OpenConstructionERP v14.6.0
Tabbing across a bill of quantities reaches the quantity cell and stays there. Moving from the unit column to quantity opened the editor and closed it again in the same keystroke, so the cell appeared to blink and then refused to accept typing, and the only way in was the mouse. The unit editor committed on Tab without cancelling the key, which left the same press to be handled a second time by the grid's own Tab handling, and the second move tore down the editor the first had just opened. The columns before it are unaffected because they use stock editors whose Tab handling has always cancelled. Cancelling the key also removes the grid's own move, so each of the three hand-written editors now makes the move itself and follows the modifier: Shift and Tab walks backwards, where before all three walked forwards regardless. The unit editor had no tests at all; it has six now, three of which go red if the cancel is removed.
A register that records who carried out an inspection prints their name rather than an identifier. The column holds a bare thirty-six character string with no foreign key behind it, and three different things write to it: seeded data records an external party, a picker on other screens records a member of the team, and the form on this screen accepts free text because the field asks for a name. Resolving it against any one of those sources leaves the other two blank, which is why it was left unresolved and a page of identifiers reached the screen, the drawer and the spreadsheet export. It resolves against parties first and people second now, and keeps whatever was stored when neither matches, so a typed name still shows as typed and a deleted party still leaves a trace. The lookup runs once per page rather than once per row, and a value that cannot be an identifier never reaches the database, which on PostgreSQL would refuse the comparison and take the whole list with it.
The check that reads what the macOS sidecar carries can no longer clear a file it never opened. It reported a member count and a verdict over the members it could read, and a member it failed to extract fell out of that count in silence, as did one whose signature it could not parse, which was filed as carrying no identity rather than as unknown. Both now fail the check rather than passing quietly, and it accepts the name of a file that has to have been inspected for the run to count, which the release workflow now sets to the framework named in the report we have. This settles nothing about that report; it removes a measurement that read as settling it.
Full Changelog: v14.5.0...v14.6.0