Issue282 handling syntax errors in projects config

[juhoinkinen]

Raise ConfigurationException instead of full traceback in following situations:

1. when trying to use (train/suggest) a project without a backend set in projects.cfg:

   $ annif train nobackend -p tests/projects.cfg any_trainingdata.tsv 
   Error: Project 'nobackend': backend setting is missing
   

   This is in line with missing vocab error: "Error: Project 'maui-fi': vocab setting is missing".

2. when trying to use projects.cfg that has duplicated sections or entries:

   $ annif train any_backend -p tests/projects_invalid.cfg any_trainingdata.tsv 
   Error: Error: While reading from 'tests/projects_invalid.cfg' [line 10]: section 'duplicatedproject' already exists
   

This closes #282.

[osma]

Travis Python 3.5 build is failing, I think VW needs to be upgraded first. Maybe merge #296?

[osma]

failing build log: https://travis-ci.org/NatLibFi/Annif/jobs/559957266

[codecov]

Codecov Report

Merging #300 into master will increase coverage by 19.59%.
The diff coverage is 100%.

@@             Coverage Diff             @@
##           master     #300       +/-   ##
===========================================
+ Coverage    79.8%   99.39%   +19.59%     
===========================================
  Files          55       55               
  Lines        2842     2969      +127     
===========================================
+ Hits         2268     2951      +683     
+ Misses        574       18      -556

Impacted Files	Coverage Δ
annif/default_config.py	100% <100%> (ø)	⬆️
tests/test_project.py	100% <100%> (+7.75%)	⬆️
annif/project.py	100% <100%> (+1.21%)	⬆️
annif/corpus/subject.py	100% <0%> (ø)	⬆️
annif/corpus/types.py	100% <0%> (ø)	⬆️
annif/corpus/document.py	100% <0%> (ø)	⬆️
annif/corpus/convert.py	100% <0%> (ø)	⬆️
tests/test_corpus.py	100% <0%> (ø)	⬆️
annif/analyzer/__init__.py	92.59% <0%> (+3.7%)	⬆️
... and 12 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d13f369...47d4f2a. Read the comment docs.

[osma]

Looks good, but there is one further case that could be covered. If analyzer is left unspecified, I get this:

  File "/home/oisuomin/git/Annif/annif/project.py", line 188, in _create_vectorizer
    tokenizer=self.analyzer.tokenize_words)
AttributeError: 'NoneType' object has no attribute 'tokenize_words'

Should this case be covered too, or left for another issue/PR?

[juhoinkinen]

Looks good, but there is one further case that could be covered. If analyzer is left unspecified, I get this:

  File "/home/oisuomin/git/Annif/annif/project.py", line 188, in _create_vectorizer
    tokenizer=self.analyzer.tokenize_words)
AttributeError: 'NoneType' object has no attribute 'tokenize_words'

Should this case be covered too, or left for another issue/PR?

I'll check if this can be easily covered too and implement, otherwise I'll make a new issue about this.

[juhoinkinen]

Looks good, but there is one further case that could be covered. If analyzer is left unspecified, I get this:

  File "/home/oisuomin/git/Annif/annif/project.py", line 188, in _create_vectorizer
    tokenizer=self.analyzer.tokenize_words)
AttributeError: 'NoneType' object has no attribute 'tokenize_words'

Should this case be covered too, or left for another issue/PR?

Implemented raising ConfigurationError for this in the previous commit. For the test I changed the [noanalyzer] project to use TFIDF backend because dummy backend does not need analyzer and would not raise the ConfigurationError. That project was included for issue #148. I did not see any reason for the backend to remain dummy, but could @osma confirm?

[osma]

In general, changing [noanalyzer] to use tfidf should be fine. It was originally added in this commit where the intent was to verify that it's possible to have a project without an analyzer. But it might be necessary to add more project configurations e.g. "fasttext-noanalyzer", "vw_multi-noanalyzer" etc. to cover all the situations where an analyzer is needed but is missing.