Follow-up to #208 (fixed for CSV export in #213). The survey done there found the same dtype-blindness — stringifying or widening typed getters without consulting unsignedness — in these consumers:
cli/src/main/java/io/github/dfa1/vortex/cli/tui/GridRender.java:61-64 — Long.toString(getLong) / Integer.toString(getInt) / Short.toString(getShort) / Byte.toString(getByte): U8–U64 high-half values render negative in the TUI grid.cli/src/main/java/io/github/dfa1/vortex/cli/tui/InspectorRender.java:51-54 — identical pattern.cli/src/main/java/io/github/dfa1/vortex/cli/FilterCommand.java:203-206 — compareValue compares raw signed getters, so filter predicates on unsigned columns are wrong for high-half values (e.g. magnesium >= 130 on uci-wine misses rows).calcite/src/main/java/io/github/dfa1/vortex/calcite/VortexTable.java:812-817 — U8/U16 correctly widen via getInt, but U32 -> getInt and U64 -> getLong surface raw signed values to Calcite.calcite/src/main/java/io/github/dfa1/vortex/calcite/VortexAggregates.java:107,112 — scanSum accumulates raw getInt/getLong into a signed sum; U32/U64 sums are wrong for high-half values.
The filter and Calcite items are silent-corruption class (wrong query results); the TUI items are display-only. ByteArray.getInt/ShortArray.getInt already zero-extend per dtype — the U32/U64 cases need the same treatment CSV export got in #213 (isUnsigned gate + unsigned widening/strings, and for sums an unsigned-aware accumulation or a documented refusal).
Found by the Raincloud conformance work (#205).
Follow-up to #208 (fixed for CSV export in #213). The survey done there found the same dtype-blindness — stringifying or widening typed getters without consulting unsignedness — in these consumers:
cli/src/main/java/io/github/dfa1/vortex/cli/tui/GridRender.java:61-64—Long.toString(getLong)/Integer.toString(getInt)/Short.toString(getShort)/Byte.toString(getByte): U8–U64 high-half values render negative in the TUI grid.cli/src/main/java/io/github/dfa1/vortex/cli/tui/InspectorRender.java:51-54— identical pattern.cli/src/main/java/io/github/dfa1/vortex/cli/FilterCommand.java:203-206—compareValuecompares raw signed getters, so filter predicates on unsigned columns are wrong for high-half values (e.g.magnesium >= 130on uci-wine misses rows).calcite/src/main/java/io/github/dfa1/vortex/calcite/VortexTable.java:812-817— U8/U16 correctly widen viagetInt, butU32 -> getIntandU64 -> getLongsurface raw signed values to Calcite.calcite/src/main/java/io/github/dfa1/vortex/calcite/VortexAggregates.java:107,112—scanSumaccumulates rawgetInt/getLonginto a signed sum; U32/U64 sums are wrong for high-half values.The filter and Calcite items are silent-corruption class (wrong query results); the TUI items are display-only.
ByteArray.getInt/ShortArray.getIntalready zero-extend per dtype — the U32/U64 cases need the same treatment CSV export got in #213 (isUnsignedgate + unsigned widening/strings, and for sums an unsigned-aware accumulation or a documented refusal).Found by the Raincloud conformance work (#205).