Datum Plot 0.3.0

A full check of the whole application: security, robustness, reliability and speed. No new features; everything that was there works more safely.

Nothing is lost

• Saving a project, the options and every export (pictures, pages, .xlsx, .pptx, .docx, notebooks) now writes a temporary file first and swaps it in only when it is complete: a crash, a full disk or a power cut never leaves a half-written file.
• If something unexpected goes wrong inside the app during an edit, that one step is refused, your project stays exactly as it was, and the details go to crash.log in the app's data folder instead of closing the app.
• A project whose internal numbers were damaged (repeated or impossible identifiers) is repaired on opening, with every graph, result and layout still linked to its data.

Safe files

• Graph settings out of range in a project file (sizes, fonts, angles, contour levels, marker sizes) are brought back within limits when it opens.
• Damaged or unreadable data files (broken workbooks, garbled text, cut-off .pzfx files) give a clear message instead of an engine error.
• Spreadsheet exports never contain live formulas: text that looks like one (starting with =) is written as text. Notebook CSV files mark text starting with =, +, - or @ so a spreadsheet will not run it.

Safe updates and window

• The updater accepts only an installer with the expected name, requires the release's checksum file, limits the download size and saves it in a new private folder.
• The window refuses frames, plug-ins, forms and remote scripts (a stricter content security policy).

Tested to break

• New robustness tests feed the engine hundreds of random and hostile projects and edit sequences, damaged and truncated files, and impossible values; they found and fixed one crash (pasting at an impossible position).
• An engine check runs every sample project through the real Python engine: every analysis, every graph type on every kind of table, every export format, layouts, slides, reports, notebooks and the console (532 requests, no failures).
• The checks now also run on Windows and macOS on every change.

Faster

• Typing a value in a very large table (a million rows) is about seven times quicker (0.05 s instead of 0.4 s): checking whether an edit changed anything no longer compares a million empty row titles one by one.