This allows to limit regions to certain classes if these are known
from the harvester context, if a given name always refers to municipality
or river or mountain.

This avoids a match if another same-class match exists, e.g. if a
populated place and a seat of an administrative division with
the same name exist, but not if the other match is an administrative
division or stream/river/lake instead.

An possible extension would be to limit alignment to a given class, e.g.
in gw_lu_bw, we are matching municipalites which in GeoName's approach
is a fourth-order administrative division for which we could filter.
This would however require a significantly extended internal API and
some modelling of the relevant classes.

Adam Reichold authored 5 days ago and Adam Reichold committed 2 days ago

Accept regression test changes due to internal changes to FxHashSet which yield different arbitrary but equally deduplication results

To be more precise, these changes are effected by upgrading rustc-hash from 2.1.0
to 2.1.1 which changes the string hashing and the fact that the four affected datasets
all have the same DOI even though their content is completely different.

Adam Reichold authored 5 days ago and Adam Reichold committed 2 days ago

This switches the TLS backend from OpenSSL to rustls
which by now provides better security and efficiency [1].

In a similar vein, it replaces zlib-ng by zlib-rs
which also recently reached performance parity [2].

[1] https://www.memorysafety.org/blog/rustls-performance-outperforms/

[2] https://trifectatech.org/blog/zlib-rs-is-faster-than-c/

Use a rewriter for compound nouns in queries to avoid phrase queries which are of limited usefulness.

Refine splitting of compound nouns to differentiate base nouns and their forms which can only appear at the end of compounds.

Niedersachsen uses the common UVP-Verbund now and their separate systems only yields some information on that now.

There is sufficient churn in Wikidata identifiers that we need to persistent placeholders to ensure sufficient performance.