Misleading error when getting empty span's `text` attribute

Slicing empty span from Doc works without errors (e.g. doc[:0] or doc[3:3]), but attempt to use text attribute of a resulting empty span throws IndexError: [E201] Span index out of range. However, text_with_ws, attribute works fine, so it’ll do as a workaround for my usage, but that seems like a very unintuitive behaviour.

How to reproduce the behaviour

import spacy

nlp = spacy.load('en')
doc = nlp('something')
print(doc[:0].text_with_ws)  # prints empty string, as expected
print(doc[:0].text)  # throws IndexError: [E201] Span index out of range

Your Environment

• spaCy version: 2.3.5
• Platform: Linux-5.8.0-38-generic-x86_64-with-glibc2.29
• Python version: 3.8.5
• Models: en

[adrianeboyd]

I think it’s both: the span bound check in __getitem__ leads to span.text failing for a 0-length span when it didn’t in earlier versions because you used to be able to access span[-1] even for 0-length spans (it would return the token before the span). That’s the only thing that changed between v2.3.2 and v2.3.5, and the lack of the bounds check was masking the .text bug.

[svlandeg]

Hi @mzeidhassan : Could be, yes, I’ll leave a message on that thread to clarify.