fix(moss-cli): keep newlines in quoted csv fields on doc import

csv.DictReader was fed content.splitlines(), so newlines inside quoted fields were dropped and words on either side were joined. parse from a stream instead, validate the header, strip a utf-8 bom, and report real file line numbers in errors.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Author: RyanRana

packages/moss-cli/src/moss_cli/documents.py

@@ -3,10 +3,11 @@
 from __future__ import annotations
 
 import csv
+import io
 import json
 import sys
 from pathlib import Path
-from typing import Any, List
+from typing import Any, List, Optional
 
 import typer
 from moss import DocumentInfo
@@ -23,7 +24,7 @@ def load_documents(file_path: str) -> List[DocumentInfo]:
         raise typer.BadParameter(f"File not found: {file_path}")
 
     suffix = path.suffix.lower()
-    content = path.read_text()
+    content = path.read_text(encoding="utf-8-sig")
 
     if suffix == ".csv":
         return _parse_csv_docs(content)
@@ -67,42 +68,70 @@ def _parse_jsonl_docs(raw: str, source: str = "input") -> List[DocumentInfo]:
 
 
 def _parse_csv_docs(content: str) -> List[DocumentInfo]:
-    reader = csv.DictReader(content.splitlines())
+    reader = csv.DictReader(io.StringIO(content, newline=""))
+    if reader.fieldnames is None:
+        raise typer.BadParameter("CSV is empty: expected a header row")
+
+    reader.fieldnames = [(name or "").strip() for name in reader.fieldnames]
+
+    seen = set()
+    dupes = set()
+    for name in reader.fieldnames:
+        if not name:
+            continue
+        if name in seen:
+            dupes.add(name)
+        else:
+            seen.add(name)
+    if dupes:
+        raise typer.BadParameter(
+            f"CSV header has duplicate column name(s): {', '.join(sorted(dupes))}"
+        )
+
+    missing = [name for name in ("id", "text") if name not in reader.fieldnames]
+    if missing:
+        raise typer.BadParameter(
+            f"CSV header is missing required column(s): {', '.join(missing)}"
+        )
+
     docs = []
-    for i, row in enumerate(reader):
-        if "id" not in row or "text" not in row:
+    for row in reader:
+        line_no = reader.line_num
+        doc_id = row.get("id")
+        text = row.get("text")
+        if not doc_id or text is None:
             raise typer.BadParameter(
-                f"CSV row {i + 1}: missing required 'id' or 'text' column"
+                f"CSV line {line_no}: empty value in required 'id' or 'text' column"
             )
-        metadata = None
-        if "metadata" in row and row["metadata"]:
-            try:
-                metadata = json.loads(row["metadata"])
-            except json.JSONDecodeError:
-                raise typer.BadParameter(
-                    f"CSV row {i + 1}: invalid JSON in 'metadata' column"
-                )
-
-        embedding = None
-        if "embedding" in row and row["embedding"]:
-            try:
-                embedding = json.loads(row["embedding"])
-            except json.JSONDecodeError:
-                raise typer.BadParameter(
-                    f"CSV row {i + 1}: invalid JSON in 'embedding' column"
-                )
+
+        metadata = _parse_csv_json(row.get("metadata"), "metadata", line_no)
+        embedding = _parse_csv_json(row.get("embedding"), "embedding", line_no)
 
         docs.append(
             DocumentInfo(
-                id=row["id"],
-                text=row["text"],
+                id=doc_id,
+                text=text,
                 metadata=metadata,
                 embedding=embedding,
             )
         )
     return docs
 
 
+def _parse_csv_json(value: Optional[str], column: str, line_no: int) -> Any:
+    if value is None:
+        return None
+    value = value.strip()
+    if not value:
+        return None
+    try:
+        return json.loads(value)
+    except json.JSONDecodeError:
+        raise typer.BadParameter(
+            f"CSV line {line_no}: invalid JSON in '{column}' column"
+        )
+
+
 def _dict_to_doc(d: Any, index: int) -> DocumentInfo:
     if not isinstance(d, dict):
         raise typer.BadParameter(f"Document at index {index}: expected object, got {type(d).__name__}")

packages/moss-cli/tests/test_documents.py

@@ -0,0 +1,52 @@
+import pytest
+import typer
+
+from moss_cli.documents import _parse_csv_docs
+
+
+def test_csv_keeps_newlines_inside_quoted_text() -> None:
+    docs = _parse_csv_docs('id,text\ndoc1,"line one\nline two"\ndoc2,plain\n')
+    assert len(docs) == 2
+    assert docs[0].text == "line one\nline two"
+    assert docs[1].text == "plain"
+
+
+def test_csv_parses_metadata_and_embedding_columns() -> None:
+    docs = _parse_csv_docs(
+        "id,text,metadata,embedding\n"
+        'doc1,hello,"{""topic"": ""ml""}","[0.1, 0.2]"\n'
+        "doc2,world,,\n"
+    )
+    assert docs[0].metadata == {"topic": "ml"}
+    assert docs[0].embedding == pytest.approx([0.1, 0.2])
+    assert docs[1].metadata is None
+    assert docs[1].embedding is None
+
+
+def test_csv_header_is_trimmed() -> None:
+    docs = _parse_csv_docs("id , text \ndoc1,hello\n")
+    assert docs[0].id == "doc1"
+    assert docs[0].text == "hello"
+
+
+def test_csv_missing_column_reports_header_error() -> None:
+    with pytest.raises(typer.BadParameter) as exc:
+        _parse_csv_docs("id,body\ndoc1,hello\n")
+    assert "text" in str(exc.value)
+
+
+def test_csv_empty_content_is_rejected() -> None:
+    with pytest.raises(typer.BadParameter):
+        _parse_csv_docs("")
+
+
+def test_csv_short_row_is_rejected() -> None:
+    with pytest.raises(typer.BadParameter) as exc:
+        _parse_csv_docs("id,text\ndoc1\n")
+    assert "line 2" in str(exc.value)
+
+
+def test_csv_invalid_metadata_reports_file_line_number() -> None:
+    with pytest.raises(typer.BadParameter) as exc:
+        _parse_csv_docs('id,text,metadata\ndoc1,"a\nb",not-json\n')
+    assert "line 3" in str(exc.value)