docling-project / docling-project/docling
free(): invalid pointer crash when loading PDF with docling_parse (v4.0.1 / CPU / Python 3.12)
- Dominant language
- Python
- Stars
- 66.4k
- Forks
- 4.8k
- Avg merge
- 2d 21h
- Merged PRs (30d)
- 84
Description
Title:
free(): invalid pointer crash when loading PDF with docling_parse (v4.0.1 / CPU / Python 3.12)
Body:
Hey Docling team,
I’m consistently running into a fatal crash when using docling_parse (version 4.0.1) with Python 3.12 and the CPU-only version of PyTorch (no CUDA).
The crash happens during document parsing — specifically when calling _load_document() from pdf_parser.py in doclin_parse version 4.0.1. Here's the error:
```bash
free(): invalid pointer
Fatal Python error: Aborted
```
And this is the relevant part of the stacktrace:
File ".../docling_parse/pdf_parser.py", line 489 in _load_document
File ".../docling_parse/pdf_parser.py", line 455 in load
...
I’ve confirmed the following:
PyTorch is installed correctly as 2.6.0+cpu, no CUDA present
Python version is 3.12.3
No crash when importing torch or pdfminer.six separately
Downgrading docling_parse to 4.0.0 and 3.4.0 does not resolve the issue
The crash happens inside your library, likely due to a native extension or unsafe memory access
Would really appreciate it if you could look into this — especially since it completely breaks CPU-only setups. Let me know if you need more info to reproduce.
Thanks.
Contributor guide
Assessment
This issue has not been assessed yet.