[Bug]: add directory when unsupported file exists

[MaojiaSheng]

Bug Description

”新问题：提示不支持 .doc .wps
这也是在说明文件里面没有提到的问题
.wps 是不支持，还没有适配。doc 应该是没有枚举到，之前支持了 docx
一个目录中，导入 550个文件， 识别出 未支持72个文件，然后服务端就会中断加入到resources，也不会有错误提示。“

this issue is also related to #83

Steps to Reproduce

add a dir with .doc, .wps files

Expected Behavior

1. add all
2. ignore and feedback unsupported

Actual Behavior

break and failed

Minimal Reproducible Example

Error Logs

OpenViking Version

v0.1.18

Python Version

3.12

Operating System

Linux

Model Backend

None

Additional Context

No response

[shaoeric]

#128 使用markitdown支持的office组件，但markitdown支持docx 不支持doc，为了跨平台应用考虑，优先考虑让用户自己转换成docx pptx xlsx文件后上传

[MaojiaSheng]

后面我们考虑提供云上的解析模型服务，来适配大量的长尾格式。似乎没有哪个开源项目能承诺解决所有格式

[r266-tech]

Hi, I looked into this issue and found that the current main branch already has a directory scan architecture (directory_scan.py) that separates files into processable and unsupported categories, so the "all files fail" crash should no longer occur.

However, the core pain point remains: .doc and .wps files are not supported — they get classified as unsupported and silently skipped. The WordParser only handles .docx.

I'd like to submit a PR to add .doc support using docx2txt (pure Python, no system dependencies). This would directly address the original reporter's use case (550 files with 72 unsupported).

Would this direction be welcome?

[shaoeric]

@r266-tech docx2txt supports tables in .doc?

[r266-tech]

Good catch — docx2txt only handles .docx format, not the legacy .doc format. And for tables, it extracts them as flat tabbed text without preserving row/column structure.

For better .doc support, python-docx + antiword or textract would be needed. For table-aware extraction, docx2python preserves full table structure as nested lists.

Since my original comment proposed docx2txt for .doc support, that was inaccurate — it wouldn't actually solve the .doc problem. A more robust approach would be:

1. .docx tables → docx2python (pure Python, good table support)
2. .doc (legacy) → antiword or libreoffice --convert-to docx as preprocessing step

Happy to discuss the best path forward!

[shaoeric]

@r266-tech Interesting repo, docx2python performs better, refer https://github.com/ShayHill/docx2python , in RAG cases, docs include headers, footers, page id, tables, images, math equations, etc.
Some useless headers, footers and page id, font, styles can be ignored,.
tables -> text in markdown format.
images -> extracted img files, the img relative path inserted into the converted file
math equation -> latex text

welcome any other ideas