Skip to content

Speculative Terminal StandardForm

enumeratio edited this page Nov 15, 2026 · 1 revision

StandardForm in the terminal

Speculative. A proposal, not adopted.

The Ideas page lists "StandardForm in the terminal" as an enriched text rendering of the boxes (Unicode, script characters, colour, a second line where the terminal allows) instead of a fall back to InputForm. Most of that is built. The CLI has a two-dimensional StandardForm on character cells, with a golden test, and the REPL still prints InputForm unless the reader asks for it. What is missing is the default, a fit to the terminal's width, a settled colour story, and tests that hold the rendering to the expression. This page says what is built, what the gaps are, and the smallest slice that closes them.

"Measured" means read or counted from the tree at origin/main (bfa63e225), with a path:line; nothing here was run (no node_modules, no builds, no tests). "Assumed" means reasoned from that reading and not executed. Paths are under the repository root.

Recommendation

Do not write a new text renderer. Keep the split that exists: StandardForm is the 2-D cell layout (layout with cell metrics, drawn by packages/cli/src/cell-layout.ts), and OutputForm is the one-line Unicode serialiser (toText). Change four things:

  1. Default. The Node REPL and the one-shot CLI print StandardForm when stdout is a TTY, and InputForm otherwise, unless a config file or :form says otherwise. Today only show mode does this.
  2. Width. Lay out against the terminal's columns. A block wider than the terminal falls back, per result, to toText of the same boxes, then to InputForm.
  3. Environment. Describe the terminal's text capabilities (columns, Unicode, colour) in the environment record and read them in one place, so the REPL, the CLI and show mode stop deciding separately. Honour NO_COLOR in the REPL too.
  4. Fidelity. Add a small set of tests that hold the drawing to the expression, beyond the nine-case golden that exists.

InputForm stays one command away (:form inputform, -f inputform) for copy and paste.

What exists

  • The CLI's StandardForm is 2-D already. renderDrawn sends "standardform" to drawLaid(makeBoxes(expr.json), …), with a comment "the web's boxes, laid out on character cells: stacked fractions, raised scripts" (packages/cli/src/engine.ts:122-126). Every other form goes through the format registry (engine.ts:97-110, :125).
  • The default is InputForm. form: Form = "inputform" (packages/cli/src/engine.ts:224); the README says inputform is "the default" and standardform "the notation the web shows, laid out on character cells", styled on a terminal and plain when piped or with NO_COLOR (packages/cli/README.md:57-63). It is chosen with :form (packages/cli/src/core.ts:191-196), -f/--form (packages/cli/src/command.ts:48) or a config file's form (packages/cli/src/config.ts:32-35).
  • Show mode opts in by capability. present sets session.form = "standardform" when can.interact(env, "keys"), with the comment "Someone at a terminal reads 2-D StandardForm; a pipe keeps the one-line form" (packages/cli/src/present.ts:64-65). That is the docs terminal (packages/components/src/notatio-terminal.ts:262-264). The interactive REPL (packages/cli/src/repl.ts:20-22) builds a NodeHost with no environment and no form default, so it prints InputForm.
  • Two one-line renderers in boxes. toText (Unicode glyphs, script characters where every character has one, fractions a/b parenthesised "wherever the linear form would misread") and toAscii (AsciiMath spelling) (packages/boxes/src/render/text.ts:1-5, :212-218; the tables in packages/boxes/src/render/linear.ts:9-158). Other renderers in packages/boxes/src/render/: latex.ts, mathml.ts, html.ts, markdown.ts (render/index.ts:5-21). None of them lays out in two dimensions; that is packages/boxes/src/layout.ts, which is renderer-neutral and takes a Metrics with a cell flag (layout.ts:1-10, :53).
  • OutputForm means one line. The format OutputForm (aliases outputform, text) encodes toText(makeBoxes(…)), "Not read back" (packages/formats/src/formats.ts:92-101). The wiki agrees: "StandardForm as one line of Unicode text … terminals, outlines, titles, logs" (Syntax-and-Formats.md:111). Wolfram's OutputForm is the 2-D character form, so our name is the opposite of its precedent. OutputForm is in packages/wolfram/src/system-names.ts:5120 as a name only; there is no reference/OutputForm record (a search of packages/*/reference finds only DisplayForm and FormBox near it, packages/boxes/reference/).
  • The cell drawer. drawLaid runs layout(box, DEFAULT_STYLE, metrics) with cell: true, em: 1, text width = code points minus combining marks, and paints items onto a canvas of cells (packages/cli/src/cell-layout.ts:56-80, :262-277); rules and frames join as box-drawing characters (cell-layout.ts:189-208, joints from cell-draw.ts). It returns the text and a baseline row so Out[n]= sits beside it (cell-layout.ts:235-259; packages/cli/src/core.ts:129-136).
  • Colour. styled mode paints weight, slant and colour as SGR: a lone letter italic, a number cyan unless coloured (cell-layout.ts:126-146). The SGR helpers are in packages/cli/src/ansi.ts:5-17, errors are red (core.ts:114), the Out[n]= label dim (core.ts:136), the prompt bold (core.ts:96). stripAnsi exists for tests (ansi.ts:22-23).
  • Colour switches. cli-main.ts sets color = isTTY && !NO_COLOR (packages/cli/src/cli-main.ts:49). runRepl sets color = Boolean(process.stdout.isTTY) with no NO_COLOR check (packages/cli/src/repl.ts:21). present uses env.colour !== "mono" (present.ts:38).
  • The environment record. Environment has engine, interaction, time, surface, form, layout, colour, theme, static (packages/frontend/src/environment.ts:33-46). TTY is surface: ["text"], layout: "grid", colour: "256", form: "AsciiMathForm" (environment.ts:78-89); PIPE is TTY with no engine, no interaction and colour: "mono" (:92-100). nodeEnvironment({isTTY, env}) returns PIPE off a TTY, and on a TTY adds raster for kitty, iTerm2 or WezTerm and sets colour from COLORTERM and TERM (:119-134). Nothing reads Environment.form: the only match for AsciiMathForm outside the record is a symbol table entry (packages/frontend/src/symbols.ts:1126), and grep of packages/cli/src for env.form finds nothing. The record has no column count.
  • Inline images. detectProtocol returns kitty or iterm from TERM, KITTY_WINDOW_ID, TERM_PROGRAM, LC_TERMINAL, only on a TTY (packages/cli/src/node-graphics.ts:30-40); otherwise plots draw on braille cells (packages/cli/src/textual.ts:1-4, packages/cli/src/cell-draw.ts). No sixel support exists.
  • Tests. packages/cli/tests/standardform.test.ts holds nine cases (powers, fraction, radical, nested fraction, sum, binomial, matrix, functions, constants) to standardform.golden.json, regenerated with UPDATE_STANDARDFORM=1; checks that SGR strips back to the plain text and that a pipe gets no escapes (standardform.test.ts:9-15, :17-30, :41-60). present-standardform.test.ts holds two cases and checks a pipe's output is one line (:21-30). packages/boxes/tests/layout.test.ts and layout.golden.json cover the layout.
  • Docs are behind the code. The speculative environments page says terminal typesetting is "the only part of the terminal column with no existing code behind it. The REPL prints notatio (InputForm) meanwhile" (Speculative-Rendering-Environments.md:278-281). The Boxes page already says that with cell metrics layout "writes a small math (stacked fractions, raised scripts)" (Boxes.md:112). Syntax-and-Formats.md:110 calls StandardForm "typeset (KaTeX)", which is the web only.

1. The box inventory

The union has 62 heads: 47 named in the BoxNode type plus the 15 in CONTROL_BOX_HEADS (packages/boxes/src/box.ts:25-41, :52-125); ARITY lists the same 62 (box.ts:133-196). The table sorts them by what a character grid can do. "Built" is what the cell layout does today (layout.ts:347-414 dispatch, :571-596 math), "text" what toText writes (text.ts:79-208).

Class Heads On a grid Built in the cell layout
(a) one line RowBox, TextBox, TextData, StyleBox, TagBox, InterpretationBox, FormBox, TooltipBox, ButtonBox, TextCell, ErrorBox, PaneSelectorBox (first pane) text runs; Greek and operators are already Unicode in the boxes; operators spaced (layout.ts:697-738) yes (layout.ts:350-384)
(a) one line when it fits SuperscriptBox, SubscriptBox, SubsuperscriptBox Unicode script characters when every character has one (digits, + − = ( ), n i up; a e o x down); no x²ⁿ-style guess for the rest yes, else stacked (layout.ts:609-623; tables :761-803). toText has a wider subscript table (h k l m n p s t, linear.ts:147-149)
(a) or (b) SqrtBox, RadicalBox √x is one line only for a one-line radicand; otherwise a bar over the radicand with ┌│√ on the left yes (layout.ts:625-642)
(a) or (b) OverscriptBox, UnderscriptBox, UnderoverscriptBox a repeat bar becomes a combining overline on each digit; otherwise stacked, centred yes (layout.ts:644-666)
(b) more than one line FractionBox numerator, a rule of ─, denominator, each centred, bar two cells wider; no bar when FractionLine -> False yes (layout.ts:597-607). A vulgar-fraction glyph (½) is not produced by the layout
(b) GridBox, and the TagBox formatters Row, Column, Grid, Labeled padded columns, optional frame and dividers as box drawing; spans yes (layout.ts:416-528)
(b) a fence beside something taller than a line ⎛⎜⎝ ⎞⎟⎠ ⎡⎢⎣ ⎤⎥⎦, │, and ╔║╚ ╗║╝ for white brackets yes (layout.ts:443-456, :750-759)
(b) PanelBox, FrameBox, PaneBox box-drawing frame; a PaneBox clips yes (layout.ts:530-565)
(c) cannot be shown GraphicsBox, GraphicsComplexBox, DiskBox, LineBox, PointBox, ArrowBox, RectangleBox, PolygonBox, PolyhedronBox, InsetBox, RasterBox braille or box-drawing cells; an inline image where raster is declared; toText writes -Graphics- cells: leafLines -> drawGraphicsBox (cell-layout.ts:40-42); toText: -Graphics- (text.ts:165-176)
(c) Graphics3DBox, GraphicsComplex3DBox, Polygon3DBox, Line3DBox, Point3DBox, CuboidBox toText writes -Graphics3D-; the cell layout reserves a slot slot; no 3-D drawing in the cell layout that I found, so ? fills it (cell-layout.ts:50)
(c) the 15 control boxes (SliderBox … RadioButtonBox) a one-line text control, driven from the keyboard in a TTY; toText writes -Slider- yes (cell-layout.ts:42-45, packages/cli/src/control-draw.ts, drive.ts)
(c) DynamicBox, DynamicModuleBox, TableViewBox DynamicBox read at the controls' starting values; TableViewBox is paged by pager; toText writes -Dynamic-, -TableView- DynamicBox yes (cell-layout.ts:46-49); table via table.ts, pager.ts
holes TemplateSlot, TemplateExpression ${name} plain (layout.ts:822-824)

Measured exceptions to read as gaps: a "?" stands in for any leaf the drawer has no picture for (cell-layout.ts:50), which names nothing; toText's -Graphics- names the family but not the head. A placeholder that names the head (-Graphics3D-, -DiskBox-) would be one line of code in leafLines.

2. The layout model

This is what Layout does today (packages/boxes/src/layout.ts), stated so a builder can extend it.

A block is {width, height, baseline, items}: a rectangle of cells, with baseline the count of rows from the top to the row text sits on (layout.ts:218-224). On cells the baseline of a leaf is ceil(height / 2) (cell-layout.ts:74), of a grid the same (layout.ts:524). across places blocks left to right aligned on their baselines (layout.ts:333-345), with a one-cell gap (CELLS.rowGap = 1, :200-212) where spaceBetween asks for one (around a binary operator, after a comma, :697-748).

Rules.

  • Text. A token row is one run, spaced by spaced(): a space either side of a binary operator in SPACED unless it begins the row or follows another operator or an opening fence; a space after a comma (layout.ts:697-738). Width is visibleWidth: code points, minus combining marks, minus the invisible function-application and times (cell-layout.ts:53-56).
  • Fraction. width = max(num, den) + 2 with a bar, max without; numerator and denominator centred by floor((width - w) / 2); height = num + 1 + den; baseline is the bar's row (layout.ts:597-607).
  • Script. If every script character has a Unicode form, the script is appended to the base as text and the block stays one line. Otherwise the superscript block sits above the base, the subscript below, both at the right edge; baseline is the base's (layout.ts:609-623).
  • Radical. A bar of ─ over the radicand, √ on the baseline row; a radicand more than one row tall gets ┌ and │ down the left; an index is a small superscript or a block to the left (layout.ts:625-642).
  • Big operators. UnderoverscriptBox stacks over, base, under, centred (layout.ts:644-666); Sum(1/k^2, {k,1,n}) draws n above ∑ k⁻² above k = 1 (golden sum).
  • Grid. Column width = widest cell, row height = tallest; centred by default, Column left; gutters of 2 cells between columns and 1 row between rows (CELLS.gridColumns = 2, rowGap = 1, layout.ts:200-212, :458-528).
  • Fence. A string (, [, |, ⟦, ⟧ next to a block more than one row tall becomes a column of bracket pieces as tall as the tallest sibling (layout.ts:443-456).

Worked examples, copied from packages/cli/tests/standardform.golden.json and present-standardform.golden.json:

 a + b
───────           (a + b)/(c + d)
 c + d

⎛n⎞ + 1           Binomial(n, k) + 1
⎝k⎠

⎛1  2⎞            Matrix([[1, 2], [3, 4]])
⎝3  4⎠

 ─────
√x + 1            Sqrt(x + 1)

Width, not yet built. The layout has no maximum width: a result takes the room it needs. The width and height of CellOptions bound a figure, not a formula (packages/cli/src/cell-draw.ts:26-33); present passes 60 as the width to textOf only for plots (present.ts:43-44). Show mode hands the terminal's columns to a driver, not to the layout (packages/cli/src/drive-node.ts:25, packages/cli/src/present.ts:35). Proposed rule: lay out, and if block.width > columns - labelWidth, try, in order, (1) toText of the same boxes if it fits on one line, (2) InputForm, which the terminal soft-wraps. A fraction does not wrap inside a block. The Out[n]= label width is the one formatOut computes (core.ts:133-136), so the available width is columns minus that.

Character width. Measured: width is a code-point count with combining marks removed (cell-layout.ts:56, :98); a wide East Asian character or an emoji counts as one cell, though a terminal gives it two. Assumed: math output rarely carries them, but TextBox and strings can. Rule to add: width is wcwidth-style, 0 for combining marks (\p{M}) and zero-width, 2 for East Asian Wide and Fullwidth (Unicode East_Asian_Width W or F), 1 otherwise; ambiguous-width characters (Greek, ∑, box drawing) count 1, which is what most terminals do and is the setting the user's terminal may change. The canvas stores one string per cell, so a wide character occupies its cell and the next one is left empty.

3. One line or many

Two names, kept, with one decision to record.

  • StandardForm is the layout that may take several rows. It is what the web shows, drawn on cells (engine.ts:123).
  • OutputForm is the one-line serialiser, never more than one row (formats.ts:92-101). It exists for places that need a line: a pipe, a title, a log, a search string (Syntax-and-Formats.md:111).

Recommend against a single renderer with a maxHeight option. Reasons: the one-line serialiser makes choices the layout does not (it parenthesises a non-unit operand of / and of a script, text.ts:54-56, :128-130; it has a second alphabet in linear.ts), and merging them couples the pipe format to the layout's spacing. What the environment can say instead is a width, and the fallback chain in §2 does the rest. The environment record gains two fields beside layout: "grid" (assumed names): columns: number | undefined and unicode: boolean; can.fit(env, width) is not needed.

Naming cost: our OutputForm is Wolfram's OutputForm flattened. A reader who knows Wolfram expects the 2-D one. The wiki already decided (Syntax-and-Formats.md:111), so this page does not reopen it; the open question below asks whether to.

The form field of Environment ("AsciiMathForm" on TTY, environment.ts:84) is unused. Either remove it or make it mean the fallback form for the environment (TTY: StandardForm; PIPE: InputForm), read by the CLI when no :form was chosen. The second gives the default in §Recommendation 1 a single home.

4. Colour and copy

  • Where colour is used today. The Out[n]= label dim (core.ts:136), the prompt bold (core.ts:96), errors red (core.ts:114, present.ts:48), a variable italic and a number cyan in StandardForm (cell-layout.ts:130-146), a StyleBox's weight and FontColor (cell-layout.ts:104-124, :129-135; named colours and #rrggbb as 24-bit), figure/braille marks (cell-draw.ts:30-32).
  • Heads, strings. Not coloured. Assumed useful: a head name before ( in a Head(args) fallback, and a string in TextBox with ShowStringCharacters. Keep to the 16 basic colours plus the 24-bit FontColor the layout already writes, since Environment.colour distinguishes mono, 16, 256 and true (environment.ts:19) but the drawer writes 24-bit codes regardless (cell-layout.ts:119-122). Assumed: a 16-colour terminal garbles those; downgrade 38;2;… to the nearest basic code when colour is 16.
  • Switches. NO_COLOR is honoured by cli-main (cli-main.ts:49) and not by the REPL (repl.ts:21); a pipe is PIPE, colour: "mono" (environment.ts:92-100), and present and the CLI pass color: false for it (present.ts:38). Proposal: one function colourOf(env, signals) returns mono for a non-TTY, for any non-empty NO_COLOR, and for TERM=dumb; the REPL, the CLI and present call it. The rule: the text with the SGR codes stripped is identical to the text without them (standardform.test.ts:41-48 already holds that for nine cases).
  • Copy. StandardForm on a grid cannot be selected as one line. :form inputform re-prints in the retypeable form, -f is repeatable (command.ts:48), --json returns the InputForm, TeX, MathJSON and WL together (command.ts:357), and % substitution is in the session (engine.ts:3; not re-read for its syntax). Proposal: when StandardForm printed a block taller than one row, a dim hint line says :form inputform once per session, and :copy writes InputForm to the clipboard through OSC 52 where the terminal accepts it (assumed; OSC 52 support varies and is not read anywhere in the tree).

5. The fidelity rule

Rule. The text rendering must not show something that reads as a different expression. Concretely: no parentheses dropped on flattening; a script is written in Unicode only when every character has a Unicode script form (scriptChars, linear.ts:151-158; raised, layout.ts:805-808), else it is laid out raised or written ^(…); a fraction that goes to one line is a/b with a non-unit operand in parentheses (text.ts:19-30, :128-130); a result that does not fit its width goes to a rendering that is still faithful, never to a truncated one.

Measured on the code.

  • toText holds the rule by construction: unit() parenthesises anything that is not a token or an already fenced row (text.ts:54-56).
  • The cell layout draws a fraction as two rows with no parentheses, which is faithful because the stacking carries the grouping; a fraction inside a script is drawn stacked, since / has no script form (layout.ts:609-623).
  • Case to test, not a finding: a script whose base is itself scripted. toText writes (x²)ⁿ (via unit), and the cell layout's scripted lays the base and appends the script, so x² then ⁿ reads as x²ⁿ. Whether makeBoxes hands it a fenced base is not something I read.
  • x^(n+1): a RowBox script of n, +, 1 is written by plain as n + 1, which has spaces and so no script form, so it is stacked (layout.ts:697-738). Faithful.
  • The golden shows the engine's canonical order and form, not what was typed: x^2 + y^3 prints y³ + x², 1/k^2 prints k⁻² (standardform.golden.json, powers, sum). That is the result of evaluation or canonicalisation, not of the renderer, but a reader may take it for the renderer's doing.

Tests that hold it (proposed; the repo's rule: a drawing is a golden, regenerated behind an UPDATE_* flag, never toMatchSnapshot, AGENTS.md "Testing"; the flag exists as UPDATE_STANDARDFORM, standardform.test.ts:11).

  1. Corpus golden. Run the shared box corpus (packages/boxes/tests/corpus.ts) through drawLaid at width 80 and 40, compared with a golden; regenerated with UPDATE_STANDARDFORM=1. A change to a rendering is a visible diff. (Assumed: the corpus is the right set; I did not read it.)
  2. Single-row agreement. For every corpus entry whose cell layout is one row, the stripped text equals toText of the same boxes after collapsing spaces. This is checkable without a parser and catches a dropped parenthesis or a script silently written another way.
  3. No-fallback-silence. For a script with no Unicode form (x^(a+b), x_k with k in the table and x_q not), the layout is stacked, or toText writes ^; never a mixed Unicode form missing a character.
  4. Fit. For widths 20, 40, 80, every result is no wider than the width, or is one InputForm line.
  5. Parse back, where meant. toText is "not read back" (formats.ts:92), so no parse-back test for StandardForm. Where a golden for the one-line rendering of a reference example should read back, the reader to use is AsciiMath's, through toAscii and the asciimath format (formats.ts:104-111), whose round trip is pinned in packages/formats/tests/asciimath.golden.json (Syntax-and-Formats.md:122-126); the CLI's text rendering is not that.
  6. Colour strip. SGR-stripped equals plain, for the corpus (extends standardform.test.ts:41-48).

A head's expected value stays in its reference examples; none of these tests touch it (AGENTS.md "Testing": golden JSON "only for output that isn't a head's value").

Plan for a builder

The smallest first slice is the default and the width fit, in packages/cli only, plus one environment field.

  1. nodeEnvironment gains columns (from the caller; the function takes signals as arguments, environment.ts:113-119, so NodeSignals gets an optional columns). Environment gains columns?: number. No other package changes.
  2. colourOf in packages/frontend/src/environment.ts (or beside ansi.ts): non-TTY, NO_COLOR, TERM=dumb give mono. repl.ts:21 and cli-main.ts:49 call it.
  3. Default form. runRepl and cli-main set defaults.form = "standardform" when stdout is a TTY and no config form is set. The one-shot CLI's piped output stays InputForm. Config form wins (config.ts:32-35).
  4. Width fallback. In Session.renderDrawn (engine.ts:344-350), when the drawn block is wider than columns less the label, return toText(makeBoxes(json)) if it fits, else InputForm. Session gains a columns field next to color (engine.ts:225).
  5. Placeholder that names the head for an undrawn leaf in leafLines (cell-layout.ts:50).
  6. Tests. Extend standardform.test.ts with the fit test (4) and single-row agreement (2) over the nine existing cases plus a few that overflow; regenerate standardform.golden.json with UPDATE_STANDARDFORM=1. Packages touched: cli (engine, repl, cli-main, cell-layout, tests), frontend (environment). boxes is untouched.

Second slice, after answers 3 and 4 below: wide-character widths, colour: "16" downgrade, the head-name colouring, and the corpus golden at several widths.

Open questions, with recommended answers

  1. Is StandardForm the REPL default on a TTY? Recommend yes. The code and the README already offer it, the docs terminal already does it, and the idea's complaint ("falls back to InputForm") is only true because the REPL's default was never changed. Risk: the ambiguous-width and font issues (box-drawing and bracket pieces ⎛⎜⎝) on terminals with a font missing them. Mitigation: :form inputform and a config form.
  2. Keep our OutputForm as the one-line form? Recommend keep. The wiki has it (Syntax-and-Formats.md:111) and the format is registered with that meaning. Cost: it is Wolfram's name for the 2-D form. If the owner wants parity, rename the one-line format TextForm and let OutputForm alias the 2-D layout; that is a rename across formats.ts:92-101, the CLI aliases (engine.ts:143) and the wiki table, best done as a codemod (AGENTS.md "Git hygiene").
  3. Is the fallback on overflow toText, then InputForm? Recommend yes. It keeps a faithful, shorter form before the retypeable one. Alternative: scroll horizontally in a pager (pager.ts exists for tables); more work, and not for a first slice.
  4. Wide characters: build wcwidth ourselves or take a standard library? Recommend a small standard library (string-width or the platform's Intl.Segmenter for graphemes plus a width table), per AGENTS.md ("common/standard libraries"), added to cli only. Assumed: string-width is not yet a dependency; I did not read packages/cli/package.json.
  5. Sixel? Recommend no. Kitty and iTerm are detected (node-graphics.ts:30-40); sixel support needs a query of the terminal (DA1) that a first slice should not add. Plots already draw on braille.
  6. Copy button. Recommend the dim hint now and OSC 52 :copy later; the hint costs one line of output, the clipboard needs terminal cooperation.
  7. Should the environment's form be the default's source? Recommend yes (see §3), replacing the unused "AsciiMathForm" on TTY with StandardForm, PIPE with InputForm.
  8. Update the wiki pages (Speculative-Rendering-Environments.md:278-281, Syntax-and-Formats.md:110) and remove the item from Ideas when the first slice lands. Recommend yes.

Measured versus assumed

Measured (read): the REPL default form and the opt-in paths (engine.ts:224, core.ts:191-196, present.ts:64-65, repl.ts:20-22); the cell layout's rules for fractions, scripts, radicals, stacks, grids, fences (layout.ts); the drawer and its width function (cell-layout.ts); toText, toAscii and their script tables; OutputForm as one line (formats.ts); the environment presets and detector (environment.ts); that Environment.form has no reader; NO_COLOR in cli-main.ts and not in repl.ts; image protocol detection; the tests and goldens named; the head count (62: 47 named, 15 controls; ARITY has 62 lines, box.ts:133-196).

Assumed (not run, not read to the end): that the goldens match what the code produces today (nothing was executed); the parenthesisation of a scripted base by makeBoxes; the contents of packages/boxes/tests/corpus.ts; that no 3-D drawing exists in the cell layout (I found none by search); how terminals render ambiguous-width and bracket-piece characters; the support for OSC 52, 16-colour downgrades and string-width; the % history's syntax. Not determined: the cost of the width fallback on very large results, and what the CLI does with a Session that has no columns today (assumed: nothing).

Home

Project

Working

Design: every design page, by subject

Clone this wiki locally