2PptBeamer 相关 State 和 Request 定义 ====================
+@dataclass
+class Paper2PptBeamerRequest(MainRequest):
+    """仅用于 PDF → Beamer PPT 工作流"""
+    paper_pdf_path: str = ""
+
+
+# ==================== Paper2PptBeamer 生成 State ======================
+@dataclass
+class Paper2PptBeamerState(MainState):
+    """用于 pagecontent → Beamer PPT 工作流（接在 paper2page_content 之后）"""
+    request: Paper2PptBeamerRequest = field(default_factory=Paper2PptBeamerRequest)
+
+    # 来自上游 paper2page_content 的产出
+    pagecontent: List[Dict[str, Any]] = field(default_factory=list)
+    result_path: str = ""
+    mineru_root: str = ""
+    minueru_output: str = ""  # 论文全文/摘要，供 table_extractor 等使用
+
+    beamer_code_path: str = ""
+    is_beamer_wrong: bool = False
+    is_beamer_warning: bool = False
+    code_debug_result: str = ""
+    ppt_path: str = ""
+    img_size_debug: bool = True
+
+    # 每页单独生成时的路径列表（页序）
+    per_page_beamer_paths: List[str] = field(default_factory=list)
+    per_page_pdf_paths: List[str] = field(default_factory=list)
+
 
 # ==================== Planning Agent 相关 State ====================
 @dataclass