Decker 1.38 is a maintenance release with numerous bugfixes, performance enhancements, and generalizations.

Fixes (Web)

• The random[] function did not properly select random elements from dictionaries.
• Exporting images from Prototype backgrounds did not work correctly.
• Images with 0xN or Nx0 dimensions were (incorrectly) serialized distinctly from 0x0 images.

Fixes (Web and Native)

• Exporting images did not correctly clip selections to the dimensions of Cards or Prototypes.

New Features

• The readcsv[] function and "by" clauses in queries are significantly more efficient in native-Decker and Lilt. In practical terms, these improvements took some simple data loading and aggregation benchmarks on 5 million rows of data from over a minute to a few seconds. Lilt won't be competing with SQLite any time soon, but this extends the range of problems Decker and Lilt can handle interactively.
• When "compact" slider widgets are at the extreme ends of their range, the appropriate arrow buttons will now be displayed in a disabled, non-interactive state, making it more clear that you've reached the end:

• When "invisible" buttons are set to "Show Transparent", they will no longer invert their background when clicked. This gives "Show Transparent" distinct behavior from the default "Show Solid", and can be useful in situations where the background flash revealing the bounding-box of the widget is aesthetically undesirable.

• The "rtext.cat[]" function now converts Image interfaces into rich-text runs consisting of an inline image, as a convenience, instead of (uselessly) attempting to convert the Images into strings. Many other places throughout Decker's DOM which attempt to cast values to rtext will now perform the same more useful coercion.
• The Image interface now exposes "image.scale[]", which can be given either a single number (for a uniform scale) or a pair (for separate x/y scaling factors). This function modifies the Image in-place, just like "image.rotate[]" and "image.translate[]". This function uses precisely the same nearest-neighbor scaling approach as "image.paste[]", but may be more convenient in some situations. For best results, consider using source images with thick lines and patterns, rather than baked-in dithering.