## Summary
- Frontend `Images` already includes `bmp`
(`web/src/constants/common.ts`), and picture parsing already accepts
`.bmp` (`internal/parser/parser/picture_parser.go`, `rag/app/picture.py`
via Pillow).
- This PR adds `bmp` to both whitelist sites and a small Go unit test.
Co-authored-by: zhangjiangshan1 <zhangjiangshan1@kingsoft.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Refactor the Extractor component into a pure, unified **5-in-1 modular extraction engine** across both Dataset (`knowledgebase.parser_config`) and Pipeline (Canvas DSL).
## Summary
- Disable DeepSeek V4 thinking (chain-of-thought) by default.
- litellm 1.82.x drops `thinking: disabled`; carry the toggle through
`extra_body.thinking.type` and strip `reasoning_effort` to avoid the
400.
- Use local timezone for agent `sys.date` instead of UTC.
Reference: https://api-docs.deepseek.com/guides/thinking_mode
---------
Co-authored-by: Claude <claude@example.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Claude <noreply@anthropic.com>
This PR modularizes the **Extractor** component configuration with dedicated feature subtabs, adds independent system prompt configuration, fixes multi-node execution determinism and parameter persistence across save and page refresh, and ensures backward compatibility with legacy flat fields.
### Summary
Pregel nodes created with `AddNodeWithOptions` retain a node-level
`RetryPolicy`, but task construction dropped that policy. When a task
had no policy, execution also bypassed the engine-level policy
configured with `WithRetryPolicy` and instantiated the default policy
directly.
Fixes the defect where selecting many files (e.g. 20+) in one dataset and starting parsing at once leaves most of them stuck in `RUNNING` forever: a few parse, the rest never do.
- The Go DLA client (`inference.Client.DLA`) kept `footer`/`header`/`reference` regions whose detection confidence was below 0.4.
- Python's production path (`parse_into_bboxes` -> `LayoutRecognizer.__call__`, `deepdoc/vision/layout_recognizer.py:97` and `:379`) drops such regions via a 0.4 garbage gate over
`garbage_layouts=['footer','header','reference']`.
- The shared `/predict/dla` backend runs `LayoutRecognizer.forward`, which does **not** apply that gate, so the Go client was the only place the filter was missing — a real Go<->Python parity gap.
## Problem
In the Go PDF pipeline, `processTablesWithReplacements`
(`internal/deepdoc/parser/pdf/table/table_post.go`) marked **every** box
overlapping a DLA table region for removal, then `insertTableBoxes`
re-inserted an HTML box only when `buildTableHTMLs` produced one.
When DLA over-labels a **text** box as a table but TSR produces **no
cells** (so `buildTableHTMLs` skips the table and `htmls[ti]` is unset),
the box was removed with nothing re-inserted — the original text was
**silently dropped**.
Fixes Go ingestion progress reporting and pipeline selection:
- Add timestamps to document progress logs.
- Keep document duration and status updated during parsing.
- Start frontend polling immediately after parsing begins.
- Prevent documents explicitly using General from inheriting an old
dataset pipeline.
- Populate missing pipeline operation log fields.
- Remove stale component progress logs between retries.
- Prevent progress values greater than `1`.
