Re-searching the same place name now returns instantly from an in-session
cache instead of re-querying the geocoding service, which is faster and
kinder to the Nominatim usage policy. Applies to both the geopy/Nominatim
and GeocodeGlib code paths.

…ion#939

Remove the monolithic _hack_query() string-substitution function and
replace it with dialect-aware overrides on the PostgreSQL class:
- _sql_type() returns 'bytea' instead of 'BLOB'
- _quote_column() properly quotes PostgreSQL reserved words (e.g. 'desc')
  rather than blindly renaming them; marked for removal once gramps PR
  #2178 (which adds _quote_column to the DBAPI base) is merged
- Connection.execute() handles only the two remaining SQL syntax
  differences: qmark→format paramstyle and REGEXP→~ operator
- Drop dead LIMIT rewriting code (not generated by current dbapi.py)
- Remove now-unused 'import re'

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…tabases

Quoting the reserved word as "desc" would require a migration for every
existing PostgreSQL database where the column was already created as
desc_ by the old _hack_query. Using desc_ keeps backward compatibility.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Add back LIMIT offset,count → LIMIT count OFFSET offset and
LIMIT -1 → LIMIT ALL rewrites in Connection.execute(), using
re.sub with a staticmethod replacer to match the style of the
other dialect translations.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

23 tests covering all four rewrite rules in Connection.execute()
(qmark→%s, REGEXP→~, LIMIT offset,count, LIMIT -1→ALL), plus
_sql_type BLOB→bytea and _quote_column reserved-word handling.
psycopg2 is stubbed so no real database is required.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Override _create_secondary_columns, _update_secondary_values, and
get_media_handles in the PostgreSQL class to call self._quote_column()
so reserved words (e.g. desc → desc_) are handled at the right sites
rather than via post-hoc string replacement.

Add a targeted Connection.execute() fix for the one CREATE INDEX site
that has no override hook (CREATE INDEX media_desc ON media(desc)).

All four overrides are marked for removal once gramps PR #2178
(_quote_column hook in DBAPI base) is merged into core.

Also inline the LIMIT rewrite lambda (removing _limit_repl staticmethod)
and drop the redundant TestLimitReplStaticmethod test class.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…tests

PostgreSQL has no BLOB type; the core _create_schema() hardcodes "BLOB"
in CREATE TABLE statements that bypass _sql_type(). Add a word-boundary
re.sub in execute() to translate BLOB→BYTEA for all SQL passing through
the adapter.

Also add TestExecuteMediaDescIndex and TestExecuteBlobType so every
rewrite rule in execute() is covered by its own test class.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

The Deep Connections Gramplet searches outward from the Home person and
reports how the active person connects back to them. When the active
person reached Home through one of Home's own direct relatives -- for
example a niece or nephew, who is the child of Home's sibling -- the
reported path named the Home person as an intermediate relationship step
("... / sibling of <Home> / ...") on top of its terminal root. The Home
person is the search origin, so it should appear in a path only as that
root, never as an interior anchor.

The search now keeps the Home person out of the body of every produced
path: a relative that is the Home person is dropped, and when the Home
person is the node being expanded its relatives attach directly to the
root rather than through a redundant "of <Home>" step. Paths still
terminate at the Home person and still connect the active person to
them; only the spurious interior mention is removed. The renderer and
the rest of the search loop are unchanged.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Fixes #8653

The tests/ suite (run by `python3 -m unittest discover -s tests`) imports
gramps.gui.* modules directly, which bypasses the Gramps GUI launcher's own
gi.require_version. On a host where GTK 4 is the default GI resolution, Gtk/Gdk
then load unpinned — emitting PyGIWarning and risking the wrong stack (and the
GUI-touching tests silently skipping).

Pin both in tests/__init__.py, which `unittest discover -s tests` imports before
any test module, so the whole repo-root suite runs on the GTK 3 / GDK 3 stack a
real Gramps session uses. tests/test_gtk_version_pin.py guards it (skips where
the 3.0 typelibs are absent).

Verified: `python3 -m unittest discover -s tests -p "test_*.py" -t .` pins both
(get_required_version == "3.0"); the guard is red without the pin.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>