v3.10.0

A text value knows how wide it is, and the editor is one you can work in.
AP 6.4.15.13 defines display width and PasUnicode.Columns answers it, so a
column in the editor is a cell rather than a byte; and the editor opens eight
documents, scrolls sideways, lands on a diagnostic in whichever of them it
names, and draws in twenty-four-bit colour where the terminal has it.

Added

• A question is a box (ADR-0392). Find, Go to line and Save as are
  centred framed dialogs instead of a line at the bottom, so the editor draws
  one kind of thing for one kind of thing. It is the second user of the panel
  primitives, which is what tests whether they are an abstraction.
• The editor has a menu bar (ADR-0391): File Edit Search Run along
  the top, F10 (or Ctrl-O) to open, arrows to move, Enter to choose, Ctrl-C to
  close — drawn as a framed drop-down in real box characters. A screen
  became one array element per display column to make that possible: ┌ is
  three bytes and one column, so a row indexed by byte could not hold a frame
  and keep the colour plane naming what a person sees. A menu item names a
  key, so choosing one is the same path as pressing it — no second dispatch.
• The editor decodes the function keys and binds two of them in Turbo
  Pascal's positions: F2 is Save and F9 is Build. Both spellings a
  terminal uses are understood — ESC O Q (SS3) and ESC [ 1 2 ~ (CSI) are
  the same key — and the decoder maps a bound one to the key it is bound to,
  so nothing above it learns that Save has two spellings. The other ten are
  decoded and reported by number (F5 is not bound) rather than ignored. The
  hint bar now reads F2 Save F9 Build ….

Added

• A second reading of the width properties (ADR-0400).
  East_Asian_Width is the one Unicode property here with no conformance
  file, so the new icu-width case compares AP 6.4.15.13's table against
  ICU's reading of the same two files — applying the clause's own rules
  to UCHAR_EAST_ASIAN_WIDTH and u_charType, so what is compared is two
  transcriptions and not two opinions about rendering. 1 112 064 code
  points, exact agreement. It abstains when ICU's Unicode version is not
  the pinned one, which is what makes exactness a claim worth making.
• The editor scrolls sideways (ADR-0397). A line wider than the window
  was cut; the window follows the cursor now, by column rather than by
  byte, so a line of Japanese scrolls by what a person sees and a wide
  character straddling the edge is dropped rather than half-drawn.
• The editor opens more than one file at once (ADR-0396). F3 opens, F6
  goes to the next, up to eight — each with its own cursor, scroll, dirty mark
  and undo journal, and the status line says which of how many. This is what
  makes it usable on this compiler: pascalc translates every
  program-component and this compiler is three of them, so a diagnostic
  naming an open document is now landed on, switching to that file.
• PasUnicode.Columns, and AP 6.4.15.13 to define it (ADR-0395). How many
  cells a text value occupies when a fixed-pitch terminal draws it: Wide and
  Fullwidth take two, a mark or a format character takes none, and the unit is
  the element, so e with a combining acute is one cell and a family
  emoji is two rather than six. AP 6.4.15 NOTE 14 had put this outside the
  language and is amended to point at the clause — it was right that no
  property of a character answers for a proportional font, and wrong that
  nothing was left to answer.
• The editor edits East Asian text correctly (ADR-0395). A column is a
  cell rather than a byte, the arrows and Backspace and Delete move by
  element, and a wide character occupies two cells with the second holding
  nothing. ed.col is still a byte and the status line still shows one,
  because pascalc reports a diagnostic's column in bytes and Ctrl-B has to
  land on it.
• PasTerm.SetRgb, twenty-four-bit colour (ADR-0394). SGR's direct
  38;2;r;g;b form over an Octet subrange, beside the eight ANSI colours
  the module has had. It has no clDefault and cannot — the terminal's own
  colour has no numeric value to name — so a program drawing a document still
  writes SetColour(clDefault, clDefault) and leaves the person's own colours
  alone. Whether a terminal understands the form is the caller's policy: there
  is no query one lacking it answers safely, and COLORTERM is what emulators
  set.
• The editor draws a published palette where the terminal can show it
  (ADR-0394). apide reads COLORTERM once and keeps two tables: the eight
  ANSI colours for every terminal, and five colours of the LUXE scheme for
  one that says it understands twenty-four bits. Five of ten is the finding
  rather than a compromise — the scheme's rose, khaki, slate, blue and
  vermilion reach at best 6.7:1 against anything else in it, and every cell of
  a text editor is text. Both tables are held to the same floor.

Fixed

• Ctrl-B no longer lands on the wrong line of the wrong file. EditFault
  parsed file:line:col: and threw the filename away, so a diagnostic naming
  one of the other program-components jumped to that line number in whatever
  document was open and displayed the message as though it had arrived there.
  pascalc translates every component and this compiler is three of them, so
  that was the ordinary case rather than a corner. A diagnostic about another
  file is now reported as the compiler wrote it, position and all, and the
  cursor stays where it was.
• The editor's colour is legible, and a role that nothing draws now fails a gate (ADR-0393). A drop-down's body was cyan on blue, which is 4.8:1 on xterm's own palette and worse on a muted theme, and the message line was yellow on the terminal's own background, which on a light terminal is yellow on white. Every role but the document's text now pairs black or white with a colour, at 7.5:1 or better. And a dialog's answer is drawn as a field again: crPrompt — the editor is waiting for you — had been declared, coloured and drawn on no screen since the prompts became boxes, with all fifteen session goldens agreeing. The new tui-palette case holds both claims in both directions.
• A menu that was legible and indistinguishable (ADR-0394). The fix above
  put the selected menu title at black on white against a black-on-cyan bar:
  16.7:1 to read and 1.6:1 away from the four titles beside it, so nothing
  said which menu was open. A contrast floor cannot see that — both pairs pass
  it — so tui-palette now also requires two regions a person sees at once to
  differ in their backgrounds by 3:1, and the three bars along the bottom
  were rechosen together.

Changed

• The editor is apide, and that is now its only name. v3.9.0 installed
  it as afterschool, on the convention pascalc sets over
  selfhost/compiler.pas — but tui/build.py names its default output after
  the program it was given, so the script wrote build/bin/apide and the
  CMake target wrote build/bin/afterschool, leaving one program under two
  names with nothing to say which one a document meant. The script's default
  is what keeps build.py '' session.pas from overwriting the editor and so
  could not move; the binary did. cmake --build now produces
  build/bin/apide and cmake --install puts it in <prefix>/bin/apide.