starplast.theme

The color palette for a named theme.

Whether a palette is a dark one, which decides contrast choices elsewhere.

Hex to the 0-1 RGBA tuple pyqtgraph's GL items want.

GL viewport background — the page color, not a literal, so themes stay coherent.

The gray for 'unknown, not absent'. Derived from fg_dim so it reads on either ground.

A pyqtgraph colormap, falling back rather than raising inside a paint call.

The color maps appropriate to one kind of data: sequential, diverging or categorical.

Which family of color map a column deserves.

Diverging only when the data actually straddles a midpoint -- a residual or a score that goes both ways. Applying one to a strictly positive quantity invents a midpoint that is not there.

Whether a theme's ground is light. Decides which direction contrast has to go.

Distance from the camera, normalized to 0 (nearest) .. 1 (farthest).

rng is the (min, max) distance to normalize against. Pass the WHOLE cloud's range when cueing a subset -- edges touching a point must fade by the same amount as that point, and a subset normalized against its own extent does not agree with the full set at the same location.

Scale alpha and size by distance from the camera. Returns (alpha_array, size_array).

n distinct colors for unordered classes, legible on this theme's ground.

A palette tuned for a dark background washes out on a light one -- the same hues that read as saturated against near-black read as pastel against near-white. Rather than ship two hand-tuned lists, the colors are darkened for light themes and lightened for dark ones, so any chosen categorical map stays legible on either ground.

pyqtgraph GL option set for a point mode.

additive is offered because density is sometimes what you want to see, but it is not the default: with 8,140 genes in dense UMAP clusters it saturates every color mode to white and destroys the encoding entirely.

The glass colour for one role as an (r, g, b, a) tuple of floats in 0-1.

Black on a dark theme, white on a light one, at GLASS_ALPHA[role]. When the display cannot composite (translucent=False) the same pane is returned OPAQUE, as it would look laid over the theme's own background: a near-black on a dark theme rather than a black square with garbage in its corners.

The glass colour for one role, as a stylesheet colour (rgba(...), or #rrggbb when opaque).

The hairline round a glass pane: a faint light edge on black glass, a faint dark one on white.

A '#aarrggbb' string, which both a stylesheet and QColor read. Opaque displays get the palette's own border, since a translucent edge over an opaque pane has nothing to show through.

rgba(r, g, b, a) from a hex colour, for a panel that has to let the background through.

Qt stylesheet for one theme. Every color comes from the palette, never a literal.

container_opacity below 1 lets whatever is painted behind the window -- the drifting blob field -- show through the panels. Only the CONTAINERS take it: a translucent field would put moving colour behind text somebody is trying to read, and the point of the background is that it is behind things.

translucent says whether the display can composite translucent windows (see starplast.glass.compositing_available). Menus, tooltips, drop-down lists and glass windows are translucent black when it can, and the same colour made opaque when it cannot.

A list of choices carried as TICKS, chosen by dragging across the rows and confirming with the right-click menu.

Selection and choice are separated deliberately, and that separation is the whole point. Both category lists in this program used the SELECTION as the choice, which has two costs a reader pays constantly: a set assembled over several ctrl-clicks is destroyed by one ordinary click, and there is no way to keep a set while clicking somewhere else to look at something. Here a drag selects, the context menu turns that selection into ticks, and the ticks survive every later click.

The one-item fast path is not lost: clicking an item's own box ticks it, which is Qt's behaviour for a checkable item and needs no menu. Space toggles the selected rows for the same reason.

checkedChanged fires ONCE for a bulk operation rather than once per row. The lists it feeds trigger a redraw of 8,140 points, and ticking forty categories should redraw once.

A tree of tick boxes where ticking a CATEGORY ticks everything under it.

The flat list this replaces had 96 rows, one per feature block, and at that length the list stops being a set of choices and becomes a scroll. What a reader actually wants to say is "all the transcription evidence", or "everything except the fitness screens" -- and both of those are statements about a LEVEL of the slot hierarchy, which already exists and already has the addresses.

Three properties make it usable for the thing it is for, which is choosing what a map is built on before holding something out of it:

• Ticking a group ticks its descendants, and a group whose children are only partly ticked shows partially rather than lying in either direction.
• checked() returns LEAVES only. A group is a way of saying something about its leaves, not a thing that can itself feed a map.
• "Everything except this" is one action, because that is the hold-out question -- build the map on all the evidence but this, then ask whether this comes back -- and doing it by hand on 96 boxes is how a reader ends up holding out something they did not mean to.

A boolean slider, the shape and colors of spacr.gui_elements.spacrSwitch.

A checkbox would have done the job. This exists because the user runs both programs and a setting that looks like a setting in one of them should look like one in the other: 40x20 track, a 12-pixel knob that slides, purple for off and teal for on, caption on the left.

A tooltip as one block of even lines, wrapped once and left that way.

Qt lays a plain tooltip out on a single line, so any real explanation becomes a strip wider than the screen. Inserting line breaks alone does not fix it: Qt re-wraps rich text at a width of its own choosing and strands two words on a row. A fixed table width does not constrain it either -- that was tried here and measured at 1,588 pixels, because Qt only wraps when it is given an explicit text width and a tooltip sets its own.

white-space: pre is what stops the second wrap: 1,188 pixels to 347 on a typical tooltip, with the breaks where they were put. Lines are padded to equal length so the block is a rectangle rather than a ragged edge; true justification is not in Qt's rich-text subset.

width is in characters, not pixels, because that is what the wrapping is done in.

Rewrite every tooltip under root as a wrapped block, and move it onto its label.

Two things, because they are the same complaint: a tooltip belongs on the SETTING, not on the box you type in. A form row is a label and a field, and hovering the label -- which is what names the thing and what the eye goes to first -- said nothing at all. Returns how many were rewritten, so a test can tell this ran.