Console and variables¶

The bottom row of the window is the kernel described twice: as a transcript you can type at, and as a table you can open.

The console¶

The Console tab is a real jupyter console attached to the same kernel Run Cell sends to. Nothing is interpreted on the way: keystrokes go to a pty, bytes come back and are written verbatim. So it behaves like IPython because it is IPython — line editing, history search, tab completion, ? and ??, and the magics all come from the other side rather than from a widget imitating them.

That is what makes the two halves of the window fit together: run a cell with Shift-Enter, then ask the console about what it left behind.

In [1]: part.volume
Out[1]: 1256.6370614359173

Ten thousand lines of scrollback are kept, so a traceback from earlier in the session is still there. Right-clicking offers copy and paste; on macOS the standard Cmd-C / Cmd-V work through the application menu, which is what gets them delivered into a pane the browser would otherwise swallow them in.

About carries a Kernel connection section at the foot, with this instance's Jupyter connection file and the command that uses it:

jupyter-console --existing "<the path About shows>"

The quotes are not decoration — the path contains a space on macOS and Windows both. Install jupyter-console yourself, outside Studio; the environment here is Settings' to manage.

The variable explorer¶

The pane beside the console lists what is in the namespace: name, type, and a short repr, with a length in brackets for anything sized and build123d's own label where an object carries one. Drag the boundaries between the columns to re-proportion them; the widths are remembered.

Rows are pushed by the sidecar whenever the kernel goes idle, so the table is current after a run and after typing in the console, with nothing to refresh and no polling. That listing is deliberately cheap — it reads only what is already there, never anything that computes geometry.

Expanding a row asks for its contents at that moment, and goes as deep as the data does: a row inside a row inside a row is addressed by position, so an assembly unrolls by its children the way it is built. Long collections are paged rather than truncated.

What a click will and will not compute¶

Expanding a build123d shape adds its face, edge and vertex counts. It does not add volume, area or the bounding box, and that is measured rather than cautious — on a suspension assembly, with OCCT caching none of it:

The counts are enough to tell two versions of a model apart, which is what the pane is for. The rest belongs in the console, where you ask for it and can see what it costs:

In [2]: rc.volume

The viewer already shows the extent bounding_box() would report, and show() prints it.

Which world the pane is describing¶

The tab on show decides, and only that:

The two worlds share no addresses, so nothing is carried across when it switches. Nothing switches back on its own either: after a run or a session ends, the output is still there and still worth reading.