Skip to content

Release Notes — Sui 0.75U

Branch: Sui0.75T → Sui0.75U
Base: Sui 0.75T (no-wrap fix, named diff presets, alias in window title)
Date: 2026-05-24

Summary of Changes

Area Type Description
Cross-Connection Broadcast New feature Run one SQL against multiple connections in one step — tick target connections, set optional per-connection schema override, get one result window per target
Cross-Connection Broadcast New feature Vendor-aware SET SCHEMA injection before each query (DB2, Oracle, PostgreSQL, SQL Server, MySQL/MariaDB)
Cross-Connection Broadcast New feature Named target groups — save, load and delete groups of connections+schemas to BroadcastGroups.pro
Cross-Connection Broadcast New feature Accessible from main menu ("Cross-Connection Broadcast…") and right-click popup ("Broadcast to connections…")
Performance Optimization ExpCSV — row builder switched from String.concat chain to StringBuilder (per-row was O(n²)); large CSV exports now run in linear time and produce far less GC pressure
Performance Optimization ExpXLS / ExpXLSRS — cell styles + fonts configured once instead of mutated per cell (was hundreds of thousands of redundant setter calls on large exports); SXSSF auto-sizing now tracked before rows are written so it actually measures all data instead of only the last 100 rows in the streaming window; BIGINT values no longer fall back to string — Long/Double parse path covers them
Performance Optimization formatJSON — character-by-character accumulator switched from ret += c to StringBuilder; eliminates the O(n²) string-copy blow-up on large JSON payloads
Performance Optimization FormatXML — output accumulator switched from static String concatenation to a local StringBuilder; the method is also re-entrant now (no shared static state between calls)
Transpose window Bug fix IllegalArgumentException: Row index out of range eliminated — F8 in the transpose window's row-header column no longer crashes when no row is selected
Transpose window New feature F7/F8 row navigation — press F8 for next row's transpose, F7 for previous row; works in both free-floating and tabbed modes
Transpose window Enhancement Old transpose is removed (tab or window) when F7/F8 opens the replacement — no orphan windows or tabs accumulate
Transpose window Bug fix Transpose window captures keyboard focus on open — F7/F8 fires on the first keypress, not the second
Version Bump 0.75T → 0.75U

Detail

Cross-Connection Broadcast (Broadcast.java)

A new dialog accessible from two entry points: - Main menu → Tools → Cross-Connection Broadcast… — pre-loads the current query editor text - Right-click popup → Broadcast to connections… — pre-loads the selected/full query text

The dialog shows every URL that has been connected in this session (from ConnHistory). Tick the target rows, optionally fill in a schema override per connection, then click Run broadcast. A multi-statement script is assembled using the existing #URL= directive to retarget between SELECTs; a vendor-appropriate SET SCHEMA statement is injected after each #URL= when a schema is specified and the vendor is known (DB2, Oracle, PostgreSQL, SQL Server, MySQL/MariaDB). The script is handed to RunSql/RunIt as a single batch, producing one QueryRep result window per target.

Named groups — tick a set of connections, click Save group…, give it a name. Groups persist to BroadcastGroups.pro in the Sui home directory and reload automatically. Per-alias default schemas are also written to SUI.CONN.<alias>.SCHEMA in ConnProp so they pre-fill the dialog on subsequent opens.

Performance optimizations

Three classic O(n²) string-concatenation hot spots were replaced with StringBuilder, mirroring the sort-algorithm replacement in earlier releases. None of these change behaviour — only allocation patterns and runtime on large inputs.

  • ExpCSV — the inner cell loop built each row via Data = Data.concat(…) and the outer row loop reset and rebuilt the buffer per row. For a 10 000 × 50 cell export this meant ~500 000 intermediate String copies. Now a single StringBuilder is reused with setLength(0) between rows. Behaviour preserved: header .trim() semantics, the SUI.CSVEMPTY blank-line prefix, and the per-column String-class quoting branch.
  • ExpXLS / ExpXLSRS — three POI-specific issues fixed:
  • Cell styles and fonts mutated per cell. The old createCell called cs.setVerticalAlignment(…), cs.setAlignment(…), cs.setDataFormat(…), cs.setFont(…) and f1.setFontHeightInPoints(…) on every single cell, even though the values are constant. For a 10 000 × 50 cell export that’s ~500 000 redundant POI setter calls on five shared CellStyle objects. Now the styles and fonts are configured once in the constructor and createCell only writes the value and assigns the pre-built style. POI's golden rule: configure CellStyle/Font once, reuse the reference.
  • Auto-sizing was effectively a no-op on big result sets. SXSSF flushes rows out of its 100-row window to disk; once flushed they are invisible to autoSizeColumn. The old code called trackColumnForAutoSizing after all rows were written, so the auto-size only ever saw the last 100 rows (and was still expensive because of POI's font-metric measurement). Tracking is now hoisted to trackAllColumnsForAutoSizing() before any row is created, and the post-write loop just calls autoSizeColumn once per column. Both correct and faster.
  • Long integers fell back to string. Integer.valueOf(x.trim()) threw for any value outside the int range (every BIGINT, every long-ish identifier). The catch wrote the raw string, defeating Excel's numeric handling. The parse chain is now Long.parseLong → Double.parseDouble → string, so 64-bit integers stay numeric. Also removed the deprecated cell.setCellType(CellType.NUMERIC) call — setCellValue(long/double) already establishes the cell type.
  • formatJSON — character-loop accumulator changed from String ret = ""; ret += c… to StringBuilder ret. Indent runs that previously called String.format("%Ns", "") to manufacture spaces are now plain append(' ') loops (no formatter allocation, no width-parsing). The backslash-lookahead chars[++i] is preserved.
  • FormatXML — dropped the shared private static String utXML accumulator and the SkrivRad() helper; the output is built into a local StringBuilder and returned. Side benefit: the method is now re-entrant (the old static state would corrupt output if two threads ever called FormatXML concurrently).

F8 crash fix (QueryRep — row-header key listener)

table.addRowSelectionInterval(-1, -1) was called when tablef.getSelectedRow() returned -1 (no row selected in the row-header column). A sel >= 0 guard was added so the interval is only set when a valid row is selected.

F7/F8 row navigation

mainPanel input-map bindings (using WHEN_ANCESTOR_OF_FOCUSED_COMPONENT for child windows, WHEN_IN_FOCUSED_WINDOW for the parent) replace the old menu-accelerator approach, which stopped working once a panel was moved into QueryRepTabbedFrame.

  • Parent window — F8 opens a transpose for the currently selected row in the main data table (table.getSelectedRow()).
  • Child (transpose) window — F8 calls MyParent.PosRows(true) (next row) then disposes or removes the current child; F7 calls PosRows(false) (previous row).
  • Tabbed child — QueryRepTabbedFrame.disposeTab(QueryRep) removes the tab entry and disposes the backing JFrame in one step, preventing orphan tabs.
  • Free-floating child — showOrTab now calls qr.toFront() after qr.show() so the new window comes in front of the parent.

Focus fix (showOrTab)

qr.table.requestFocusInWindow() is called after the new child is shown or added to the tabbed frame. This satisfies the WHEN_ANCESTOR_OF_FOCUSED_COMPONENT condition immediately, so the first F7/F8 keypress navigates rather than just moving focus.