FileArtifactService changes CRLF and CR line endings in text artifacts #7914

🔴 Required Information

Describe the Bug:

FileArtifactService changes the line endings of text artifacts. A text artifact containing \r\n or a lone \r loads back with \n in their place. The payload is read with Path.read_text, whose universal newline mode rewrites both. It is written with Path.write_text, which translates \n to os.linesep, so on Windows every \n is stored as \r\n.

InMemoryArtifactService and GcsArtifactService return the saved text unchanged, so the same artifact differs only on the file backend. That backend is what adk web and adk run use by default (use_local_storage=True, via create_local_artifact_service). CSV files (RFC 4180 uses CRLF), patches, and Windows-authored text a tool saves therefore load back different from what was saved.

Steps to Reproduce:

1. pip install google-adk (reproduced on main at 8cd3b12)
2. Run:

import asyncio, tempfile
from google.genai import types
from google.adk.artifacts.file_artifact_service import FileArtifactService
from google.adk.artifacts.in_memory_artifact_service import InMemoryArtifactService

async def main():
  text = "id,name\r\n1,a\r\n2,b\rc\n"
  for svc in (FileArtifactService(root_dir=tempfile.mkdtemp()), InMemoryArtifactService()):
    await svc.save_artifact(app_name="a", user_id="u", session_id="s",
                            filename="t.csv", artifact=types.Part(text=text))
    got = await svc.load_artifact(app_name="a", user_id="u", session_id="s",
                                  filename="t.csv")
    print(type(svc).__name__, repr(got.text), got.text == text)

asyncio.run(main())

Expected Behavior:

Both services return the saved text unchanged.

Observed Behavior:

FileArtifactService 'id,name\n1,a\n2,b\nc\n' False
InMemoryArtifactService 'id,name\r\n1,a\r\n2,b\rc\n' True

Environment Details:

• ADK Library Version (pip show google-adk): 2.13.0 (main at 8cd3b12)
• Desktop OS: macOS (the read-side rewrite is platform independent; the write-side translation affects Windows)
• Python Version (python -V): 3.11

Model Information:

• Are you using LiteLLM: No
• Which model is being used: N/A (artifact service only)

🟡 Optional Information

Regression:

Not that I know of. The file backend has read and written text payloads with read_text / write_text.

Additional Context:

I would like to contribute a focused fix for this. Would writing artifact.text.encode("utf-8") with write_bytes and decoding read_bytes() on load be an acceptable approach? This keeps the stored payload exactly equal to the saved text. The error handling stays the same (OSError / UnicodeDecodeError are still caught and logged). Metadata JSON reads and writes are unaffected, since JSON escapes CR/LF inside strings.

How often has this issue occurred?:

• Always (100%)