DSH HUB
HomePlugin StorePlugin PacksCommunityRankingsResourcesPublish Guide
Plugin source
Back to catalog

huanglianqi /

huanglianqi/dsh-math-snippets

Verified

LaTeX snippet expansion in the DSH web composer: Tab expands a trigger into a formula template with the caret in the first hole, then walks the rest. // expands with no Tab at all.

★ 1 Stars0 Forks0 IssuesN/A Community rating0 Confirmed installs
View on GitHub
READMESource: main@0227e9d6

dsh-math-snippets

LaTeX-Suite-style snippet expansion in the DSH web composer. The trigger table is built from Obsidian LaTeX Suite's own defaults — the same triggers, the same auto/Tab split, the same tabstop numbering — so the muscle memory carries over.

$//        →  $\frac{}{}$          no Tab: an auto snippet expands as it is typed
$sum       →  $\sum_{i=1}^{n} $
$ff<Tab>   →  $\frac{}{}$          this plugin's short pair, on Tab or on its own
mk         →  $$                   outside math, caret between the dollars
dm         →  $$⏎⏎$$               display math, caret on the empty line
$@a        →  $\alpha              LaTeX Suite's Greek spelling, also auto
$sr        →  $^{2}                and `_`, `cb`, `rd`, `**`, `->`, `<=`, `!=`, `oo` …

Tab then walks the tabstops, in the order LaTeX Suite numbers them: $ff<Tab>3<Tab>4<Tab> puts the caret in the numerator, then the denominator, then after the fraction.

Install

dsh plugin --profile <profile> add dsh-math-snippets

Restart the harness afterwards (a profile's bundle list is read at boot), then reload the GUI page.

Installing from a checkout instead

Add a row to the profile's cordis.patch.yml:

- insert:
    - id: math-snippets
      name: 'file:///path/to/dsh-math-snippets/index.js'

LaTeX Suite parity

The table is not a hand-written approximation: it is generated from LaTeX Suite's src/default_snippets.js, keeping every entry that can be expressed as a plain string snippet. 142 tags, 138 of them auto, 4 left on Tab.

Same the trigger vocabulary (//, @a, sr, cb, rd, _, **, ->, <=, >=, !=, oo, sum, lim, dint, hat, bar, vec, norm, ceil, floor, text, …), which triggers are auto and which need Tab, w word-boundary flags, and the $0,$1,$2 tabstop order
Same the ${1:\infty} placeholder syntax parses (the default text is dropped — see below)
Different the four LaTeX Suite deliberately keeps on Tab — par, \sum, \prod, \int — stay on Tab here too, which is the point
Different dm does not prepend the newline LaTeX Suite's does. In a note you always want display math on a line of its own; in a chat composer the common case is that you are already on an empty line, and that newline would leave a blank first paragraph
Different no conceal/render of formulas while typing, no auto-enlarging brackets, no backwards expansion, no regex/visual/function snippets, and ${1:x} placeholder text is not preselected: the shell publishes no selection seam, so a hole is empty instead

Snippet table

On first start the host half writes an editable table to ~/.dsh/math-snippets.json containing every built-in trigger, and serves the effective table at /plugin/math-snippets/table. The browser half re-reads that route once a minute, so edits take effect without a restart or a page reload.

Each entry is { "body": "...", "auto": true }:

Field Meaning
body what the trigger becomes. $0, $1, $2 … are holes and Tab visits them in numeric order; ${1:\infty} is accepted too. A body with no hole leaves the caret after the text.
auto true expands on the keystroke that completes the trigger, with no Tab. This is the default here, as it is in LaTeX Suite.
open true allows the trigger outside math (used by mk and dm, which are how you enter math). Everything else only fires inside $…$, $$…$$, \(…\) or \[…\].
word true also requires a delimiter after the trigger, so dm inside dmx stays a word. Taken from LaTeX Suite's w flag.
desc free-form note for yourself; unused by the expander.

Example — add your own:

{
  "snippets": {
    "capm": { "body": "$E[R_i]=R_f+\\beta_i(E[R_m]-R_f)$", "auto": true },
    "npv":  { "body": "\\mathrm{NPV}=\\sum_{t=0}^{n}\\frac{CF_t}{(1+r)^t}" },
    "par":  { "body": "\\frac{\\partial $0}{\\partial $1}", "auto": true }
  }
}

A tag in the file overrides the built-in of the same name field by field, so an entry that only restates a body keeps the built-in's open/auto/word switches — and an explicit "auto": false turns one off. That last line is the escape hatch: if an auto trigger fires when you were still typing a word, switch that one tag to Tab.

Two kinds of tag

Tag Kind Matched as
ff, sum, alp bare word the whole word before the caret (xff is the token xff, not ff)
//, @a, _, sr literal run the exact characters before the caret; longest table entry wins

A tag is a bare word when it is all letters, and a literal run otherwise, so //, _, **, ->, @a and :e all work. Two carve-outs keep literal runs out of URLs, mirroring the shell's own / rules: a slash directly after another slash does not fire (/// is dead), and a slash pair after a scheme: colon does not either, so https:// never arms the fraction.

How it works

Four seams the shell already publishes, no DOM editing and no fork of the composer:

  1. The caret comes from shell.projection, the composer's own projection product ({ detectText, clipboardText, selection, caret }). caret is an offset into detectText — the same coordinate space the input span uses — so nothing is measured from the DOM.
  2. The replacement is dispatched as the scoped input event the shell installs for its own input-trigger sources: actx.bail(actx, "slash/input-insert-text", { text, span }) with actx = sessions.scope(sessionId) and span = { start, end, draftRev }. The shell refuses a span whose draftRev no longer matches, so a stale keystroke can never corrupt the draft.
  3. The hole needs no caret API, because the shell always leaves the caret after what it inserted. So the expansion goes in tail first, then head at the collapsed point where the tail starts — the head pushes the tail right and the caret ends up exactly in the hole. This is why the second insert uses start === end.
  4. Typing without Tab rides keyup, not a draft subscription: by then the character is committed and the shell has re-projected, and it is an ordinary event rather than a notification raised inside the editor's own update, so the insert travels the same scoped path a Tab would have taken.

Behaviour and limits

  • Auto by default, Tab available always. Every tag except par, \sum, \prod and \int (the four LaTeX Suite keeps on Tab) expands as it is typed; Tab expands any of them and is never needed for the rest.
  • Tab is only consumed when something actually applied. Every other keystroke, and every trigger that does not match, passes through untouched — including Tab's normal meaning.
  • Tab walks the holes. The next hole is re-found each time — the walk searches forward from the caret for the literal that sat between the two holes in the body — so typing any amount in a hole shifts nothing, and an expansion typed inside a hole still wins over the walk. Moving the caret back before the hole you were placed in, or a draft that carries a file chip (detect and clipboard coordinates diverge there), stands the walk down and hands Tab back. The walk is not free: the shell publishes no bare caret move, so each jump is an edit that writes back the exact text it replaces — the draft ends up unchanged, but the jump costs one undo step.
  • Prose is safe. A snippet without "open": true never fires outside math, so typing sum in a sentence does nothing. $ glued to a digit is read as a currency sign, so costs $5 and $10 is not math. Code spans and fenced blocks are skipped whole.
  • A LaTeX command is never a trigger. The word path refuses a trigger directly behind \, /, @ or #, so typing \operatorname cannot fire op. Only the three explicit literal tags \sum, \prod and \int match a backslash, and they are Tab-only.
  • Auto can still surprise on a longer word. Typing par is fine (Tab-only), but a bare word that starts with an auto trigger expands at the trigger, so integral would become \int…egral. word: true only guards the case where a character already follows the trigger. Fix it per tag with "auto": false, or press Cmd/Ctrl+Z — the undo step restores the typed trigger.
  • The whole word is the trigger. xff yields the token xff, not ff: a word snippet can never fire from the middle of a word. A literal trigger still fires there, because x// is a fraction in LaTeX Suite too.
  • $ in a body is a hole marker. A literal $ cannot currently be written in a body, and ${1:default} loses its default text.

Diagnostics

curl -s http://127.0.0.1:<port>/plugin/math-snippets/status   # mounted + counts
curl -s http://127.0.0.1:<port>/plugin/math-snippets/table    # the effective table

Loopback Host only.

Tests

npm test

62 assertions, no harness and no browser. Two layers: the keystroke paths run against a fake composer that mirrors the shell's span semantics (span replacement, caret after the insertion, collapsed inserts, stale-revision refusal), so hole placement, the tabstop walk and the auto path are asserted at the level that matters — the resulting text and where the caret ends up — and the seat itself is then rendered against a fake document and fake shell, so the dependency wiring, the composer gate, the capture listeners and their cleanup are exercised too.

Compatibility

Written and verified against dsh 0.1.5-rc.2 (client packages 0.1.5-rc.2, DSH Desktop 2.0.13). The projection product and the slash/input-insert-text contract were read out of that version's sources; a newer shell may move either.

License

MIT

—/ 5

No ratings yet

Verified DSH bundle

Commit 0227e9d685a9

Community comments

No comments yet. Be the first to write one.

DSH HUB

A community index for DSH plugins. Not an official GitHub or DeepSeek AI product.

CommunityResourcesAPIAbout