[1.2.2][Ubuntu] Sentiment analysis raises an exception with an empty input

[Armillus]

Describe the bug
Sentiments analysis on an empty input raises the following exception: IndexError: tuple index out of range.
I didn't see this error mentionned in current issues.

The pipeline used to obtain this result:

stanza.Pipeline(lang='en', processors='tokenize,sentiment')

The actual result:

2021-07-20 14:21:11 INFO: Loading these models for language: en (English):
========================
| Processor | Package  |
------------------------
| tokenize  | combined |
| sentiment | sstplus  |
========================

2021-07-20 14:21:11 INFO: Use device: gpu
2021-07-20 14:21:11 INFO: Loading: tokenize
2021-07-20 14:21:18 INFO: Loading: sentiment
2021-07-20 14:21:18 INFO: Done loading processors!
Traceback (most recent call last):
  File "test.py", line 4, in <module>
    doc = nlp('')
  File "/home/amillo/.local/lib/python3.8/site-packages/stanza/pipeline/core.py", line 253, in __call__
    doc = self.process(doc)
  File "/home/amillo/.local/lib/python3.8/site-packages/stanza/pipeline/core.py", line 247, in process
    doc = process(doc)
  File "/home/amillo/.local/lib/python3.8/site-packages/stanza/pipeline/sentiment_processor.py", line 53, in process
    labels = cnn_classifier.label_text(self._model, text, batch_size=self._batch_size)
  File "/home/amillo/.local/lib/python3.8/site-packages/stanza/models/classifiers/cnn_classifier.py", line 449, in label_text
    text, orig_idx = sort_with_indices(text, key=len, reverse=True)
  File "/home/amillo/.local/lib/python3.8/site-packages/stanza/models/common/utils.py", line 223, in sort_with_indices
    return result[1], result[0]
IndexError: tuple index out of range

To Reproduce
To get the error, you can try this basic snippet of code, which is a slightly modified example from the official documentation.
However, this example assumes that you've already downloaded the corresponding model.

import stanza

nlp = stanza.Pipeline(lang='en', processors='tokenize,sentiment')
doc = nlp('')
for i, sentence in enumerate(doc.sentences):
    print(i, sentence.sentiment)

Expected behavior
The expected result would be to get back a working document, without any exception.
The expected output (from my limited experience) should be:

2021-07-20 14:06:44 INFO: Loading these models for language: en (English):
========================
| Processor | Package  |
------------------------
| tokenize  | combined |
| sentiment | sstplus  |
========================

2021-07-20 14:06:44 INFO: Use device: gpu
2021-07-20 14:06:44 INFO: Loading: tokenize
2021-07-20 14:06:52 INFO: Loading: sentiment
2021-07-20 14:06:52 INFO: Done loading processors!

Environment (please complete the following information):

• OS: Ubuntu (WSL 2 - 21H2 - Build 22000.71)
• Python version: Python 3.8.10
• Stanza version: 1.2.2

Additional context
I don't know if this error is happening with older versions of Stanza as well, since I didn't try to feed any pipeline with an empty input before. Nevertheless, I've never met this error with Stanza 1.2.0.

[AngledLuffa]

Oops! This is why I'll never write code for the space shuttle.

47889e3

I can push a version to testpypi if you find yourself frequently running into this

[Armillus]

No problem, it can be fixed quickly by checking that the input is empty before calling the pipeline, so I can wait until the next version :)

Thank you for your quick reaction time and for this excellent library!

[flatplate]

I run into the same problem with a string with only a space character: nlp(" "). Does this commit fix that also?

[AngledLuffa]

Yes!