Print the repr these feature examples actually produce - #8564
Print the repr these feature examples actually produce#8564vineethsaivs wants to merge 2 commits into
Conversation
Image, Pdf, Nifti and Video all declare `id` with repr=False, so none of them can print the `X(decode=True, id=None)` their examples show. Image also has a `mode` field the example omits, and Video has five more fields that do print. Also close the string literal in Video's last example, which made the line unparseable, and add the `>>> ds[0]["image"]` line Image's last block needs to produce the dict under it, the way pdf.py and nifti.py already do.
Same drift as the four classes already in this PR, in two files it did not reach.
iterable_dataset.py prints Audio with mono=True, id=None. Audio has no mono
field at all, so the line is not just stale, it cannot be constructed, and id
carries repr=False. It also prints Image(mode=None, decode=False, id=None).
features.py Features.flatten prints List(Value('int32'), id=None) twice, and
List keeps id out of its repr as well.
Features.reorder_fields_as is a different miss. Its example builds f1 and f2
from plain dicts, but its own comment says "here List is defined at the root
level", it imports List without using it, and the documented output is
{'root': List({'b': ..., 'a': ...})}. The List wrapper was dropped from the two
constructor lines at some point and the output was never updated. Putting it
back makes the example agree with its comment, its import and its output.
Checked against main:
Audio(sampling_rate=8000) -> Audio(sampling_rate=8000, decode=True, num_channels=None, stream_index=None)
Image(mode=None, decode=False) -> Image(mode=None, decode=False)
List(Value('int32')) -> List(Value('int32'))
f1.reorder_fields_as(f2) -> {'root': List({'b': Value('string'), 'a': Value('string')})}
and f1.reorder_fields_as(f2).type == f2.type still holds, so the closing assert
in that example is unchanged.
|
Pushed 101bb1a: the same drift in two files this PR had not reached, plus one example that lost its point.
Every line here was run against main, not reasoned about. |
1 participant
What breaks
Four feature classes show a repr in their
Examplesblock that they cannot print.Image,Pdf,NiftiandVideoall declareid: Optional[str] = field(default=None, repr=False), soid=Nonenever appears. On top of thatImagehas amodefield the example omits, andVideohas five more fields that do print.What they actually print:
Image(decode=True, id=None)Image(mode=None, decode=True)Pdf(decode=True, id=None)Pdf(decode=True)Nifti(decode=True, id=None)Nifti(decode=True)Video(decode=True, id=None)Video(decode=True, stream_index=None, dimension_order='NCHW', num_ffmpeg_threads=1, device='cpu', seek_mode='exact')Two more in the same blocks:
video.pyends with...[0]["video](unterminated string), so that line does not parse.image.py's last block prints a dict directly afterds = ds.cast_column(...), which is an assignment and outputs nothing.pdf.pyandnifti.pyhave the>>> ds[0]["image"]line it is missing.What changed
Docs only, no code. Each shown repr replaced with what the class prints, the string literal closed, and the missing line added.
Verification
By execution, not by eye: the four classes were built from this branch's source and their
repr()compared against the docstring text, and every>>>line in the four blocks wascompile()d.4 wrong reprs and 1 line that does not compile on main, 0 and 0 after.
ruff format --checkandruff checkclean on all four files.