From e54f05cfd1782d236b2edc2d67199ea0c274b117 Mon Sep 17 00:00:00 2001
From: Rodion Kazennov <nefayran@gmail.com>
Date: Mon, 12 Oct 2026 11:41:07 +0000
Subject: [PATCH] fix(jobs): write and read job.json as UTF-8

The default name of an imported model contains U+00B7, which CP932 cannot
encode, so on Japanese Windows the import failed while writing job.json.
Reading falls back to the platform encoding for job.json files written by
earlier versions.

Fixes #131
---
 lelab/jobs.py      | 18 ++++++++++++++----
 tests/test_jobs.py | 43 +++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 57 insertions(+), 4 deletions(-)

diff --git a/lelab/jobs.py b/lelab/jobs.py
index 10810a16c..6a4e9ea60 100644
--- a/lelab/jobs.py
+++ b/lelab/jobs.py
@@ -636,6 +636,16 @@ def _job_meta_path(output_root: Path, job_id: str) -> Path:
     return _job_dir(output_root, job_id) / "job.json"
 
 
+def _read_job_meta(path: Path) -> dict:
+    """job.json is UTF-8. Earlier versions wrote it in the platform's default
+    encoding (CP932 on Japanese Windows), so fall back to that for old files."""
+    try:
+        text = path.read_text(encoding="utf-8")
+    except UnicodeDecodeError:
+        text = path.read_text()
+    return json.loads(text)
+
+
 class JobAlreadyRunningError(Exception):
     """Raised when start() is called while another local job is running."""
 
@@ -739,13 +749,13 @@ def _rewrite_output_dir_in_meta(self, job_dir: Path) -> None:
         if not meta.is_file():
             return
         try:
-            data = json.loads(meta.read_text())
+            data = _read_job_meta(meta)
         except Exception as exc:
             logger.warning("Migration: could not parse %s: %s", meta, exc)
             return
         data["output_dir"] = str(job_dir / "run")
         tmp = meta.with_suffix(meta.suffix + ".tmp")
-        tmp.write_text(json.dumps(data, indent=2))
+        tmp.write_text(json.dumps(data, indent=2), encoding="utf-8")
         os.replace(tmp, meta)
 
     def set_on_change(self, callback: Callable[[], None] | None) -> None:
@@ -1115,7 +1125,7 @@ def _load_from_disk(self) -> None:
             if not meta.exists():
                 continue
             try:
-                data = json.loads(meta.read_text())
+                data = _read_job_meta(meta)
                 record = JobRecord.model_validate(data)
             except Exception as exc:
                 logger.warning("Skipping malformed job.json at %s: %s", meta, exc)
@@ -1272,7 +1282,7 @@ def _write_meta(self, record: JobRecord) -> None:
         # Atomic write so a crash mid-write never strands a half-written file
         # that would skip the job on next _load_from_disk.
         tmp = path.with_suffix(path.suffix + ".tmp")
-        tmp.write_text(record.model_dump_json(indent=2))
+        tmp.write_text(record.model_dump_json(indent=2), encoding="utf-8")
         os.replace(tmp, path)
 
 
diff --git a/tests/test_jobs.py b/tests/test_jobs.py
index 10384b1f5..1db2384a9 100644
--- a/tests/test_jobs.py
+++ b/tests/test_jobs.py
@@ -271,3 +271,46 @@ def list_repo_files(self, repo_id, repo_type):
     assert rec.output_dir == ""
     cks = reg.list_checkpoints(rec.id)
     assert [c.ref for c in cks] == ["user/some-model@root"]
+
+
+def _default_to_cp932(monkeypatch) -> None:
+    """Make Path.read_text/write_text default to CP932, as on Japanese Windows."""
+    from pathlib import Path
+
+    real_write_text, real_read_text = Path.write_text, Path.read_text
+
+    def write_text(self, data, encoding=None, *args, **kwargs):
+        return real_write_text(self, data, encoding or "cp932", *args, **kwargs)
+
+    def read_text(self, encoding=None, *args, **kwargs):
+        return real_read_text(self, encoding or "cp932", *args, **kwargs)
+
+    monkeypatch.setattr(Path, "write_text", write_text)
+    monkeypatch.setattr(Path, "read_text", read_text)
+
+
+def test_job_json_survives_a_cp932_default_encoding(monkeypatch, tmp_path) -> None:
+    """CP932 cannot encode the "·" in the default name of an imported model (#131)."""
+    from lelab.jobs import JobRegistry
+
+    model = tmp_path / "model"
+    _make_pretrained(model)
+    _default_to_cp932(monkeypatch)
+
+    rec = JobRegistry(tmp_path / "root").register_imported(str(model))
+
+    assert "·" in rec.name
+    assert JobRegistry(tmp_path / "root").get(rec.id).name == rec.name
+
+
+def test_cp932_job_json_written_before_the_fix_still_loads(monkeypatch, tmp_path) -> None:
+    from lelab.jobs import JobRegistry
+
+    model = tmp_path / "model"
+    _make_pretrained(model)
+    rec = JobRegistry(tmp_path / "root").register_imported(str(model), name="日本語のモデル")
+    meta = next((tmp_path / "root").rglob("job.json"))
+    meta.write_bytes(meta.read_text(encoding="utf-8").encode("cp932"))
+    _default_to_cp932(monkeypatch)
+
+    assert JobRegistry(tmp_path / "root").get(rec.id).name == "日本語のモデル"