autotui

autotui

This uses type hints to convert NamedTuple's (short struct-like classes) to JSON/YAML, and back to python objects.

It also wraps prompt_toolkit to prompt the user and validate the input for common types, and is extendible to whatever types you want.

• Supported Types
• Install
• Usage
  • Enabling Options
  • Partial prompts
  • YAML
  • Picking items
  • Editing items
  • Custom types/prompting
• Testing

Supported Types

This has built-ins to prompt, validate and serialize:

• int
• float
• bool
• str
• datetime
• Enum
• Decimal
• Optional[<type>] (or <type> | None)
• List[<type>] (or list[<type>])
• Set[<type>] (or set[<type>])
• other NamedTuples (recursively)

I wrote this so that I don't have to repeatedly write boilerplate-y python code to validate/serialize/deserialize data. As a more extensive example of its usage, you can see my ttally repo, which I use to track things like calories/water etc...

Install

This requires python3.8+, specifically for modern typing support.

To install with pip, run:

pip install autotui

Usage

As an example, if I want to log whenever I drink water to a file:

from datetime import datetime
from typing import NamedTuple

from autotui.shortcuts import load_prompt_and_writeback

class Water(NamedTuple):
    at: datetime
    glass_count: float

if __name__ == "__main__":
    load_prompt_and_writeback(Water, "~/.local/share/water.json")

Which, after running a few times, would create:

~/.local/share/water.json

[
  {
    "at": 1598856786,
    "glass_count": 2.0
  },
  {
    "at": 1598856800,
    "glass_count": 1.0
  }
]

(datetimes are serialized into epoch time)

If I want to load the values back into python, its just:

from autotui.shortcuts import load_from

class Water(NamedTuple):
    #... (same as above)

if __name__ == "__main__":
    print(load_from(Water, "~/.local/share/water.json"))

#[Water(at=datetime.datetime(2020, 8, 31, 6, 53, 6, tzinfo=datetime.timezone.utc), glass_count=2.0),
# Water(at=datetime.datetime(2020, 8, 31, 6, 53, 20, tzinfo=datetime.timezone.utc), glass_count=1.0)]

A lot of my usage of this only ever uses 3 functions in the autotui.shortcuts module; dump_to to dump a sequence of my NamedTuples to a file, load_from to do the opposite, and load_prompt_and_writeback, to load values in, prompt me, and write back to the file.

Enabling Options

Some options/features can be enabled using global environment variables, or by using a contextmanager to temporarily enable certain prompts/features.

As an example, there are two versions of the datetime prompt

• The one you see above using a dialog
• A live version which displays the parsed datetime while typing. Since that can cause some lag, it can be enabled by setting the LIVE_DATETIME option.

You can enable that by:

• setting the AUTOTUI_LIVE_DATETIME (prefix the name of the option with AUTOTUI_) environment variable, e.g., add export AUTOTUI_LIVE_DATETIME=1 to your .bashrc/.zshrc
• using the options contextmanager:

import autotui

with autotui.options("LIVE_DATETIME"):
    autotui.prompt_namedtuple(...)

Options:

• LIVE_DATETIME: Enables the live datetime prompt
• CONVERT_UNKNOWN_ENUM_TO_NONE: If an enum value is not found on the enumeration (e.g. you remove some enum value), convert it to None instead of raising a ValueError
• ENUM_FZF: Use fzf to prompt for enums
• CLICK_PROMPT - Where possible, use click to prompt for values instead of prompt_toolkit

Partial prompts

If you want to prompt for only a few fields, you can supply the attr_use_values or type_use_values to supply default values:

# water-now script -- set any datetime values to now
from datetime import datetime
from typing import NamedTuple

from autotui import prompt_namedtuple
from autotui.shortcuts import load_prompt_and_writeback

class Water(NamedTuple):
    at: datetime
    glass_count: float

load_prompt_and_writeback(Water, "./water.json", type_use_values={datetime: datetime.now()})
# or specify it with a function (don't call datetime.now, just pass the function)
# so its called when its needed
val = prompt_namedtuple(Water, attr_use_values={"at": datetime.now})

Since you can specify a function to either of those arguments -- you're free to write a completely custom prompt function to prompt/grab data for that field however you want

For example, to prompt for strings by opening vim instead:

from datetime import datetime
from typing import NamedTuple, List, Optional

from autotui.shortcuts import load_prompt_and_writeback

import click


def edit_in_vim() -> str:
    m = click.edit(text=None, editor="vim")
    return m if m is None else m.strip()


class JournalEntry(NamedTuple):
    creation_date: datetime
    tags: Optional[List[str]]  # one or more tags to tag this journal entry with
    content: str


if __name__ == "__main__":
    load_prompt_and_writeback(
        JournalEntry,
        "~/Documents/journal.json",
        attr_use_values={"content": edit_in_vim},
    )

Can also define those as a staticmethod on the class, so you don't have to pass around the extra state:

class JournalEntry(NamedTuple):
    ...

    @staticmethod
    def attr_use_values() -> Dict:
        return {"content": edit_in_vim}


# pulls attr_use_values from the function
prompt_namedtuple(JournalEntry, "~/Documents/journal.json")

Yaml

Since YAML is a superset of JSON, this can also be used with YAML files. autotui.shortcuts will automatically decode/write to YAML files based on the file extension.

# using the water example above
if __name__ == "__main__":
    load_prompt_and_writeback(Water, "~/.local/share/water.yaml")

Results in:

- at: 1645840523
  glass_count: 1.0
- at: 1645839340
  glass_count: 1.0

You can also pass format="yaml" to the namedtuple_sequence_dumps/namedtuple_sequence_loads functions (shown below)

Picking

This has a basic fzf picker using pyfzf-iter, which lets you pick one item from a list/iterator:

from autotui import pick_namedtuple
from autotui.shortcuts import load_from

picked = pick_namedtuple(load_from(Water, "~/.local/share/water.json"))
print(picked)

To install the required dependencies, install fzf and pip install 'autotui[pick]'

Editing

This also provides a basic editor, which lets you edit a single field of a NamedTuple.

$ python3 ./examples/edit.py
Water(at=datetime.datetime(2023, 3, 5, 18, 55, 59, 519320), glass_count=1)
Which field to edit:

	1. at
	2. glass_count

'glass_count' (float) > 30
Water(at=datetime.datetime(2023, 3, 5, 18, 55, 59, 519320), glass_count=30.0)

In python:

from autotui.edit import edit_namedtuple

water = edit_namedtuple(water, print_namedtuple=True)
# can also 'loop', to edit multiple fields
water = edit_namedtuple(water, print_namedtuple=True, loop=True)

Any additional arguments to edit_namedtuple are passed to prompt_namedtuple, so you can specify type_validators to attr_validators to prompt in some custom way

To install, pip install 'autotui[edit]' or pip install click

Custom Types

If you want to support custom types, or specify a special way to serialize another NamedTuple recursively, you can specify type_validators, and type_[de]serializer to handle the validation, serialization, deserialization for that type/attribute name.

As a more complicated example, heres a validator for timedelta (duration of time), being entered as MM:SS, and the corresponding serializers.

# see examples/timedelta_serializer.py for imports

# handle validating the user input interactively
# can throw a ValueError
def _timedelta(user_input: str) -> timedelta:
    if len(user_input.strip()) == 0:
        raise ValueError("Not enough input!")
    minutes, _, seconds = user_input.partition(":")
    # could throw ValueError
    return timedelta(minutes=float(minutes), seconds=float(seconds))


# serializer for timedelta, converts to JSON-compatible integer
def to_seconds(t: timedelta) -> int:
    return int(t.total_seconds())


# deserializer from integer to timedelta
def from_seconds(seconds: int) -> timedelta:
    return timedelta(seconds=seconds)


# The data we want to persist to the file
class Action(NamedTuple):
    name: str
    duration: timedelta


# AutoHandler describes what function to use to validate
# user input, and which errors to wrap while validating
timedelta_handler = AutoHandler(
    func=_timedelta,  # accepts the string the user is typing as input
    catch_errors=[ValueError],
)

# Note: validators are of type
# Dict[Type, AutoHandler]
# serializer/deserializers are
# Dict[Type, Callable]
# the Callable accepts one argument,
# which is either the python value being serialized
# or the JSON value being deserialized

# use the validator to prompt the user for the NamedTuple data
# name: str automatically uses a generic string prompt
# duration: timedelta gets handled by the type_validator
a = prompt_namedtuple(
    Action,
    type_validators={
        timedelta: timedelta_handler,
    },
)


# Note: this specifies timedelta as the type,
# not int. It uses what the NamedTuple
# specifies as the type for that field, not
# the type of the value that's loaded from JSON

# dump to JSON
a_str: str = namedtuple_sequence_dumps(
    [a],
    type_serializers={
        timedelta: to_seconds,
    },
    indent=None,
)

# load from JSON
a_load = namedtuple_sequence_loads(
    a_str,
    to=Action,
    type_deserializers={
        timedelta: from_seconds,
    },
)[0]

# can also specify with attributes instead of types
a_load2 = namedtuple_sequence_loads(
    a_str,
    to=Action,
    attr_deserializers={
        "duration": from_seconds,
    },
)[0]

print(a)
print(a_str)
print(a_load)
print(a_load2)

Output:

$ python3 ./examples/timedelta_serializer.py
'name' (str) > on the bus
'duration' (_timedelta) > 30:00
Action(name='on the bus', duration=datetime.timedelta(seconds=1800))
[{"name": "on the bus", "duration": 1800}]
Action(name='on the bus', duration=datetime.timedelta(seconds=1800))
Action(name='on the bus', duration=datetime.timedelta(seconds=1800))

The general philosophy I've taken for serialization and deserialization is send a warning if the types aren't what the NamedTuple expects, but load the values anyways. If serialization can't serialize something, it warns, and if json.dump doesn't have a way to handle it, it throws an error. When deserializing, all values are loaded from their JSON primitives, and then converted into their corresponding python equivalents; If the value doesn't exist, it warns and sets it to None, if there's a deserializer supplied, it uses that. This is meant to help facilitate quick TUIs, I don't want to have to fight with it.

(If you know what you're doing and want to ignore those warnings, you can set the AUTOTUI_DISABLE_WARNINGS=1 environment variable)

There are lots of examples on how this is handled/edge-cases in the tests.

You can also take a look at the examples

Testing

git clone https://github.com/seanbreckenridge/autotui
cd ./autotui
pip install '.[testing]'
mypy ./autotui
pytest