[DanielDarrenJones]

The entries_count term index (used when terms are filtered or sorted by it, such as min_count on the taxonomy tag) is built by asking for each term's published entries one term at a time. Each of those is a query over every entry, so the build grows with terms × entries.

This finds which of the associated entries are published with one query, then counts each term from the associations the index already reads.

On a generated site of 11,771 entries and 2,446 terms (6.42.0, CPU time, two runs each):

Taxonomy	Terms in index	Before	After
categories	24	0.74 s	0.08 s
tags	400	8.6 s	0.13 s
keywords	2,022	43.1 s	0.29 s

The index values are identical before and after (compared serialized, all three taxonomies).

Where this build happens during a request rather than a warm, that request is the one waiting for it.

Kept as they were:

• A single term being updated (a save) still counts through Term::entriesCount().
• A term repository that overrides entriesCount() is still asked for each term.

The new test counts entry queries while the index is built: 11 before, 1 after.

Found by @godismyjudge95, who measured this on a much larger site and shared the approach.