In a container running the tests on two processors, the Quarto
gallery's proved example was rendered with the warning that time ran
out: the generated inputs that cross-check a proof had been cut short by
the 3-second clock, with every claim already proved. core 941af6b counts
that only where a claim is still open. The tests that hold what a proof
finds give it time to spare, and the browser-side request builder takes
prove-seconds as the Lua filter does.
It was blank until every module had loaded and the page's examples had
been run, a fuzzer's report among them (about a second of work): first
paint came 1.3 seconds into a repeat visit and about 3 into a first one.
- The assembler draws the home page once, in Node, with the app's own
homeView, and puts it inside #app. The browser paints it with the
HTML and the stylesheet.
- When the app has loaded it draws the page again, with its controls,
and puts it in the prerendered one's place; the fuzzer's reports and
the proof's counts are copied from the prerendered page, not worked
out again.
- The prerendered page shows only on the home address: the page's first
inline script marks it before the first paint, and without the mark it
stays hidden, so an embed or a problem's address never shows it. That
script changed, and its hash in docs/Caddyfile with it.
Served locally, first paint is at 0.1 to 0.2 seconds and the page with
its controls at 0.2 to 0.3; both were at 1.2 to 1.4.
The Prove view gives each routine 3 seconds on the clock. On a slow or
busy device that runs out before the proof does, and claims stay open
with nothing to say why.
- The view says when the time, not the proof, stopped the checker, that
what was not finished is open and not disproved, and offers to prove
again with the next longer time. The setting is at the foot of the
view: 3, 10 or 30 seconds or 2 minutes for each routine, kept in the
browser (pplus.proveclock).
- A proof rendered with Quarto takes prove-seconds on the shortcode or
pplus-prove-seconds for the site, and the render warns when the time
ran out.
- The home page's own example is proved with no time on the clock, and
the tests that hold a proof's outcome give it time to spare, so
neither depends on the machine.
- core moves to 38e6e2c, whose report says which routines the clock
stopped.
Walking the site as a student (desktop, a phone-width screen, dark) found
these. Each is small; together they are what got in the way.
Defects:
- Opening a problem in the Bridge saved the generated Mojo over the
exercise's own program. The workspace fills every pane's editor as it
opens, the editor reported that as typing, and the Bridge's read-only
Mojo pane holds the generated text. Developer then showed generated
Mojo labeled "your edits", and the problem counted as edited. The
value setter now marks its change as a load, the change listener
leaves loads out, and a Bridge session refuses a Mojo edit.
- A page opened at #/#problems or #/#reference stopped about 45px past
the heading: the fonts arrived after the place was computed. A page
built this moment goes straight to its section and finds it again
when the fonts are in; a page already on screen still scrolls there.
The workspace:
- With no run to step, the Run view said "Nothing to run yet." It and
the stepper now give one reason (src/core/norun.ts): how many problems
the stepped pane has and the first with its line, with a button to the
line and the Problems list; or that the pane is empty.
- In the Bridge the dock starts at about a quarter of the height (it
holds a call, a line of output, a short list), so the code and the run
get the room. Developer keeps a third for the compiler's answer. A
double-click on a divider returns to the area's default, not to the
student's last position.
- The reason a pane is not in step wraps instead of ending in an
ellipsis, and a button in the pane's head keeps its words on one line.
- A problem that does not exist keeps the site's navigation and links
the problem list.
- Help, the theme switch and the stepper's first, back, step and last
buttons have names a screen reader can say.
A phone-width screen:
- The toolbar is three rows in every area and on every tab (it was four
or five): the mark, the area, help and the theme; the problem,
Versions and Reset; the actions. About 30px more of the code shows.
An embedded workspace keeps its own toolbar.
- The Share dialog puts each row's name above what it holds (a note had
been squeezed to a word per line), and the Versions menu opens across
the screen instead of running off its right edge.
- The stepper's speed label stays with its slider.
Checked by hand, in headless Chromium at 1440x900, 1366x768, 390x844 and
320x640, light and dark: each state above before and after; the editor
has line numbers and highlighting; the stepper steps; embeds
(playground, trace, Developer) are unchanged; 31 smoke scenarios clean
but the ones that err on purpose.