minform 0.1.3

WTForms/struct integration to validate and serialize to packed buffers of binary data.

The declarative ease of WTForms, the small data footprint of struct.

For more detailed examples, read the full docs at https://minform.readthedocs.org.

Why does Minform exist?

Let’s talk about data size.

Say you have a fleet of devices with cell modems that report to a web server. Here’s an example of a data packet struct you might use:

struct sensor_data {
    char version[8];
    char serial[12];
    int32_t latitude;    // fixed-point * 100000
    int32_t longitude;   // fixed-point * 100000
    int16_t temperature; // fixed-point * 100
    uint16_t pings;
    uint8_t battery_pct;
};

Let’s say they’re reporting their data in JSON format, because that’s lighter than XML but still coherent to most server frameworks. Here’s a data packet:

{
    "version": "1.0",
    "serial": "DEADBEEF",
    "latitude": 4071270,
    "longitude": -7400590,
    "temperature": 3200,
    "pings": 123,
    "battery_pct": 62
}

If you take out all the whitespace, that’s 144 bytes. (ASCII encoding). Maybe that’s all you need, but maybe you need to store billions or trillions of these little guys. Worse, maybe you need to pay through the tear ducts for cellular data. It would be nice for that data to be smaller, and to be predictably sized.

Besides, let’s face it: serializing to JSON from C is a pain in the patoot. Depending on the library, you could be looking at ten or twenty lines of code (or a truly epic sprintf), just to serialize a record.

On the Python side, we can probably just use a form library like WTForms to validate incoming data, but we’ve already paid a price to make that data server-friendly.

What can Minform do for me?

Let’s build a Minform form to handle incoming sensor data.

from minform import *

class SensorData(BinaryForm):

    order = LITTLE_ENDIAN  # let's say our devices are little-endian
    version = BytesField(max_length=8, length=AUTOMATIC)
    serial = BytesField(max_length=12, length=AUTOMATIC)
    latitude = Int32Field()
    longitude = Int32Field()
    temperature = Int16Field()
    pings = UInt16Field()
    battery_pct = UInt8Field()
    maintainer = BytesField(max_length=3, length=FIXED)
    padding = BlankBytes(3)

Here’s the C code that will serialize your structure:

#include <string.h>

void serialize(char *send_buffer, struct sensor_data data) {
    memcpy(send_buffer, &data, sizeof(struct sensor_data));
}

And here’s the Python that will receive it:

form = SensorData.unpack(serialized_data)

That serialized record is 36 bytes. 36 on the wire, 36 in a file. You may need to tweak the form definition, depending on your C compiler and the target architecture, but Minforms gives you the tools to cope with padding bytes, and even mixed byte ordering.

Let’s fill in some gaps

Minforms are an awful lot like WTForms: you subclass minform.BinaryForm, and add BinaryFields as class properties. Here’s another quick example:

import minform

class MyForm(minform.BinaryForm):
    '''
    This is a subclass of wtforms.Form: you can validate data with it,
    construct it from an HTML form, extract the data as a Python dict, etc.
    '''
    first_name = minform.BytesField('First Name', max_length=10)
    last_name = minform.BytesField('Last Name', max_length=10)
    age = minform.UInt8Field('Age')

#               first_name (10)          last_name (10)           age (1)
packed_data = b'David\x00\x00\x00\x00\x00Donna\x00\x00\x00\x00\x00\x18'
form = MyForm.unpack(packed_data)
assert form.data == {
    'first_name': b'David',
    'last_name': b'Donna',
    'age': 24,
}

next_form = MyForm(first_name=b'Foo', last_name=b'Barsson', age=100)
packed = next_form.pack()
assert packed == b'Foo\x00\x00\x00\x00\x00\x00\x00Barsson\x00\x00\x00\x64'

Because the library is built on struct, binary serializations of a form’s data are in fixed-length buffers. This makes them easier to store, and easy to map onto relatively naive serializations of C structs. It also allows for clear documentation of the binary format, because the data maps predictably onto different positions in a packed buffer.

Compound BinaryFields allow you to create nested structures that still serialize into flat buffers.

class MyBigBadForm(minform.BinaryForm):
    """
    This is taking a turn for campy criminality.
    """
    riches = minform.Int16Field()
    goons = minform.BinaryFieldList(Person, max_entries=4, length=minform.EXPLICIT)

squad = MyBigBadForm(riches=55223, goons=[
    {'first_name': 'Joey', 'last_name': 'Schmoey', 'age': 32},
    {'first_name': 'Manny', 'last_name': 'The Man', 'age': 40},
    {'first_name': 'Gerta', 'last_name': 'Goethe', 'age': 52},
])
assert squad.pack() == (b'\xd7\xb7' +                                  # riches
                        b'\x03' +                                      # goons prefix
                        b'Joey\0\0\0\0\0\0Schmoey\0\0\0\x32' +         # goons[0]
                        b'Manny\0\0\0\0\0The Man\0\0\0\x40' +          # goons[1]
                        b'Gerta\0\0\0\0\0Goethe\0\0\0\0\x52' +         # goons[2]
                        b'\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0') # goons[3]

Even with an entire set of blank bytes for goons[3], that’s 87 bytes, vs 185 for the JSON representation.

History

0.1.0 (10/28/2015)

Initial release