Error loading pipeline yaml with empty strings

[danielfleischer]

YAML supports empty strings as "" or ''. I need to provide an empty string as a parameter to one of the classes. When loading a pipeline from that YAML file using haystack.Pipeline.load_from_yaml I get an exception. This is due to the following code:

haystack/haystack/pipelines/config.py

Line 136 in af06519

if not valid_regex.match(str(pipeline_config)):

because the regex for valid values does not recognize the (pythonic) empty string; top of the file contains its definition

VALID_VALUE_REGEX = re.compile(r"^[-\w/\\.:* \[\]]+$")

Error message

haystack.errors.PipelineConfigError: '' is not a valid variable name or value. Use alphanumeric characters or dash, underscore and colon only.`

Expected behavior
Empty string have their uses and are valid JSON/YAML/Python objects and should be valid as values.

System:

• Haystack version (commit or version number): 1.11.0

Thanks!

[danielfleischer]

Suggestion: change the pattern to re.compile(r"^[-\w/\\.:* \[\]]*$").

[mayankjobanputra]

Hi @danielfleischer, Thanks for pointing out this issue. We are looking into it. Can you also share your exact use case? so that it can help us reproduce this issue?

[danielfleischer]

Sure. Simplest example is creating an elasticsearch component:

components:
- name: Store
  params:
    host: localhost
    index: text_1
    password: ""
    port: 80
    username: ""                  
  type: ElasticsearchDocumentStore
- name: Retriever
  params:
    document_store: Store
  type: ElasticsearchRetriever

pipelines:
- name: my_pipeline
  nodes:
  - inputs:
    - Query
    name: Retriever

version: 1.11.0

Thanks for looking into it.

[masci]

Going to add this to the backlog.

@danielfleischer your proposed solution seems the right one, feel free to go ahead and open a PR to speed up things.

[ZanSara]

@masci fyi about the origin of that regex: #2214 (comment)