もう事故らせない：PythonでCSV/JSON/Excelを安全に読み書きする実務レシピ

from pathlib import Path
import tempfile, os, logging
log = logging.getLogger(__name__)

def atomic\_write\_text(path: Path, text: str, encoding: str = "utf-8") -> None:
path.parent.mkdir(parents=True, exist\_ok=True)
with tempfile.NamedTemporaryFile("w", delete=False, dir=path.parent, encoding=encoding, newline="") as tmp:
tmp.write(text)
tmp\_path = Path(tmp.name)
os.replace(tmp\_path, path)

import pandas as pd
from pathlib import Path

p = Path("data/sales.csv")
usecols = \["date","store","sku","qty","price"]
dtypes  = {"store":"category","sku":"string","qty":"int32","price":"float32"}
na\_vals = \["","NA","N/A","null","-","--"]

df = pd.read\_csv(
p,
encoding="utf-8",     # ダメなら 'cp932'
usecols=usecols,      # メモリ節約
dtype=dtypes,         # 先頭ゼロ保護=string
na\_values=na\_vals,
parse\_dates=\["date"], # 日付→datetime
thousands=",",        # 3桁区切り除去
skip\_blank\_lines=True
)

agg = {}
for chunk in pd.read_csv(p, usecols=usecols, dtype=dtypes, parse_dates=["date"], chunksize=200_000):
    g = chunk.groupby("store").qty.sum()
    for k, v in g.items():
        agg[k] = agg.get(k, 0) + int(v)

# pandas → CSV
from pathlib import Path
out = Path("out/sales_clean.csv")
df.to_csv(out, index=False, encoding="utf-8", line_terminator="\n")

# 標準csvで厳密制御（空行問題の回避）

import csv
rows = \[\["sku","qty"],\["00123",10],\["A-9",5]]
with Path("out/items.csv").open("w", encoding="utf-8", newline="") as f:
w = csv.writer(f, quoting=csv.QUOTE\_MINIMAL)
w\.writerows(rows)

import json
from pathlib import Path

p = Path("data/config.json")
config = json.loads(p.read\_text(encoding="utf-8"))

# or:

with p.open("r", encoding="utf-8") as f:
config = json.load(f)

text = json.dumps(config, ensure\_ascii=False, indent=2, sort\_keys=True)
Path("out/config.json").write\_text(text, encoding="utf-8")

# NDJSON（1行=1JSON）

import pandas as pd
df = pd.read\_json("data/events.ndjson", lines=True)
Path("out/events.ndjson").write\_text(
"\n".join(df.to\_json(orient="records", lines=True).splitlines()),
encoding="utf-8"
)

import pandas as pd
from pathlib import Path

# 読む

p = Path("data/report.xlsx")
df = pd.read\_excel(p, sheet\_name="売上", usecols="A\:F",
dtype={"店舗":"category","商品コード":"string"})

# 書く：複数シート＋書式

with pd.ExcelWriter(Path("out/report.xlsx"), engine="xlsxwriter") as w:
df\_summary.to\_excel(w, sheet\_name="サマリ", index=False)
df\_detail.to\_excel(w, sheet\_name="明細", index=False)
wb  = w\.book
ff  = wb.add\_format({"bold": True})
num = wb.add\_format({"num\_format": "#,##0"})
ws1 = w\.sheets\["サマリ"]
ws1.set\_row(0, None, ff)
ws1.set\_column("D\:D", None, num)

import pandas as pd

df.to\_parquet("work/data.parquet", compression="zstd")
df2 = pd.read\_parquet("work/data.parquet")

def optimize\_dtypes(df: pd.DataFrame) -> pd.DataFrame:
for c in df.select\_dtypes(include="int64").columns:
df\[c] = pd.to\_numeric(df\[c], downcast="integer")
for c in df.select\_dtypes(include="float64").columns:
df\[c] = pd.to\_numeric(df\[c], downcast="float")
for c in df.select\_dtypes(include="object").columns:
if df\[c].nunique() / max(len(df),1) < 0.5:
df\[c] = df\[c].astype("category")
return df

from pathlib import Path
import pandas as pd

root = Path("data/daily")
files = sorted(root.glob("sales\_2025-\*.csv"))

frames = \[]
for p in files:
df = pd.read\_csv(p, usecols=\["date","store","sales"], parse\_dates=\["date"], dtype={"store":"string"})
df\["source"] = p.name
frames.append(df)
all\_df = pd.concat(frames, ignore\_index=True)

from pathlib import Path
import pandas as pd

class DataError(Exception): ...
class ExternalError(Exception): ...

def load\_csv\_safe(path: Path) -> pd.DataFrame:
try:
return pd.read\_csv(path, encoding="utf-8")
except UnicodeDecodeError as e:
raise DataError(f"encoding error: {path}") from e
except OSError as e:
raise ExternalError(f"file io error: {path}") from e

import pandas as pd

def test\_csv\_roundtrip(tmp\_path):
p = tmp\_path/"a.csv"
df = pd.DataFrame({"sku":\["001","002"],"qty":\[1,2]})
df.to\_csv(p, index=False, encoding="utf-8")
back = pd.read\_csv(p, dtype={"sku":"string"})
assert list(back.sku) == \["001","002"]