everalgo-parser 0.1.0__tar.gz → 0.2.0__tar.gz

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/CHANGELOG.md

@@ -5,12 +5,13 @@ All notable changes to this distribution will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
-Entries are generated by [git-cliff](https://git-cliff.org/) from Conventional
-Commit messages on `main`. See the repository root [`README.md` § Cutting a release](../../README.md#cutting-a-release)
-for the full workflow.
-
 <!-- git-cliff-unreleased-start -->
 ## [Unreleased]
+<!-- git-cliff-unreleased-end -->
+
+## [0.2.0] - 2026-05-27
+
+> First archived changelog. Entries below accumulated since the initial `0.1.0` PyPI release (published manually, without a git tag or per-package changelog), so this `0.2.0` section also consolidates the previously-unarchived `0.1.0` surface.
 
 ### Added
 
@@ -73,4 +74,5 @@ for the full workflow.
 - Optional `cairosvg >= 2.7.0` via `[svg]` extra.
 - System: LibreOffice (`soffice` on PATH) for Office document parsing.
 
-<!-- git-cliff-unreleased-end -->
+[Unreleased]: https://github.com/EverMind-AI/EverAlgo/compare/everalgo-parser/v0.2.0...HEAD
+[0.2.0]: https://github.com/EverMind-AI/EverAlgo/releases/tag/everalgo-parser/v0.2.0

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/PKG-INFO

@@ -1,22 +1,26 @@
 Metadata-Version: 2.4
 Name: everalgo-parser
-Version: 0.1.0
+Version: 0.2.0
 Summary: EverAlgo parser: multimodal raw-file parsing (image / audio / document / video / url) into ParsedContent.
 Project-URL: Homepage, https://github.com/EverMind-AI/EverAlgo
 Project-URL: Repository, https://github.com/EverMind-AI/EverAlgo
 Project-URL: Issues, https://github.com/EverMind-AI/EverAlgo/issues
 Project-URL: Documentation, https://github.com/EverMind-AI/EverAlgo/tree/main/packages/everalgo-parser
 Project-URL: Changelog, https://github.com/EverMind-AI/EverAlgo/blob/main/packages/everalgo-parser/CHANGELOG.md
-Author: EverMind
+Author-email: EverMind <Evermind@shanda.com>
 License-Expression: MIT
 License-File: LICENSE
+Keywords: everalgo,llm,multimodal,parsing
 Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
 Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
 Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Requires-Python: >=3.12
 Requires-Dist: asgiref>=3.8.0
 Requires-Dist: beautifulsoup4>=4.12.0
-Requires-Dist: everalgo-core<2.0.0,>=0.1.0
+Requires-Dist: everalgo-core<2.0.0,>=0.2.0
 Requires-Dist: httpx>=0.27.0
 Requires-Dist: pillow>=10.0.0
 Provides-Extra: svg

everalgo_parser-0.2.0/docs/architecture.md

@@ -0,0 +1,307 @@
+# everalgo-parser — 模块 & 接口图
+
+供 code review 用,展示当前实现的模块边界、调用关系、对外接口、类型契约、依赖拓扑。
+
+GitHub 默认渲染下面的 Mermaid 图。
+
+---
+
+## 1. 包内文件结构
+
+```
+packages/everalgo-parser/
+├── src/everalgo/parser/
+│   ├── __init__.py        ← 顶层 facade:dispatch + re-export 公开类型
+│   ├── _utils.py          ← 私有助手:fetch_uri / decode_bytes / aspect-ratio / html-clean
+│   ├── image.py           ← IMAGE modality 算子(含长图分块 + BMP/TIFF 转码 + SVG 栅格化)
+│   ├── audio.py           ← AUDIO modality 算子
+│   ├── document.py        ← PDF / HTML / EMAIL / Office(LibreOffice)四合一
+│   ├── url.py             ← URL modality 算子(HTTP 抓取 + OG 抽取 + 委托 HTML)
+│   ├── video.py           ← stub,等 ADR
+│   └── prompts/
+│       ├── __init__.py
+│       ├── en/{image,audio,document}.py    ← PROMPT_FOR_{PICTURE,MERGE,AUDIO,FILE,HTML}
+│       └── zh/{image,audio,document}.py    ← 中文对偶
+├── tests/
+│   ├── fixtures/          ← 17 个真实 sample.* 文件
+│   └── parser/
+│       ├── test_audio.py
+│       ├── test_dispatch.py
+│       ├── test_document.py
+│       ├── test_e2e_all_formats.py
+│       ├── test_e2e_pdf_openrouter.py
+│       ├── test_image.py
+│       ├── test_public_api.py
+│       ├── test_url.py
+│       ├── test_utils.py
+│       └── test_video.py
+├── pyproject.toml
+├── README.md
+└── CHANGELOG.md
+```
+
+---
+
+## 2. 公开 API 表面(facade)
+
+`from everalgo.parser import ...` 一处导入,2 个函数 + 3 个类型。
+
+| 符号 | 种类 | 签名 / 说明 |
+|---|---|---|
+| `aparse` | async fn | `async def aparse(raw_file: RawFile, *, llm: LLMClient \| None = None) -> ParsedContent` |
+| `parse` | sync fn | sync 桥接(`asgiref.async_to_sync`),不要在事件循环里调 |
+| `RawFile`, `ParsedContent`, `Modality` | type re-export | 来自 `everalgo.types` |
+
+5 个 modality 子模块(`image` / `audio` / `document` / `url` / `video`)是顶层 dispatcher 的内部实现,不在 `__all__` 中;通过 Python attribute lookup 仍可访问,但不是公开承诺(对标 sklearn / numpy 模式)。
+
+---
+
+## 3. 顶层 dispatch 流(`parser.aparse`)
+
+```mermaid
+flowchart TD
+  A[aparse raw_file] --> B{has content?}
+  B -- no --> C{uri is http/https?}
+  C -- yes --> URL[url.aparse: fetch then dispatch by Content-Type]
+  C -- no --> E1[ValueError: no content, no fetchable uri]
+  B -- yes --> M1[try get_modality extension]
+  M1 --> M2{UNKNOWN and mime set?}
+  M2 -- yes --> M3[fall back to get_modality_from_mime]
+  M2 -- no --> M4[use extension-derived modality]
+  M3 --> N{Modality}
+  M4 --> N
+  N -- still UNKNOWN --> E2[ValueError: cannot map to Modality]
+  N -- IMAGE --> IMG[image.aparse]
+  N -- AUDIO --> AUD[audio.aparse]
+  N -- PDF / DOCUMENT / HTML / EMAIL --> DOC[document.aparse]
+  N -- DIRECT --> DIR[passthrough: bytes utf-8 decode]
+  URL --> R[ParsedContent modality=URL]
+  IMG --> R2[ParsedContent modality=IMAGE]
+  AUD --> R3[ParsedContent modality=AUDIO]
+  DOC --> R4[ParsedContent modality=PDF/DOCUMENT/HTML/EMAIL]
+  DIR --> R5[ParsedContent modality=DIRECT]
+```
+
+**关键规则**:
+- `uri=http(s)://` + `content=b""` → 走 `url.aparse`(内部抓取 + 按 Content-Type 再分派)。
+- 已经有 `content`:**extension 优先**,extension 解析不到时**用 mime 兜底**(`get_modality_from_mime`)。这样 design.md §2.1 的 `RawFile(uri=..., mime="application/pdf")` 走 URL 抓取后能再走到 PDF handler。
+- 子模块内部派发用 extension(`_MIME_MAP[ext]` 查 mime)。当顶层用 mime 解出 modality 时,自动 `model_copy(extension=get_extension_from_mime(mime))` 给子模块。
+- `file://` 在 `_utils.fetch_uri` 里直接拒绝(AGENTS.md §1)。
+- `Modality.URL` 是顶层 dispatch 产生的"来源标签",不出现在 extension 派发表里。
+
+---
+
+## 4. 模块依赖图(import 关系)
+
+```mermaid
+graph LR
+  PARSER[parser.__init__<br/>facade] --> IMAGE[image.py]
+  PARSER --> AUDIO[audio.py]
+  PARSER --> DOC[document.py]
+  PARSER --> URL[url.py]
+  PARSER --> VIDEO[video.py stub]
+
+  IMAGE --> UTILS[_utils.py]
+  IMAGE --> PROMPTS_EN[prompts.en]
+  DOC --> UTILS
+  DOC --> PROMPTS_EN
+  AUDIO --> PROMPTS_EN
+  URL --> UTILS
+  URL -. lazy .-> DOC
+
+  IMAGE --> CORE[everalgo-core:<br/>llm + types]
+  AUDIO --> CORE
+  DOC --> CORE
+  URL --> CORE
+
+  IMAGE -.-> PIL[Pillow]
+  IMAGE -.-> CAIRO[cairosvg<br/>optional svg extra]
+  DOC -.-> BS4[beautifulsoup4]
+  DOC -.-> SOFFICE[soffice<br/>system dep]
+  UTILS -.-> BS4
+  UTILS -.-> PIL
+  UTILS -.-> HTTPX[httpx]
+```
+
+**说明**:
+- `url.py` → `document.py` 是 **lazy import**(写在函数内,避免循环 import)
+- 5 个 parser 模块都依赖 `everalgo-core`(LLM 调用 + 类型)
+- `cairosvg` 是 `[svg]` extra,延迟 import(只在解析 svg 时触发)
+- `soffice` 是**系统包**,运行时 detection(`shutil.which`)
+
+---
+
+## 5. 类型契约
+
+### `RawFile`(输入)
+
+| 字段 | 类型 | 默认 | 说明 |
+|---|---|---|---|
+| `content` | `bytes` | `b""` | 文件 bytes。可选,但 `content` / `uri` 至少有一个 |
+| `mime` | `str` | `""` | MIME hint,如 `application/pdf` |
+| `extension` | `str` | `""` | 不带点小写扩展名,如 `pdf`,**派发依据** |
+| `uri` | `str` | `""` | `http`/`https` 才会被抓取;`file://` 拒绝 |
+
+### `ParsedContent`(输出)
+
+| 字段 | 类型 | 说明 |
+|---|---|---|
+| `text` | `str` | 提取后的主文本(Markdown / 纯文本) |
+| `modality` | `Modality` | 来源分类 |
+| `mime` | `str` | 源 MIME(无则空串) |
+| `metadata` | `dict[str, Any]` | 按模态填:`model` / `finish_reason` / `tall_image_parts` / `inline_image_ocr` / `og_tags` / `intermediate_pdf_bytes` 等 |
+
+### `Modality` 枚举(StrEnum,9 值)
+
+```
+IMAGE  PDF  AUDIO  DOCUMENT  HTML  EMAIL  URL  DIRECT  UNKNOWN
+```
+
+`get_modality(ext: str) -> Modality` 派发函数 + `EXTENSION_TO_MODALITY` 映射,都在 `everalgo-core/types/modality.py`。
+
+---
+
+## 6. 各 modality 调用链
+
+### 6.1 IMAGE(`image.aparse`)
+
+```mermaid
+flowchart LR
+  A[bytes + ext] --> B{ext in<br/>bmp/tif/tiff?}
+  B -- yes --> C[PIL transcode → PNG]
+  B -- no --> D{ext == svg?}
+  D -- yes --> E[cairosvg<br/>rasterise → PNG]
+  D -- no --> F[as-is]
+  C --> G[check_aspect_ratio]
+  E --> G
+  F --> G
+  G --> H{ratio > 10?}
+  H -- no --> I[single LLM call<br/>PROMPT_FOR_PICTURE]
+  H -- yes --> J[split_image_with_overlap<br/>N parts]
+  J --> K[per-part OCR<br/>N x LLM calls]
+  K --> L{N > 1?}
+  L -- no --> M[return part]
+  L -- yes --> N[LLM merge<br/>PROMPT_FOR_MERGE]
+  I --> R[ParsedContent]
+  M --> R
+  N --> R
+```
+
+### 6.2 DOCUMENT(`document.aparse`)
+
+```mermaid
+flowchart TD
+  A[bytes + ext] --> M{Modality from ext}
+  M -- PDF --> P[image_url part + PROMPT_FOR_FILE → 1 LLM call]
+  M -- HTML --> H[decode_bytes → clean_html_for_llm bs4 → text part + PROMPT_FOR_HTML → 1 LLM call]
+  M -- EMAIL --> E[stdlib email parse → headers + body]
+  E --> EI{inline images?}
+  EI -- yes --> EOCR[per-image OCR via LLM<br/>PROMPT_FOR_PICTURE]
+  EOCR --> ES[substitute cid placeholders]
+  EI -- no --> ES2[text only]
+  M -- DOCUMENT Office --> O[soffice convert-to pdf]
+  O --> P2[reuse PDF path]
+```
+
+### 6.3 AUDIO(`audio.aparse`)
+
+```
+RawFile.content (mp3/wav/m4a/...) → image_url data URI + PROMPT_FOR_AUDIO → 1 LLM call → ParsedContent.text
+```
+
+### 6.4 URL(`url.aparse`)
+
+```mermaid
+flowchart LR
+  A[uri or content] --> B{has content?}
+  B -- no --> C[fetch_uri httpx]
+  C -- http/https --> D[bytes + Content-Type]
+  C -- other scheme --> X[ValueError reject]
+  B -- yes --> D
+  D --> CT[get_modality_from_mime fetched Content-Type]
+  CT --> SW{inner Modality}
+  SW -- IMAGE --> IM[image.aparse]
+  SW -- AUDIO --> AU[audio.aparse]
+  SW -- PDF / DOCUMENT / EMAIL --> DC[document.aparse]
+  SW -- HTML / UNKNOWN-fallback --> HT[document.aparse extension=html]
+  HT -.OG/meta extract.-> META[og_tags / twitter_tags / favicon / ...]
+  IM --> R[ParsedContent modality=URL]
+  AU --> R
+  DC --> R
+  HT --> R
+  META --> R
+```
+
+**关键规则**:
+- 抓回来的 Content-Type **驱动 inner dispatch**。`application/pdf` → PDF handler;`image/png` → image handler;`audio/mpeg` → audio handler;`text/html` 或未知 → HTML handler(兜底)。
+- **OG / Twitter / `<meta>` 抽取仅 HTML 响应做** —— 非 HTML 响应的 `ParsedContent.metadata` 不带 `og_tags` / `title` 等字段。
+- 返回的 `modality` 永远是 `Modality.URL`,真实的 inner 类型放在 `metadata["inner_modality"]`,调用方既能知道"这是从网络来的",也能知道"实际是 PDF"。
+
+---
+
+## 7. 依赖拓扑
+
+### Pip 依赖(`pyproject.toml`)
+
+| 依赖 | 用途 | 何时拉 |
+|---|---|---|
+| `everalgo-core` | LLM 客户端 + 公共类型 + fake_llm | 必装 |
+| `asgiref` | sync 桥接(`async_to_sync`)| 必装 |
+| `pillow` | image 转码 + 长图分块 | 必装 |
+| `beautifulsoup4` | HTML 清洗 + URL OG 抽取 | 必装 |
+| `httpx` | URL HTTP 抓取 | 透传(`everalgo-core` 已经依赖 openai SDK 拉了) |
+| `cairosvg` | SVG 栅格化 | `pip install 'everalgo-parser[svg]'` 才装 |
+
+### 系统依赖
+
+| 程序 | 用途 | 检测 |
+|---|---|---|
+| `soffice`(LibreOffice)| Office 文档转 PDF | `shutil.which("soffice")` + macOS `/Applications/LibreOffice.app/Contents/MacOS/soffice` |
+
+---
+
+## 8. 测试覆盖
+
+| 文件 | 数量 | 类型 | LLM |
+|---|---|---|---|
+| `test_public_api.py` | 2 | 公开 API 形状 | 不用 |
+| `test_dispatch.py` | 25 | 顶层派发 | `FakeLLMClient` |
+| `test_image.py` | 12 | PIL 路径 + 长图分块 / 合并 | `FakeLLMClient` |
+| `test_audio.py` | 5 | audio 子模块 happy + error path | `FakeLLMClient` |
+| `test_document.py` | 12 | PDF / HTML / EML 单测 | `FakeLLMClient` |
+| `test_url.py` | 21 | OG 抽取 + httpx mock | `FakeLLMClient` + `respx` |
+| `test_utils.py` | 18 | `_utils` 助手(no LLM,no network)| 不用 |
+| `test_video.py` | 2 | video stub NotImplementedError | 不用 |
+| `tests/types/test_modality.py` | 12 | Modality enum + get_modality | 不用 |
+| `tests/types/test_parsed.py` | 3 | ParsedContent 字段 | 不用 |
+| `test_e2e_all_formats.py` | 7 | 全格式 e2e | 真 LLM(env `OPENROUTER_API_KEY`)|
+| `test_e2e_pdf_openrouter.py` | 1 | PDF spike e2e | 真 LLM(env `OPENROUTER_API_KEY`)|
+
+**测试总数**:`391`(parser+core)/`923`(workspace 全量)
+**E2E**(需 `OPENROUTER_API_KEY`):`8`(`test_e2e_all_formats.py` 7 + `test_e2e_pdf_openrouter.py` 1)
+
+---
+
+## 9. 公开规范契约
+
+| 规范 | 对应 |
+|---|---|
+| `a` 前缀代表 async | ✅ `aparse` async / `parse` sync 桥接(ADR-010) |
+| Protocol over ABC | ✅ 没有 `ABC` 基类,LLMClient 是 Protocol(ADR-011) |
+| 无 retry / fallback / metrics | ✅ 算法层不做工程高可用(ADR-012) |
+| 无状态(不读文件系统) | ✅ 接 `bytes` / HTTP,`file://` 拒绝(AGENTS.md §1) |
+| 懒 `%`-format 日志 | ✅ `logger.debug("count=%d", n)` 风格(ADR-013) |
+| Prompt = `prompts/{en,zh}/<name>.py` 模块常量 | ✅ |
+| Numpy-style docstring | ✅ |
+| 英文 docstring / 标识符 / commit | ✅(docs/ 下 design.md 例外,中文 OK) |
+
+---
+
+## 10. 已知未实现 / 推迟
+
+| 项 | 状态 | 备注 |
+|---|---|---|
+| `video.py` | stub | 上游 evermemos-multimodal **也没有**实现,Gemini Video vs Whisper+帧抽取 选型待 ADR |
+| `audio` BMP-like 转码 | 没做 | mp3/wav 测过;m4a/amr/aiff/aac/ogg/flac 未在 e2e 覆盖,Gemini 兼容度未全测 |
+| `Office` 系统依赖动态检测降级 | 仅运行时报错 | 没有 import 时友好警告;`_find_soffice` 缺失就 `RuntimeError` |
+| `T1` 主条目其他 schema | 待 RESOLVED | 我们只 RESOLVED 了 `T1.parser` 子集(`ParsedContent` / `Modality` / `RawFile`) |

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/pyproject.toml

@@ -4,19 +4,23 @@ build-backend = "hatchling.build"
 
 [project]
 name = "everalgo-parser"
-version = "0.1.0"
+version = "0.2.0"
 description = "EverAlgo parser: multimodal raw-file parsing (image / audio / document / video / url) into ParsedContent."
 requires-python = ">=3.12"
 license = "MIT"
 readme = "README.md"
-authors = [{ name = "EverMind" }]
+authors = [{ name = "EverMind", email = "Evermind@shanda.com" }]
+keywords = ["llm", "multimodal", "parsing", "everalgo"]
 classifiers = [
   "Development Status :: 3 - Alpha",
+  "Intended Audience :: Developers",
   "License :: OSI Approved :: MIT License",
+  "Programming Language :: Python :: 3",
   "Programming Language :: Python :: 3.12",
+  "Topic :: Scientific/Engineering :: Artificial Intelligence",
 ]
 dependencies = [
-  "everalgo-core>=0.1.0,<2.0.0",
+  "everalgo-core>=0.2.0,<2.0.0",
   "asgiref>=3.8.0",
   "pillow>=10.0.0",
   "beautifulsoup4>=4.12.0",

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/src/everalgo/parser/__init__.py

@@ -3,7 +3,6 @@
 Public surface (re-exports for ergonomic ``from everalgo.parser import ...``):
 
 - Functions: ``aparse`` / ``parse`` — dispatch entry points.
-- Submodules: ``image`` / ``audio`` / ``document`` / ``url`` / ``video``.
 - Types: ``ParsedContent`` / ``Modality`` / ``RawFile`` (re-exported from
   ``everalgo.types`` for one-stop import).
 """
@@ -16,7 +15,7 @@ from urllib.parse import urlparse
 
 from asgiref.sync import async_to_sync
 
-from everalgo.parser import audio, document, image, url, video
+from everalgo.parser import audio, document, image, url
 from everalgo.types import (
     Modality,
     ParsedContent,
@@ -34,12 +33,7 @@ __all__ = [
     "ParsedContent",
     "RawFile",
     "aparse",
-    "audio",
-    "document",
-    "image",
     "parse",
-    "url",
-    "video",
 ]
 
 logger = logging.getLogger(__name__)

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/src/everalgo/parser/_utils.py

@@ -1,8 +1,5 @@
 """Internal parser helpers.
 
-Migrated (with English docstrings) from
-``evermemos-multimodal/src/parser/utils.py`` (prod-20260306-0331-v1):
-
 - ``decode_bytes`` — multi-encoding decoder for unknown-charset bytes.
 - ``check_aspect_ratio`` / ``split_image_with_overlap`` — long-image (tall
   screenshot) handling for the image parser.

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/src/everalgo/parser/audio.py

@@ -1,7 +1,6 @@
 """Audio parser — ASR via multimodal LLM.
 
-Mirrors ``evermemos-multimodal/src/parser/audio/gemini_parser.py``: send the
-audio bytes as a single ``image_url`` data URI (OpenRouter / Gemini accepts
+Sends the audio bytes as a single ``image_url`` data URI (OpenRouter / Gemini accepts
 audio MIMEs through the same content-part shape) and the LLM transcribes.
 """
 

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/src/everalgo/parser/document.py

@@ -4,8 +4,7 @@ Per ``AGENTS.md`` and the engineering alignment for the parser migration, the
 ``document`` submodule covers every non-image / non-audio file shape:
 
 - ``Modality.PDF`` — sent as a single ``application/pdf`` data URI to the
-  multimodal LLM (mirrors the path used in
-  ``evermemos-multimodal/src/parser/pdf/gemini_parser.py``).
+  multimodal LLM.
 - ``Modality.HTML`` — bs4-cleaned (``_utils.clean_html_for_llm``) before
   sending the structured HTML to the LLM with the HTML prompt.
 - ``Modality.EMAIL`` — pure-python ``email`` stdlib parsing; when inline
@@ -50,8 +49,7 @@ _PDF_MIME = "application/pdf"
 _HTML_MAX_INPUT_CHARS = 1_000_000
 _MACOS_SOFFICE = "/Applications/LibreOffice.app/Contents/MacOS/soffice"
 
-# Verbatim from ``evermemos-multimodal/src/parser/email/eml_parser.py``:
-# inline image OCR caps at 8K tokens (single-image scope, no need for the
+# Inline image OCR caps at 8K tokens (single-image scope, no need for the
 # 32K merge budget used by the long-image stitcher).
 _INLINE_OCR_MAX_TOKENS = 8_000
 _INLINE_IMAGE_MIMETYPES = frozenset(
@@ -160,8 +158,6 @@ async def _aparse_html(raw_file: RawFile, *, llm: LLMClient) -> ParsedContent:
 async def _aparse_email(raw_file: RawFile, *, llm: LLMClient) -> ParsedContent:
     """EML → ``ParsedContent`` via stdlib ``email`` + inline-image OCR.
 
-    Mirrors evermemos-multimodal/src/parser/email/eml_parser.py:
-
     1. Extract headers.
     2. Collect inline images (Content-ID → bytes).
     3. Extract body: prefer ``text/plain`` when there are no cid images
@@ -261,7 +257,7 @@ async def _aparse_office(
 
 
 def _extract_email_headers(msg: EmailMessage) -> str:
-    lines = []
+    lines: list[str] = []
     for field in ("Subject", "From", "To", "Cc", "Date"):
         value = msg.get(field)
         if value:
@@ -367,7 +363,7 @@ async def _ocr_inline_images(
         except Exception as exc:
             # Best-effort per-image: a single OCR failure should not abort the
             # whole email parse — log and continue with a placeholder marker.
-            logger.warning("OCR failed for inline image %s: %s", filename, exc)
+            logger.warning("OCR failed for inline image %s: %s", filename, exc, exc_info=True)
             results[idx] = f"[image: {filename}, OCR failed]"
     return results
 

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/src/everalgo/parser/image.py

@@ -1,7 +1,5 @@
 """Image parser — multimodal LLM OCR with tall-image splitting.
 
-Mirrors ``evermemos-multimodal/src/parser/image/gemini_parser.py``:
-
 - Normal-ratio images go through a single LLM call.
 - Tall screenshots (height/width > 10) are split into overlapping slices
   via ``_utils.split_image_with_overlap``; each slice is OCR'd independently
@@ -54,10 +52,8 @@ _SVG_EXTENSIONS: frozenset[str] = frozenset({"svg"})
 
 _TALL_IMAGE_RATIO_THRESHOLD = 10.0
 
-# Token budgets — verbatim from
-# ``evermemos-multimodal/src/parser/image/gemini_parser.py``. Per-slice OCR
-# caps at 8K; the merge step gets 4x because it joins multiple OCR'd chunks
-# and the merged text is longer than any individual slice.
+# Token budgets. Per-slice OCR caps at 8K; the merge step gets 4x because it
+# joins multiple OCR'd chunks and the merged text is longer than any individual slice.
 _MAX_OCR_OUTPUT_TOKENS = 8_000
 _MAX_MERGE_OUTPUT_TOKENS = 32_000
 
@@ -138,9 +134,6 @@ async def _aparse_tall(
 ) -> ParsedContent:
     r"""Split-then-merge OCR for tall screenshots.
 
-    Mirrors ``evermemos-multimodal/src/parser/image/base.py:parse`` +
-    ``gemini_parser._merge_parts``:
-
     1. Split into N overlapping vertical slices.
     2. OCR each slice; **drop empty results** so the merge step sees only
        real text (an all-empty slice from a blank PDF region would otherwise
@@ -223,7 +216,9 @@ async def _aparse_tall(
             merge_fallback_used = True
         last_response = merge_resp
     except Exception as exc:
-        logger.warning("LLM merge of tall-image OCR parts failed (%s); falling back to deterministic join", exc)
+        logger.warning(
+            "LLM merge of tall-image OCR parts failed (%s); falling back to deterministic join", exc, exc_info=True
+        )
         merged_text = "\n\n".join(ocr_results)
         merge_fallback_used = True
 

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/src/everalgo/parser/prompts/en/audio.py

@@ -1,8 +1,4 @@
-"""Audio parser prompt.
-
-Lifted verbatim from ``evermemos-multimodal/src/parser/prompts.py``
-(prod-20260306-0331-v1).
-"""
+"""Audio parser prompt."""
 
 PROMPT_FOR_AUDIO = """This is a voice message sent by a user in a chat application. Transcribe it word by word.
 

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/src/everalgo/parser/prompts/en/document.py

@@ -1,9 +1,4 @@
-"""Document parser prompts (PDF / Office / HTML).
-
-Lifted verbatim from ``evermemos-multimodal/src/parser/prompts.py``
-(prod-20260306-0331-v1) where these strings have been iterated on against
-real customer documents.
-"""
+"""Document parser prompts (PDF / Office / HTML)."""
 
 PROMPT_FOR_FILE = """Read this document and return its content as if you were reading it naturally.
 

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/src/everalgo/parser/prompts/en/image.py

@@ -1,8 +1,4 @@
-"""Image parser prompts.
-
-Lifted verbatim from ``evermemos-multimodal/src/parser/prompts.py``
-(prod-20260306-0331-v1).
-"""
+"""Image parser prompts."""
 
 PROMPT_FOR_PICTURE = """Read this image and return its content as if you were reading it naturally.
 

everalgo_parser-0.2.0/src/everalgo/parser/video.py

@@ -0,0 +1,31 @@
+"""Video parser — deferred.
+
+No implementation yet; selection between Gemini Video and Whisper + frame
+extraction is a follow-up decision. Calling ``aparse`` / ``parse`` today
+raises ``NotImplementedError`` so dispatch in ``everalgo.parser.aparse``
+surfaces the deferred status cleanly to callers.
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from everalgo.llm import LLMClient
+    from everalgo.types import ParsedContent, RawFile
+
+__all__ = ["aparse", "parse"]
+
+
+async def aparse(raw_file: RawFile, *, llm: LLMClient) -> ParsedContent:
+    """Parse video → ``ParsedContent``. Deferred — see module docstring."""
+    raise NotImplementedError(
+        "video parser is deferred; selection between Gemini Video / Whisper + frame extraction pending an ADR."
+    )
+
+
+def parse(raw_file: RawFile, *, llm: LLMClient) -> ParsedContent:
+    """Sync bridge. Deferred — see module docstring."""
+    raise NotImplementedError(
+        "video parser is deferred; selection between Gemini Video / Whisper + frame extraction pending an ADR."
+    )

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/tests/parser/test_dispatch.py

@@ -246,7 +246,7 @@ async def test_aparse_raises_for_unsupported_modality(monkeypatch: pytest.Monkey
     from everalgo.types import Modality
 
     # Modality.URL is valid but not handled by the dispatch switch (IMAGE/AUDIO/PDF/etc.)
-    monkeypatch.setattr(pkg, "get_modality", lambda _ext: Modality.URL)
+    monkeypatch.setattr(pkg, "get_modality", lambda _ext: Modality.URL)  # type: ignore[misc]
     with pytest.raises(ValueError, match="Unsupported modality"):
         await parser_pkg.aparse(RawFile(content=b"x", extension="mp4"), llm=_fake())
 

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/tests/parser/test_document.py

@@ -130,7 +130,7 @@ async def test_aparse_html_empty_after_clean_returns_empty_parsed_content(monkey
     import everalgo.parser.document as doc_mod
 
     # Monkeypatch clean_html_for_llm to return "" so the empty-after-cleanup branch fires
-    monkeypatch.setattr(doc_mod, "clean_html_for_llm", lambda _: "")
+    monkeypatch.setattr(doc_mod, "clean_html_for_llm", lambda _: "")  # type: ignore[misc]
 
     fake = _make_fake()
     html = b"<html><body><p>some content</p></body></html>"

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/tests/parser/test_public_api.py

@@ -10,16 +10,8 @@ def test_top_level_routing_aparse_parse() -> None:
     assert callable(parse)
 
 
-def test_five_submodules_have_dual_interface() -> None:
-    from everalgo.parser import audio, document, image, url, video
-
-    for mod in (image, audio, document, video, url):
-        assert inspect.iscoroutinefunction(mod.aparse)
-        assert callable(mod.parse)
-
-
 def test_dunder_all_lists_public_symbols() -> None:
-    """Re-exports cover the 5 submodules + dispatch fns + 3 facade types."""
+    """``__all__`` lists only facade-level symbols; submodules are internal."""
     from everalgo.parser import __all__
 
     assert sorted(__all__) == sorted(
@@ -28,11 +20,6 @@ def test_dunder_all_lists_public_symbols() -> None:
             "ParsedContent",
             "RawFile",
             "aparse",
-            "audio",
-            "document",
-            "image",
             "parse",
-            "url",
-            "video",
         ]
     )

{everalgo_parser-0.1.0 → everalgo_parser-0.2.0}/tests/parser/test_utils.py

@@ -27,26 +27,26 @@ def _png_bytes(width: int, height: int) -> bytes:
 def test_check_aspect_ratio_normal_image_not_tall() -> None:
     needs_split, ratio = check_aspect_ratio(_png_bytes(100, 200))
     assert needs_split is False
-    assert ratio == pytest.approx(2.0)
+    assert ratio == pytest.approx(2.0)  # pyright: ignore[reportUnknownMemberType]
 
 
 def test_check_aspect_ratio_tall_image_needs_split() -> None:
     needs_split, ratio = check_aspect_ratio(_png_bytes(100, 2000))
     assert needs_split is True
-    assert ratio == pytest.approx(20.0)
+    assert ratio == pytest.approx(20.0)  # pyright: ignore[reportUnknownMemberType]
 
 
 def test_check_aspect_ratio_custom_max_ratio() -> None:
     # 200 / 100 = 2.0; with max_ratio=1.5 it should need split
     needs_split, ratio = check_aspect_ratio(_png_bytes(100, 200), max_ratio=1.5)
     assert needs_split is True
-    assert ratio == pytest.approx(2.0)
+    assert ratio == pytest.approx(2.0)  # pyright: ignore[reportUnknownMemberType]
 
 
 def test_check_aspect_ratio_square_image() -> None:
     needs_split, ratio = check_aspect_ratio(_png_bytes(100, 100))
     assert needs_split is False
-    assert ratio == pytest.approx(1.0)
+    assert ratio == pytest.approx(1.0)  # pyright: ignore[reportUnknownMemberType]
 
 
 # ---- split_image_with_overlap ----

everalgo_parser-0.1.0/src/everalgo/parser/video.py

@@ -1,38 +0,0 @@
-"""Video parser — deferred.
-
-evermemos-multimodal has no video parser implementation; the upstream
-``FILE_TYPE_DEFAULT_EXTENSION`` map reserves ``"video": "mp4"`` but no
-parser ever consumed it. Selection between Gemini Video and Whisper +
-frame extraction is a follow-up decision; when that lands, this stub is
-replaced with the real handler. Calling ``aparse`` / ``parse`` today raises
-``NotImplementedError`` so dispatch in ``everalgo.parser.aparse`` surfaces
-the deferred status cleanly to callers.
-"""
-
-from __future__ import annotations
-
-from typing import TYPE_CHECKING
-
-if TYPE_CHECKING:
-    from everalgo.llm import LLMClient
-    from everalgo.types import ParsedContent, RawFile
-
-__all__ = ["aparse", "parse"]
-
-
-async def aparse(raw_file: RawFile, *, llm: LLMClient) -> ParsedContent:
-    """Parse video → ``ParsedContent``. Deferred — see module docstring."""
-    raise NotImplementedError(
-        "video parser is deferred; selection between Gemini Video / Whisper "
-        "+ frame extraction pending an ADR. No upstream implementation exists "
-        "in evermemos-multimodal to port."
-    )
-
-
-def parse(raw_file: RawFile, *, llm: LLMClient) -> ParsedContent:
-    """Sync bridge. Deferred — see module docstring."""
-    raise NotImplementedError(
-        "video parser is deferred; selection between Gemini Video / Whisper "
-        "+ frame extraction pending an ADR. No upstream implementation exists "
-        "in evermemos-multimodal to port."
-    )