Skip to content

Prototype: ReactiveScript and the Solid ecosystem (router 2.0, TanStack, SSR) on top of upstream PR #69 - #1

Draft
yumemi-thomas wants to merge 227 commits into
pr69from
claude/magical-bell-pfhb1w
Draft

yumemi-thomas wants to merge 227 commits into
pr69from
claude/magical-bell-pfhb1w

Conversation

@yumemi-thomas

@yumemi-thomas yumemi-thomas commented Nov 19, 2026 •

Copy link
Copy Markdown
Owner

Prototype, for exploration, on top of upstream PR solidjs#69. The base branch pr69 is that PR's branch (Brenley Dueck's commit a54830a, unchanged) merged with this fork's main, so this diff shows only the prototype. Nothing here is proposed upstream yet; spec rules it adds are marked proposed, and the design choices taken while prototyping are recorded with their reasons in spec/09, Decisions §9.

Every proposal stays at the JS boundary: a declaration word, a check on declarations, or a runtime hook for the host. Nothing adds an unchecked escape inside .rx (CN-5). Each language change is in both ref/ and the crate, which agree.

Router (@solidjs/router 2.0.0)

  • Hooks need the router (MO-12): export context router, uses(router) on hooks, provides(router) on createRouter or a provider; provider order checked. S-module-7.
  • Typed params (MO-11g/h, MO-8d): a page declares props: { params: { id: string } }; a route whose path lacks id is a type error. Nested routes are typed with their parents' params, alternatives (["/a/:id", "/b/:id"]) included. S-module-8.
  • Layout routes (MO-11i): nests(children in component), so a layout's errored/provide reach its routes. S-module-9.
  • Exported routers and typed links (MO-11j/m), typed search params without an import cycle (MO-8g). S-module-10, S-module-17.
  • Data layer: a router query declared with the rows of the fn it caches, a formAction bound to a <form>, useSubmissions. S-module-19.
  • SSR (SV-4a): the server runtime's createApp({ request }); the Vite suite builds client and server, serves per request, hydrates and navigates in Chromium.

Other libraries

  • TanStack Query: holds fails(…) on result fields (MO-10f), declared callbacks as handlers (MO-10g). S-module-11, S-module-12.
  • TanStack Virtual / headless table: render props {view (row, i) { … }} (MO-11k), ref on an element as a handler, renders (MO-11l). S-module-14 to 16.
  • Libraries with their own root (toasters): renders(root, ctx…) (MO-11n). S-module-18.
  • .d.rx: optional parameters (MO-8e), version pins (MO-8f).

Types, runtime and tooling

  • rx-tsc checks each .d.rx against the module's real types (TY-6), and a production Vite build runs it (TY-1d): a type error fails the build.
  • A declared JS error class crosses SSR as itself, identified by its declaration (SV-5a).
  • rx-declare drafts a .d.rx from types, then runs the package to settle uses/provides.
  • rx-verify runs a declaration: exports, uses, provides, pure, and with a probe, holds/fails on every result field. It found that TanStack Query 6's isFetching holds on first load.
  • Harness: the Vite suite finds a plugin that loads .rx; the server runtime warns on two copies of @solidjs/web.

Results

  • ref/: 145 scenarios; compiled and oracle runtimes 145/145; types 196/196; SSR 138 (7 older skips), on ref/ and the crate's output.
  • Crate parity: 449/449 programs, normally and client-only; projection 236/236, lowering 232/232; cargo test passes.
  • Tooling: language server 16, rx-tsc 7, rx-declare 15, rx-verify 16.
  • Vite suite 22/22; compiler/test/ssr-router.mjs 5/5.

Fixed along the way

  • ref/scripts/solid-next.sh (the next branch is gone), a literal escape in compiler/src/prelude.ts, rx-tsc on .d.rx in include, examples in guide/diagnostics.md that didn't hold.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BVdYZqRoBTRfgj2qMzfd7X

claude and others added 30 commits November 14, 2026 03:40
eval: regrade .rx apps on Solid next with #4840 and #4841 fixed
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
A write that lands after its page was left is not judged on the next visit
(Ryan 2026-11-14: late writes are out of scope). A4 was only applied to rows.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
…XState references follow it

Ryan 2026-11-14: with two accepted writes to one field, the server's last answer is the
only correct value at rest. --r15 drops allowance A2; FUZZ_ROUND=--r15 sh harness/grade.sh.
All six desk9 references grade 0 under it (20 seeds; React, Exact, XState, Foldkit and
vanilla also 0/60 on seeds 21-80).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
…erence, fuzz seed 45)

Also replay.mjs: DW=<n> widens the detail text it prints.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
…ence (grades 0)

Brief frozen 2026-11-14 (Ryan). 26 visible, 45 hidden, 26 overlap tests; fuzz-board.mjs
with oracles for duplicates, headings, counts, the work-in-progress limit, the detail,
and settled state. TASK=board sh harness/grade.sh grades it. The fuzzer catches two
planted bugs in the reference (limit off by one, an accepted toggle not kept).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
…el under a pending guess

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
…rite visit

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
…15 prereg draft

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
… answered add's placeholder while work is in the air

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
…, minimal trail)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
…s/solid#4874, rx runtime catchRead, pending probe)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
…); guide: one store over the server

- Runtime: `catchRead` with no observer (a component fn called from a handler or
  action) is a snapshot (04 R-8): the committed value, or the alternative on a
  matching failure. A fresh error boundary there had no value while a refresh
  was out and threw NotReadyError (S-catch-snapshot).
- Checker (ref/ and crate): ARGUMENT_COUNT (06 AC-1f) for a call of a declared
  action with the wrong number of arguments; `props` in `component C()` is
  UNBOUND (03 RG-4b). AST gains `noProps`. S-check-1 rejected variants, a
  check.rs unit test, guide/diagnostics.md sections.
- pending? inside a held `loading on(sel)` region reads false: Solid's, under
  05 DV-22 (plain-Solid repro saved). Guide and GUESS_WHILE_LOADING text say
  to test `pending? x` in the handler too (R-12).
- guide/async.md: "One store over the server".

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
…k shown column), fuzz B3 declined count hold, repro for solidjs/solid#4887

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
…on, two Solid repros

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
…and by key; record seed)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
…(seed 28); repro for solidjs/solid#4897

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0117BoWLtkg281f1NkGEDjCt
yumemi-thomas and others added 30 commits November 21, 2026 20:59
…n, or a string type argument, is skipped whole, so a `}` in it keeps a generic call a call

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ot there instead of printing ok

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…d.d.rx (.mjs and .jsx alike); the checker, rx-tsc, the boundary check and rx-verify read the same names in the same order

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…rt that uses c alone and fails without it; a component's handler, tracked and pure props are run, so strict mode can accept them without trusted; records are version 4

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…read is different by content (rx-tsc --rx-inputs: files hashed as read, paths looked for, folders listed), not by mtime

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ports MODULE_CYCLE at its own import, not only the module where the cycle closes

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ssion may hold — aliases, record and array literals (a static index precisely), destructuring, ?:, ??, closures, fn returns, dynamic keys; drafts, renders values and undeclared methods follow it, and an exported record holding a renders value is refused

DRAFT_ESCAPES now sees [d], { d }, let [a] = [d], c ? d : x, d ?? x and what
an .rx fn or a closure returns; RENDERS_ESCAPES sees ?:, ??, closures, fn
returns, dynamic keys and exports, and arr[1] of [table, 5] is 5 (R2);
SNAPSHOT_ARGUMENT and SNAPSHOT_KEPT see an undeclared method through a
record, an array, a branch or a closure. Ref and the crate share the design.
Scenarios gain a plain `-- accept` block, checked on both sides.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…tter per field (__rt.lazy), so only the fields read are touched; a result consumed whole (spread, shown, passed to a component or a JS fn, written) carries every held field's obligations, through MO-10k

The checker counted q.isFetching after q = orig as a read of isFetching alone,
while the compiled memo read every field, so a failing data crashed SSR.
Codegen is the side changed: the checker's per-field count is now what runs.
Aliases, records and closures only move the result; a field read through
them is that field's read; consuming it whole touches every held field.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…erns (./*), as the crate's packageDeclaration does

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…cript's for … of, a statement, so its body is the run it is written in: it may await in an action and return from the run, which a closure handed to forEach cannot

The run's checks (AC-1c..g) read a loop as a body that may not run, run once,
or run again over what its last turn wrote, so `total := total + x` in a loop
is STALE_OWN_READ and `total := t => t + x` composes (W-6). Both parsers,
checkers, emitters and projections, the interpreter (each turn a scope of its
own) and the oracle; the tree-sitter grammar already read it as JavaScript's.
S-surface-10 runs it on the interpreter, Solid, the compiled module, the
server and the types.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…7, 18, 34): no value is followed — a result with held fields is read a field at a time (RESULT_AS_VALUE), a draft is a place whose reads are copies, a renders value and an instance are confined by mention

Three rounds answered "where may this value go" with one analysis of what
values an expression may hold, followed through aliases, records, calls and
closures (values_of). Each review found a way through it and a program it
refused for nothing: with first-class functions and no types the question has
no answer that is both sound and precise. 08's stance is the stop (sound,
over-approximate, modular, silent about JS beyond its declaration), so each
rule is now one whose soundness is local (08 CN-7, 09 MO-10k):

- RFC 06 (MO-10f): a JS result whose fields hold or fail is not a value. Its
  name stands only as the object of a field read, as another name given by
  an equation, or compared or tested; anything else is RESULT_AS_VALUE. So
  every read of a held field is a path the checker sees. The closure prop,
  the provision, the export and the source write that were accepted and
  threw in the server's render are refused by that one rule.
- RFC 09 (MO-10i): a draft is a place. A read of it compiles to a copy
  (__rt.copy), except where it goes straight back to the library; its name
  stands only in its own callback (not in a closure made there, nor after an
  await); and the function that hands it is called by its own name with the
  callback written in the call. Nothing that only reads is refused any more.
- RFC 17, 18 (MO-11l, MO-11o): a value may hold a part of a renders value
  when its expression mentions its name, unless what is written makes a
  primitive; it leaves its region through none of the ways out (a prop, a
  provision, an export, a source, a thrown value, a call that may keep it).
  A renders prop is given only what the checker can name the components of.
- RFC 14, 34 (E-10, E-5a): a field read, an index or a method call on
  anything that may be the instance is a read of what may change; calling
  the closure a name is bound to reads what its body reads (E-10's own gap:
  `f = () => n; createLabel(f())` was accepted for any source n).

Removed: values_of / valuesOf and its overlay, held_touched, placed,
bare_draft, undeclared_method_via, the alias propagation of drafts and of
renders values. Added: flows (the names an expression's value may come
from, syntactic), tainted (a least fixpoint over names), the deferred sinks.

Each guarantee is tested as one, not as the cases a review found: the value
is put through every way an expression hands a value on and every pair of
them, into every way out (check.rs tests and ref/units.ts, the same matrix);
compiler/test/drafts.mjs runs the draft matrix against a library that
revokes its draft, and Solid 2's own store; compiler/test/ssr-held.mjs
renders the accepted programs on the server and shows the refused ones are
the ones that threw. S-module-11, 16, 20, 22, 25 are rewritten where the
rules change their programs, and say why; examples/formisch reads a copy out
of Solid's draft and loops over it in Chromium.

Also: `pending` and `fails(…)` written where no record field takes them (a
tuple's part, a union's member) were ignored; they are a PARSE error.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…laration's own place — its named package and the path below, else the path below the nearest package.json, else the file's name — never from the entry; two that would be named alike in one program are DUPLICATE_WIRE_ID

The id of a class in an unnamed package was kept from the project root, and
the root was the folder of the nearest package.json above the entry: with
unnamed scopes nested (`{"type":"module"}` in app/, app/web/, app/server/) the
same class was `web/err.d.rx#Gone` from one entry and `err.d.rx#Gone` from
another, and two classes could both be `err.d.rx#Gone` in one build.

An unnamed package has nothing to tell its classes from another unnamed
one's, and the compiler no longer guesses from where the build started: the
id depends only on the file and the package.json files above it, and a
program in which two declarations would name a class alike is refused where
they are loaded, naming both and the fix (a "name" in one package.json). So
in an accepted program every declared class has an id of its own, the same
from every entry, in every checkout, on every machine.

Refused now, accepted before: two unnamed scopes side by side, each declaring
a class of the same name in a file of the same name, imported by one program
(they were told apart by the entry's folder).

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…, as ./x.jsx is (./x.mts and ./x.cts as ./x.mjs and ./x.cjs) — one list, read in one order by the checker and by the tooling

For `./x.tsx` the checker read only `x.tsx.d.rx` and the tooling (rx-tsc's
declarations, the boundary check, rx-verify) only `x.d.rx`, so each ignored
the file the other read: a module declared for one was undeclared for the
other. The crate's `declaration_names`, the reference's `declarationNames`
and `beside.js` now hold the same extensions; test/tsc.js holds the tooling's
list against the crate's and runs both on a `calc.tsx`.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…hat a type-check could take — its folders, and the files of the kinds the project's program holds — so a README or a build's assets under dist/ no longer make the gate stale

The gate is stale when anything the check read is different by content, and
it read every folder it listed. With `"include": ["**/*"]` TypeScript lists
every folder of the project, so any new name anywhere made the next build
start check again: a README, an html or css file a build writes, a hashed js
asset in a project with no allowJs.

A listing now counts the folders in it (a glob may descend into them) and
the files with an extension the program can hold: TypeScript's, .rx,
JavaScript's under allowJs, .json under resolveJsonModule, and package.json
(`rx-tsc --rx-inputs` writes the list it used beside the hashes, `takes`).
A source file added anywhere below is still one; under allowJs so is a .js.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…dule (MODULE_NOT_FOUND, as rxc says), not as a read error (parse error: EISDIR)

The reference's command lines read a module with `existsSync(p) ?
readFileSync(p)`, which a folder answers, then throws on. `ref/disk.ts` reads
a file or nothing, as the crate's `read_disk` does; as the entry and as an
import, on both sides (`a_folder_named_as_a_module_is_no_module` in
ref/units.ts and compiler/tests/cli.rs).

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… 17, 18, 23, 25, 32 and 34, the guarantee each rule now states, what is refused that was accepted, and the limits that remain

The text for the issues of the tracking list (#37), as the second review's
notes were appended to them, kept here until it is posted.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…does with the ecosystem's libraries (Formisch, TanStack Query, Router, Start and Table, Kobalte, Solid's store), each a good program that both checkers must accept, the library's types must type and Chromium must run, and one-line mistakes that must be refused; the scorecard at d6e640e is the baseline (good 31/39, types 48/55, runtime 35/39, bad caught 31/59)

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…r each confined kind — what a JS method answers, keeps or calls is no longer read off the syntax around the call. A result with held fields is read by its declared fields (q.valueOf(), q.toJSON(), q[k]() are RESULT_AS_VALUE); a draft is called by the methods of plain data, a callback given to one is written there with its parameters places of the draft, an iterator is kept as one over copies (anything else DRAFT_ESCAPES); a method keeps nothing of a renders value only when methods(…) names it on the value's own name; a callee read out of a value reads what the closures it holds read. A declared primitive leaves a renders value with its type, a literal's own slot is that slot, and the loop of 07 §5 visits what its list held when it started, as the spec said. compiler/test/confined.mjs keeps the fourth review's four programs: both verdicts, and what each does when it runs

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…r is the library's own object, not a getter-per-field copy of it (its methods are called on it, a field its declaration leaves out is still there); a method a declaration says answers nothing (() => void) is io, so calling it while the view computes is IO_IN_REACTIVE; a handler attribute given a call of something io (onClick={save()}, onNext={table.nextPage()}) is refused with the closure it meant; a type a .d.rx declares is no longer imported from the JS module as a value. bench: the table task declares the table's interface and runs

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… it is, to a prop that renders what it carries renders it in turn — the prop is one that renders in the component's summary and its signature (router: renders provides(…)), and the components it carries are named and checked where the component is applied, under the component's own boundaries around the provider and then the application's; in this module or another. A router wrapped in the app's own Shell is accepted (S-module-22, bench router-provider); what the Shell is given must still be a value whose components can be named

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…s fields' rows — a component's prop whose type is a record with field rows (query: { data: pending fails(Error) T; … }) is read in that component as the result is, a field at a time with each held read checked, and a result (or a record of held fields below one) is given whole to such a prop when its type marks every held field at least as the declaration does; the interface is in the signature (query: takes(data pending fails(Error))), and the row words are dropped from the type TypeScript sees. A query passed down to a child is accepted, typed and run (S-module-11, bench query-forward); a prop that takes the field for plain data still refuses it

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…on or a local whose name is written only as the argument of a function that hands that position a draft is that function's callback: its parameter is the draft, its body the callback's, and it is shared by every handler that gives it to the setter (bump = (d) => { d.done := d.done + 1 }; setTodos(bump)). Any other use of the name (called by hand, passed elsewhere, stored, exported) is DRAFT_ESCAPES where it is written; a function that is not such a closure is still refused. A declared fn whose result hands out drafts is a primitive (instantiates inferred), so its setter is io and calling it while the view computes is IO_IN_REACTIVE. S-module-25, compiler/test/drafts.mjs, bench store-updater

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…N_PLACE: an Array's mutating method (push, sort, reverse, splice, …) called on the value of a source, an equation, a prop, a provision, a row of one, or the value a functional write is given changes it in place, telling no reader (rows.push(2); props.items.sort()); DISPOSER_DROPPED: a declared JS call whose result is () => void (what undoes it) written as a statement, or bound to a local nothing uses, leaves its listener behind when the component ends. guide/diagnostics.md has both; bench source-mutation and listener-cleanup

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…th two alike keys are refused (rows from data stay the data's: nothing checks them at run time, as nothing does in Solid); and a component a JS host renders (pages.rx?rx-lib) is checked as 11 CP-4d says it is, applied under a root of its own: a failure, a context or a first load its own view does not handle is UNHANDLED_FAILURE, CONTEXT_NOT_FOUND or the loading error in the host's module (whole-program builds, which know the hosts). bench duplicate-keys, start-route

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…dvice beside its diagnostics (rxc check prints advice lines and lists them in --json; ref/check.ts the same; the two are compared), never an error: ASYNC_WATERFALL for a pure question awaited after another whose answer it does not read. The spec for the round: what a prop renders and takes (09 MO-3, MO-10f, MO-11o), a named draft callback and the methods of a draft (MO-10i), a handler is passed not called (02 SC-4), W-1a, EF-5a, ST-4d, a host's pages (11 CP-4d), what Solid does and does not report of a mismatch (10 SV-7c), and the decision on each mistake the benchmark found unprotected (spec/rfc-revisions-2026-11-22.md); the guides say the same. bench: the scorecard (good 39/39, types 55/55, runtime 55/55, bad caught 57/59), the two that stay red with their reasons, and CI holds every red to a reason

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… in place (five shapes, and what the code made itself stays its own), a handler given a call, a method that answers nothing called in a view, a disposer dropped two ways, two rows with one key — so the scenario suite and the crate/ref comparison hold them, not the benchmark alone; CHANGELOG

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…result is now the library's own object (README, runtime/README, 09, S-module-11, ssr-held.mjs); a declared type's numeric literal is told the same way by the crate and ref/

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…erexamples — a router at its provider, the same through the app's own component, and a page with no boundary at the outlet still refused through it; 13 TY-1c says what a props type is to TypeScript

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…ows/suites.yml) — cargo test; ref/'s interpreter, oracle, compiled, ssr and types runners; the crate against ref/ in each mode (checked, --spa, --whole-program, --lower, --project); compiler/test/drafts.mjs, ssr-held.mjs and confined.mjs. Solid at ref/SOLID_NEXT is built once per pin and kept in a cache

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…ges (bench/baseline-rebased.txt) — the probes a review writes to find the edge of a rule (an inherited valueOf(), a draft's iterator kept, a forEach callback that keeps rows, a callback kept in a record, [f][0]()) are compiler/test/confined.mjs's; a variant the language cannot write (PARSE, ASSIGNMENT, UNBOUND, UNKNOWN_COMPONENT, a TypeScript error alone) is counted apart, as unrepresentable; shadcn's DataTablePagination table={table} is part of the table's good program; each task's .d.rx files are measured against what rx-declare drafts for the same module (hand-written words, and the anys they hold), and --check holds both numbers; a type probe is judged on the line it wrote; query-forward runs a refetch and a failing query, sequential-awaits asserts two loads in flight at once (its loaders as rx-declare drafts them, io). Eleven tasks added: @solidjs/router 2.0's data layer, TanStack Query's mutation with dehydrate/hydrate, Formisch field arrays, a Kobalte Select with object options, solid-primitives with a TanStack Virtual adapter on an element ref, a store's draft given to a JS helper, an optimistic todo, a selection as a projection, state derived through an effect, a keyed list edited, a context with its provider component. 24 tasks: good 68/72, types 103/107, runtime 97/106, bad 84/106, unrepresentable 8/8, 437 hand-written .d.rx words, 195 anys

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants