keep the extracted affiliations if none found from consolidation.

[Aazhar]

No description provided.

[coveralls]

Coverage increased (+0.3%) to 37.314% when pulling f1c3cf9 on keep_affiliations_after_consolidation into c7b4b20 on master.

[lfoppiano]

Would be possible to rename bibo and bib with more appropriate names?
Might be a lot of work but this class deserves to have a unit test perhaps :-)

See my other comment below ;-)

[lfoppiano]

what about the initial is with a dot? so it has a length of 2?
What about the initial is present in the aut2 instead of aut? shouldnt you check that condition too?

[Aazhar]

yes actually we should compare the initials for the firstnames either for aut2 or aut , , so the check would be either they are equal or they have the same initial letter in their firstname , WDYT ?

[lfoppiano]

the check (whch was like this before, so I might be wrong) first check if aut has length 1, and if so it compares aut with the first character of aut2

what if aut does not have length 1, but aut2 is just the initial?

[Aazhar]

yes I didn't change this check actually, it was like this..
basically if it is not the same firstname or aut firstname has more than one character then it 's not considered the same author form

[Aazhar]

what I suggest is for firstnames to compare only the first char (initials)

[lfoppiano]

OK

I noticed another thing, at line 4205, there is bib.setFullAuthors(bibo.getFullAuthors()); so the authors are copied anyway from the second biblio... or should I better be in bed?

[lfoppiano]

I added a couple of tests, there is an ignored one that is failing and I'm wondering what is the rule here

[Aazhar]

yes the data about authors coming from consolidation is considered more accurate than the parsed one,
in this PR I would like to keep the affiliations if none fetched from the consolidation which seems to be more frequent

[Aazhar]

regarding the test, so far the affiliations from crossref are just raw and not deserialised, so maybe we could keep the affiliations parsed from the pdf ? I'm not sure how good are the affiliations from crossref actually

[kermitt2]

So as explained on Mattermost, I updated this part with the new Person object deduplication, and this part becomes simply now:

       if (bibo.getFullAuthors() != null) {
            if (CollectionUtils.isEmpty(bib.getFullAuthors()))
                bib.setFullAuthors(bibo.getFullAuthors());
            else {
                // we have the complete list of authors so we can take them from the second
                // biblio item and merge some possible extra from the first when a match is 
                // reliable
                List<Person> thePersons = bib.getFullAuthors();
                thePersons.addAll(bibo.getFullAuthors());
                Person.deduplicate(thePersons);
            }
        }

and the deduplication handles the merging of affiliation.

[kermitt2]

With this approach, it works fine when authors are well recognized initially, but when the authors are badly extracted from the PDF, we will have as authors at the same time the erroneous extraction and the correct consolidated ones. So maybe this could be refined to prioritize the consolidated authors?

[kermitt2]

ok too many authors added with this approach, I'll try to find something safer...