typedcsv

Typed CSVs via header-embedded types (sigils or :type) plus optional header-embedded validation — stdlib-only, Python 3.10+.

Default extensions (by convention):

• .tcsv for comma-separated typed CSV
• .ttsv for tab-separated typed TSV

The core lives in a single module and is copy-pasteable into projects.

pip install typedcsv-lib

---

Header typing

Declare types using either suffix sigils or explicit :type (not both on the same column).

Type	Sigil (suffix)	Explicit
int	#	:int
float	%	:float
bool	?	:bool
datetime	@	:datetime
str	$	:str

Untyped columns default to str.

Optional constraint sigils (suffix, after the type sigil):

Constraint	Sigil
required	!
unique	+

Order is flexible, but the recommended style is: type + optional unique + optional required.

Examples:

• id#+ (unique int)
• email$! (required str)
• slug$+! (unique + required str)

Logical column names are the header names with the type marker removed:

• age# becomes key "age"
• created:datetime becomes key "created"

---

Validators

Add an optional validator clause after the type marker:

age# [min=0 max=120]
ratio% [min=0 max=1]
status$ [in=OPEN|CLOSED|PENDING]
code$ [re=^[A-Z]{3}\d{2}$]
created@ [min=2020-01-01T00:00:00 max=2030-12-31T23:59:59]
email:str [required unique]

Notes:

• Validators are space-separated key=value pairs inside [ ... ].
• required and unique are flag-only validators (no =).
• re= uses Python re.fullmatch.
• in= uses | as separator.
• Unknown validator keys raise an error.

---

Missing values (nullable by default)

• An empty cell ("") is missing.
• For str columns, missing stays "".
• For non-str columns, missing becomes None.
• Missing values skip validation.
• required (sigil ! or validator) rejects missing values.
• unique ignores missing values (multiple empty cells are allowed).

---

Reading

import typedcsv

with open("data.tcsv", newline="") as f:
    for row in typedcsv.DictReader(f):
        print(row)

Example CSV:

id#,name$,active?,created@
1,Alice,true,2021-05-01T12:30:00
2,Bob,false,

Produces:

{'id': 1, 'name': 'Alice', 'active': True, 'created': datetime(2021, 5, 1, 12, 30)}
{'id': 2, 'name': 'Bob', 'active': False, 'created': None}

---

Writing (canonical formatting)

• None → empty cell
• bool → true / false
• datetime → isoformat()
• float → repr(f)
• Header preserved exactly as provided to DictWriter(fieldnames=...)

import typedcsv
from datetime import datetime

rows = [
    {"id": 1, "name": "Alice", "active": True, "created": datetime(2021, 5, 1, 12, 30)},
    {"id": 2, "name": "Bob", "active": False, "created": None},
]

with open("out.tcsv", "w", newline="") as f:
    w = typedcsv.DictWriter(f, fieldnames=["id#", "name$", "active?", "created@"])
    w.writeheader()
    w.writerows(rows)

---

More examples

Validators with quoted values

name$ [in="Alice Smith"|Bob]

Regex validation (fullmatch)

code$ [re=^[A-Z]{3}\d{2}$]
ABC12

Type inference for untyped columns

import typedcsv
import io

data = "a,b\n1,true\n2,false\n"
rows = list(typedcsv.DictReader(io.StringIO(data), infer_types=True))

---

Errors

Parsing/validation failures raise TypedCSVError with context:

• row (1-based; header row is 1)
• col (0-based)
• column (logical name)
• header (raw header cell)
• value (raw cell)
• reason (short message)

---

API reference (csv-compatible)

typedcsv mirrors Python's csv module API and is designed to be a drop-in replacement where you want typed rows.

• typedcsv.reader(f, ...) → yields typed list rows (header consumed)
• typedcsv.DictReader(f, ...) → yields typed dict rows keyed by logical names (header consumed)
• typedcsv.writer(f, ...) → returns a standard csv.writer
• typedcsv.DictWriter(f, fieldnames, ...) → writes typed dict rows with canonical formatting

---