csv.Sniffer returns a dialect that drops empty space-delimited fields #161400

Bug description:

With a space as the delimiter, csv.Sniffer.sniff() returns skipinitialspace=True for a sample with an unquoted empty field when one other row is shorter, so the sniffed dialect loses the empty field:

import csv, io
data = ('name age city\n'
        'alice 30 paris\n'
        'bob  london\n'
        'carol 25 "new york"\n'
        'dave 40 berlin\n'
        'erin 35\n')
dialect = csv.Sniffer().sniff(data)
print(dialect.skipinitialspace, list(csv.reader(io.StringIO(data), dialect))[2])

On main:

True ['bob', 'london']

On 3.15.0:

False ['bob', '', 'london']

The shorter row can also be the last row of a cut-off sample, as in the documented sniff(csvfile.read(1024)): with the last row 'erin 35 rome\n', sniffing data[:-6] gives the same result. The docs say sniff() picks the combination "with the most consistent number of fields", which here is skipinitialspace=False. This came with gh-154744.

Expected: skipinitialspace=False, as in 3.15.

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-161400: Fix csv.Sniffer dropping empty space-delimited fields #161401