Read three more shapes of code as literal blocks

[tamnd]

Closes #50.

L01 reports an entry whose translation is identical to its source. 5 of its 9 findings were entries there was nothing to translate in, so being identical was the correct answer and the check was reporting a classifier bug rather than a translation bug.

The rule was missing three shapes.

A block opener quoted without its body. There is no second line to be indented and no call to recognise, so every rule read it as a sentence ending in a colon.

def f(pos1, pos2, /, pos_or_kwd, *, kwd1, kwd2):
case (Point(x1, y1), Point(x2, y2) as p2): ...
class Bag[T]: ...

A line with a comment on the end. The code is there, and the English on the end of it stopped any rule from seeing it.

parrot(1000)                                          # 1 positional argument
parrot()                     # required argument missing

A call whose arguments are quoted strings. What separates a call from a sentence here is spacing, code putting a space after a comma and nowhere else, and the spaces inside the quotes were being counted.

parrot('a million', 'bereft of life', 'jump')
Popen(["/usr/bin/git", "commit", "-m", "Fixes a bug."])

The keyword list

It stops where English starts. if, for, while, with and else open blocks and are also ordinary words, and a colon is how the documentation introduces a list. The wide list was measured before it was dropped: it reaches 4 more entries and 2 sentences, while a positional argument could be created like:: and if it is 3, implements::. Those two are in the tests as negatives. The comment rule wants two spaces before the hash, which is how PEP 8 says to write one and how every one of these in the corpus is written; one space reaches 2 more entries and reads a hash anywhere in a sentence as a comment.

Measured

	before	after
prose	75,660	75,593
literal_block	2,856	2,923
batches on a full run	2,802	2,801
human segments	1,299	1,294

67 entries move, 0.08 per cent of the corpus. That is small enough to read one at a time, and all 67 were read before this was written: no prose among them.

The 5 human segments that go are exactly the 5 L01 findings. They were translated by being copied, because copying was correct. human says who typed a string, not that the string wanted typing.

Checks

make check green, 1,399 tests, 97.45 per cent coverage, classify.py at 100 per cent. H03 clean.