[facontidavide]

What

Six defects from the pre-1.0 review. None changes an ABI layout, a wire format or a public signature in a source-incompatible way. Each code fix has one regression test that was seen failing before the fix.

#	Defect	Fix
1	toStringView returned a view of size bytes past a one-byte literal for {nullptr, size > 0}. It is applied to plugin-supplied views in the host-side adapters.	Returns an empty view when the pointer is null. The existing test accepted the bug and is now stricter.
2	The dialog marker decoders (chartMarkers(), timelineMarksFromJson) used nlohmann's throwing value(). A plugin sending {"kind": 1} threw through the host, which calls them without a try/catch.	They skip the malformed entry, as the sibling list decoders (chartSeries, rangeSliderMarkers) do.
3	DialogHandle kept a null context when the plugin's create() failed, then passed it to plugin trampolines that dereference it.	New valid(); each method returns its empty value without calling the plugin. Before the fix the new test hit a null dereference under ASAN.
4	One non-UTF-8 string made WidgetData::toJson() throw; the trampoline then replaced the whole update with {}, silently.	Invalid bytes become U+FFFD. The 31 WidgetEventBuilder serializers had the same unguarded dump() and use the same helper.
5	The Conan assert_throws option did not reach consumers, so their inline PJ_ASSERT disagreed with the archives (ODR).	package_info() adds the define to the base component, as the conan-center recipe already did.
6	Builtin object types were invisible to the feature-floor checker, so a plugin emitting MeshMap with a floor below 1.0.0 passed.	The extractor now emits PJ_builtin_object_type_t enumerators; feature_floors.json gains a 1.0.0 group for kMeshMap and kMeshMapUpdate.

Things a reviewer should know

• Fix 2 removes a duplicate. chartMarkersFromJson had no callers while the host view re-implemented it, despite a comment calling it the single source of truth. The view now calls it, so ChartMarkerView is an alias of ChartMarker (same fields, order and defaults). PlotJuggler only uses the result through auto.
• Fix 2 is slightly stricter. A timeline mark with a float in an integer field (start: 3.7) used to be truncated and is now skipped. The SDK's writers always emit integers.
• Fix 6, baseline. All 18 pre-existing types are classified as baseline. I checked against git tags that each first shipped by 0.26.0, below the 0.28.0 minimum floor, so that is accurate.
• Fix 6, over-matching. The aliases MeshMap / kMeshMap also match a plugin that only reads meshes, which would work on an older host. It will be asked to raise its floor. That was chosen over letting an emitting plugin slip through.
• Fix 6, changelog. The two sentences in the 1.0.0 Host contract: paragraph that said the checker cannot see builtin types are rewritten.

PlotJuggler follow-up (optional)

pj_dialog_host/src/dialog_engine.cpp constructs a parser DialogHandle and uses it at once. With fix 3 that no longer crashes, but a dialog whose constructor threw would show an empty UI. It should check valid().

Verification

• ./build.sh --debug and ./test.sh (Debug + ASAN): 107 of 107 pass
• check_feature_floors.py: OK, 162 surfaces. test_declared_versions.py: 7 tests pass, including a new MeshMapFloorTest
• pre-commit clean
• Fix 5: conan create with assert_throws=True, then a consumer that fails to compile without PJ_ASSERT_THROWS: passes

Rebased onto main after the 1.0.0-dev.1 commit; not rebuilt after the rebase. This PR and #230/#231 edit nearby lines in feature_floors.json, extract_host_surfaces.py and conanfile.py; whichever merges later needs a small rebase.

🤖 Generated with Claude Code