mirror of
https://github.com/infiniflow/ragflow.git
synced 2025-12-08 20:42:30 +08:00
8c28587821
### What problem does this PR solve? ##### Problem Description When parsing HTML files, some page content may be lost. For example, text inside nested `<font>` tags within multiple `<div>` elements (e.g., `<div><font>Text_1</font></div><div><font>Text_2</font></div>`) fails to be preserved correctly. ###### Root Cause #1: Block ID propagation is interrupted 1. **Block ID generation**: When the parser encounters a `<div>`, it generates a new `block_id` because `<div>` belongs to `BLOCK_TAGS`. 2. **Recursive processing**: This `block_id` is passed down recursively to process the `<div>`’s child nodes. 3. **Interruption occurs**: When processing a child `<font>` tag, the code enters the `else` branch of `read_text_recursively` (since `<font>` is a Tag). 4. **Bug location**: The first line in this `else` branch explicitly sets **`block_id = None`**. - This discards the valid `block_id` inherited from the parent `<div>`. - Since `<font>` is not in `BLOCK_TAGS`, it does not generate a new `block_id`, so it passes `None` to its child text nodes. 5. **Consequence**: The extracted text nodes have an empty `block_id` in their `metadata`. During the subsequent `merge_block_text` step, these texts cannot be correctly associated with their original `<div>` block due to the missing ID. As a result, all text from `<font>` tags gets merged together, which then triggers a second issue during concatenation. 6. **Solution:** Remove the forced reset of `block_id` to `None`. When the current tag (e.g., `<font>`) is not a block-level element, it should inherit the `block_id` passed down from its parent. This ensures consistent ownership across the hierarchy: `div` → `font` → `text`. ###### Root Cause #2: Data loss during text concatenation 1. The line `current_content += (" " if current_content else "" + content)` has a misplaced parenthesis. When `current_content` is non-empty (`True`): - The ternary expression evaluates to `" "` (a single space). - The code executes `current_content += " "`. - **Result**: Only a space is appended—**the new `content` string is completely discarded**. ### Type of change - [x] Bug Fix (non-breaking change which fixes an issue)