Bug report

csv.Sniffer detects skipinitialspace by a simple rule: the spaces are a part of the format if every field which follows a delimiter starts with a space. It is wrong in both directions.

A space before a quoted field is not detected. A quote is only a quote at the very start of a field, so not skipping the space splits the field:

>>> sample = 'a, "b,c"\nd,e\nf,g\n'
>>> dialect = csv.Sniffer().sniff(sample)
>>> dialect.skipinitialspace
False
>>> list(csv.reader(io.StringIO(sample), dialect))
[['a', ' "b', 'c"'], ['d', 'e'], ['f', 'g']]

A space which cannot be skipped is counted as evidence. Spaces inside quoted or escaped fields are never skipped, so they say nothing about the padding:

>>> csv.Sniffer().sniff('a," b"\nc, d\n').skipinitialspace
True
>>> csv.Sniffer().sniff('a,\\ b\\,c\nd, e\n').skipinitialspace
True

In both samples the first field is not padded at all, so the sample gives evidence against the padding, not for it.

Linked PRs

• gh-154744: Improve detection of skipinitialspace in csv.Sniffer #154745