This PR overhauls the dialect selection algorithm to fix #128. In that issue, the problem is that there are equal numbers of dialects where the field separator is ; and ; (that latter one has a trailing space). Currently, whatever is first in the file wins. One possible solution would be to use a majority-wins dialect selection using the first checklines features. But specifically in #128 that won't help, because there are equal numbers of features. So this PR focuses more on features with a larger number of attributes keys when choosing a dialect, under the assumption that the dialect will be easier to accurately detect in features with more key/val pairs.
Turns out there is now an issue with building the docs, specifically in Python 3.8. Still not sure what's going on there, but now the docs are only built and doctested with 3.6, 3.7, and 3.9. Built docs are also uploaded as an artifact via GitHub Actions.
This PR overhauls the dialect selection algorithm to fix #128. In that issue, the problem is that there are equal numbers of dialects where the field separator is
;and;(that latter one has a trailing space). Currently, whatever is first in the file wins. One possible solution would be to use a majority-wins dialect selection using the firstchecklinesfeatures. But specifically in #128 that won't help, because there are equal numbers of features. So this PR focuses more on features with a larger number of attributes keys when choosing a dialect, under the assumption that the dialect will be easier to accurately detect in features with more key/val pairs.Turns out there is now an issue with building the docs, specifically in Python 3.8. Still not sure what's going on there, but now the docs are only built and doctested with 3.6, 3.7, and 3.9. Built docs are also uploaded as an artifact via GitHub Actions.