Routines that many programs need (a mean, a gap penalty, the
Lennard-Jones potential, the minimum image convention) were rewritten
and re-verified each time, and there was no way to open one and see
exactly what it does. They now live in pplus-library, pinned here as
the submodule library/, and a program brings them in with USE.
- USE module / USE module.Routine (pseudo-study resolveUses): each
routine is studied once in its own file (its own USE lines resolved
the same way, cycles refused) and handed to the checker as a library
routine; unknown-library, unknown-routine, library-clash, use-place.
- A run steps over a library routine (one step at the call, one at the
return, none of its lines); a broken contract is reported at the call,
naming the routine. The prover uses its REQUIRE and ENSURE at the
call and does not prove it again; it is not a unit of the program.
- pseudo-to-mojo writes each library routine the program calls once,
dependencies first, labeled as coming from the library.
- src/library: the registry USE reads, generated from library/
(scripts/library.mjs emit), so the site needs no files at run time.
- scripts/library.mjs check: every routine at every level: structure,
cases, prover, fuzzer, generated Mojo, each reference implementation
(CPython; cc at -O0 and -O2; core's Mojo interpreter; the pinned
compiler's stored answer) held to the pseudo's results, and the Dafny
proof with the pinned Dafny. Writes library/status.json and STATUS.md.
scripts/library_godbolt.py stores the compiler's answers (network).
- pixi tasks library-check, library-emit, library-verify-mojo,
library-harness; CI checks out the submodule and runs library-check.
- Tests: tests/library.test.mjs, corpus 53-library; docs/library.md and
a library section in trust.md.
Gate: 332 pass; library-check: no failing check (16 routines; the Mojo
references are unverified until library-verify-mojo runs on the network).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ
Pseudo could say what a procedure computes only up to INTEGER and FLOAT,
so it could not state what a Mojo or Python program with Int32, UInt8 or
Float32 values will compute, and it had no words for rounding itself.
- lang/parse: INT8..INT64, UINT8..UINT64, FLOAT32, FLOAT64 as types
(INTEGER or FLOAT with a width, so every "whole number" rule accepts
them); conversions written as Mojo writes them, INT32(X), FLOAT32(X);
1_000_000, 0xFF and 0b1010 literals; a type name as an argument
(MAX_VALUE(INT8)); the USE line (USE IEEE now, library modules next).
- check: widths and precisions never mix silently (mixed-width,
mixed-precision); a literal takes the type of the sized value beside
it, as in Mojo, and one that does not fit is an error before any run
(literal-range); typing for every new built-in.
- interp: a sized result, store, parameter, RETURN or conversion that
does not fit stops the run (kind overflow), WRAP(X, T) wraps on
purpose; FLOAT32 results are rounded to single precision; USE IEEE
lets NaN and infinities flow.
- numeric.js: exact helpers: FMA rounded once, EXACT(X) with every digit
a FLOAT holds, ULP and NEXT_AFTER for both precisions, bit operations
in two's complement.
- Built-ins: ASIN ACOS ATAN SINH COSH TANH HYPOT POW CBRT EXPM1 LOG1P E
TRUNC SIGN COPYSIGN CLAMP FMA IS_NAN IS_INF IS_FINITE INF NAN ULP
NEXT_AFTER IS_CLOSE EXACT MAX_VALUE MIN_VALUE EPSILON GCD LCM BIT_AND
BIT_OR BIT_XOR BIT_NOT SHIFT_LEFT SHIFT_RIGHT WRAP. They stay usable
as names (a variable E, a FUNCTION GCD): the program's own wins.
- vcgen: an overflow obligation on every sized result, then the range
as a fact (also for sized parameters and loop-cut variables); FLOAT32
arithmetic as its own uninterpreted functions, constants rounded
exactly, TOFLOAT32(v) otherwise; argument obligations for SQRT, the
logarithms, ASIN/ACOS, LOG1P, CLAMP and the shifts; nothing proved
under USE IEEE.
- testing: the fuzzer tries the edges of a sized type and generates
FLOAT32 values.
- pseudo-to-mojo: Int8..Float32, conversions, the new built-ins through
std.math (each name in the harvested index), a note where Mojo wraps.
- Tests: numeric.test.mjs; corpus 49-52; prover fixtures int8-overflow,
float32-store, ieee-nan; the reference documents every new form.
Gate: 324 pass (core). New Mojo uses std.math names the interpreter has
not been verified on (listed in sync-mojo's KNOWN_GAPS); they need
pixi run mojo-verify on the network before the interpreter runs them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ
What programs mean (the languages, translators, Mojo checker and
interpreter, prover, fuzzer, machine model, catalog problems and the Mojo
corpus with godbolt's stored answers) was mixed into the site's tree, so a
change to the prover and a change to a button went through the same gate
and the same docs. This moves all of it under core/, laid out as its own
repository, ready to be extracted with its history and pinned here as a
git submodule.
- core/: src/{engines,services,sync,mojo,exercises,curriculum,common},
mojo/, syntax/mojo.xml, its tests (with Workspace and workForProblem in
place of the site's Session), scripts (corpus pipeline, play, the
exercise and probe starters, engines export/diff), and its docs
(engines, trust, mojo-corpus, exercises, coverage). Its own pixi.toml,
package.json, README, AGENTS.md and CI workflow.
- core/tests/boundary.test.mjs: library modules use no DOM, import no
page widget, and import nothing outside src/.
- core/scripts/cli.mjs: pixi run cli check|run|prove|fuzz|translate.
- core/scripts/dafny.py: Dafny 4.11.0 pinned by version and SHA-256 per
platform, fetched into .tools/ (pixi run dafny-fetch, pixi run dafny).
- The site imports core only as #core/<path>.js: an import map in
index.html, "imports" in package.json for node, core's declarations for
tsc. The site's TypeScript builds core first (npm run build).
- assemble.sh, the Quarto engine copy, the render cache key, check-pages,
quarto-export and quarto-css read core/dist and core/mojo; quarto-css
copies core's syntax/mojo.xml into the extension.
- Site tests that covered engines moved to core; the rest keep Session.
- README, AGENTS.md, docs/app.md, docs/deploy.md and the CI workflow
describe the split (checkout with submodules; core's tests and corpus
status run against the pinned commit).
Gate: site 110 pass; core 307 pass; corpus status --strict passes.
Manual check: the assembled site opens Bridge (count-base-proved),
Developer and Hardware (count-base) with the editor, the stepper and the
board, and no module errors.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ
Two engine modules reached into app modules: the exercise type named the
app's AreaId, and the assembly service read the machine presets from the
Hardware area's board module. And the catalog mixed core's problems with
the app's areas and saved state.
- core/machine.ts holds the machine presets and their names (parseMachine,
machineSpec, machineLabel); hardware.ts re-exports them for the board.
- AreaId lives in core/types.ts; areas.ts re-exports it.
- core/problems.ts builds the problems from the exercises and the corpus
(PROBLEMS, problemById, workForProblem for a Workspace). core/catalog.ts
adds each problem's area and stateForProblem.
No behavior changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ
The engines are moving into their own project. Session mixed the two jobs:
deciding what a program means (the hub's cascade, parsing every language,
diagnostics, the run, the proof, the fuzzer) and app concerns (the area,
saving to localStorage, versions, compile history, the best-practices
helper, change events). Tests of meaning could only reach it through the
app's state.
src/core/workspace.ts is the first job alone, with no DOM, storage or
areas: Workspace reads a WorkState (sources, sync, lang, call, input, setup,
machine), and edit() returns whether anything changed. Session now extends
it, keeps every app concern, and overrides only what the app changes: the
Bridge's Mojo is derived (derivesMojo), and the Mojo diagnostics come from
the compiler's answer and the helper when there is one.
No behavior changes; every test passes unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ
docs/trust.md lists every badge and verdict the app shows (where code came
from, what a compiler said, what a proof, the fuzzer, a translation and the
model machine say): what earns each one, what it means, and what it does
not mean. New badges get a row there in the same commit.
pixi run package zips the files git tracks, plus new files it does not
ignore, into package/programming-plus-<commit>.zip (marked -dirty with
uncommitted changes). The zip shared before this held .DS_Store files,
__pycache__ folders and __MACOSX entries (3,661 files); the package holds
none of them (1,694 files). Python standard library only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ
Every compile went to godbolt.org from the session directly. The roadmap
moves compiles, and later Dafny and Lean checks, to a pinned service of the
app's own, and godbolt has no service guarantee, so where a compile goes is
now one replaceable piece.
- src/services/compile-backend.ts: a CompileBackend (id, host, compiler,
compile), explorerBackend() for any service that speaks Compiler
Explorer's API at a base URL, GODBOLT as the default, and
compileBackend() / setCompileBackend().
- godbolt.ts: compileMojo takes a CompileTarget (base, compiler, host) and
names that host in its messages; the request body is unchanged, so
tests/mojo.test.mjs still holds it to what the corpus pipeline sends.
- Session.compile asks the current backend and records which host answered
(CompileRecord.host; older records came from godbolt.org).
- The Compiler view names the host it is asking and the host that answered.
The "Compiled on godbolt" badge keeps its wording until a second backend
is deployed.
A test sends a compile through a fake backend and a refused local address.
Manual check list: open a Mojo exercise in the Developer area, Compile &
run, and see "Asking godbolt.org (mojo_1_1_0)…" then "Mojo 1.1.0 on
godbolt.org" in the Compiler view.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ
Whether the engines' changes went back to the course or the course took
them from here was an open question (docs/engines.md). Soundness fixes to
the prover cannot wait for a round trip through a repository with no engine
tests, so this repository now owns the engines.
- pixi run engines-export <course>/static builds and writes every engine
module into the course's static/js/: JavaScript modules as written,
TypeScript ones (logic, solver, vcgen) as built, without source maps.
- pixi run engines-diff now shows what the course has not taken yet,
comparing the file the course would receive with its copy. Both scripts
share the list of modules in scripts/engines-files.mjs.
- docs/engines.md describes the ownership, the prover's three modules and
its number model, the soundness tests and the bugs found so far, the
changes the course receives with the next export, and how to export.
- AGENTS.md gains a hard rule: never let the prover call something proved
that a run breaks, with how to add a soundness regression first.
- docs/app.md's Known limits no longer sends engine fixes upstream.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ
vcgen.ts types the symbolic executor under strict mode: the store (a term,
or the mark of a variable not yet set), the translation context, the raw
obligations, the loop hooks the invariant inference uses, and the report it
returns (Obligation, UnitReport, Report, VerifyOptions, exported for the
next stages). The parsed program is still built by parse.js and check.js,
which stay JavaScript, so its nodes are typed as PNode: a kind, a position
and untyped fields.
The behavior is unchanged: every obligation of every exercise, corpus
program, lesson solution and example, and the regression programs, comes
out byte for byte as it did before the conversion. Places where the code
read a store value it knew to be a term now say so (valOf), and an unset
value read as a term becomes a fresh unknown (asTerm), which is what the
name lookup already did.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ
The way out of a FOR EACH loop includes the case of no pass at all, but the
checker added "X is in the collection" to the loop's head state and then used
that same state as the way out. After a loop over an empty list the fact is
FALSE, so the checker held a contradiction and proved anything after the
loop, ASSERT 1 == 2 included (tests/prover/foreach-empty.pseudo). The fact is
now assumed only for the body of a pass.
The soundness test gains a canary: after every program that runs to its end
without input, ASSERT 1 == 2 is appended and must not be proved, since a
proof there means a contradiction on a path every run takes. A one-off scan
with the canary after every top-level statement of every program (476
points) found nothing else.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ
This repository now owns the engines, and the next stages extend the prover
(a claims model, translation to Dafny and Lean, precision claims). Doing that
under strict mode needs the terms it reasons about to have types.
logic.ts gives the term language a discriminated union (Num, Lit, Sym, Bv,
Op, And, Or, Not, Cmp, Ite, App, Quant, QuantC) and types every
constructor, rewrite and printer. solver.ts types the rationals, linear
constraints, Fourier-Motzkin, union-find with congruence, the case split
and quantifier instantiation, and exports ProveResult and ProveOptions.
Walks that looked inside any kind of term through x[k] now go through a
small fields() helper that keeps the same order.
The behavior is unchanged: every obligation of every exercise, corpus
program, lesson solution and example, and the regression programs, comes
out byte for byte the same as before the conversion (kind, line, column,
label, status, paths, goal and facts as shown, inferred invariants, notes).
The few places where the JavaScript would have thrown on a null it never
receives (an identity for a quantifier that has one) now skip instead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ
The engines had no tests of their own in the course repository; this
repository now owns them, so a change to solver.js or vcgen.js needs
something to answer to besides the end-to-end tests.
tests/engine-solver.test.mjs holds:
- a table of goals the solver must prove (transitivity, integer
tightening, case splits, congruence on positions, DIV and MOD by a
constant, lengths at least 0) and goals it must leave unknown (a false
step, FLOAT tightening, products of unknowns, rounded FLOAT addition,
different positions); one known gap, 2X == 1 not seen as impossible, is
recorded so an improvement shows up;
- 400 random linear integer formulas with DIV, MOD, if-then-else and
boolean structure; every one the solver proves is checked against all
729 assignments of X, Y, Z in [-4, 4];
- programs whose obligations must come out proved or not: a counting loop
with its invariant, an index one past the end, a division that needs a
REQUIRE, FILL and RANDINT arguments, and the int-model note.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ
The prover reasons about a FLOAT as a number: Y == Y holds, and X < Z or
X >= Z. The interpreter let 1.0E308 * 10.0 become infinity and X - X become
NaN and carried on, so a run broke claims the prover had proved
(tests/prover/float-overflow.pseudo proved ASSERT Y == Y with Y NaN).
The notation already stops on a FLOAT division by zero and on a whole number
past 2^53 - 1. It now also stops, with an 'overflow' error, when an
arithmetic result, a built-in's result or a literal is not a finite number
(finiteCheck in interp.js). Every run then stays inside what a proof covers.
Regression programs whose fix was to the language rather than to a rule now
say where the run must stop, with a RUN-STOPS header the soundness test
checks.
A unit that does FLOAT arithmetic now carries a 'float-model' note: FLOAT is
treated as a finite number, rounding is opaque, and the lab stops past about
1.8e308, while the generated Mojo goes on with inf or nan (stated for a
division by zero, the case probes/float-special has verified).
New probe mojo/probes/float-overflow.mojo asks the compiler what a Float64
product past the largest double gives; it is unverified and needs
pixi run mojo-verify-item 'probes/float-overflow' (network).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ
The prover could mark an obligation proved that a run of the same program
breaks. Two rules did it, both found with programs now in tests/prover/:
- Invariant inference (vcgen.js infer) stopped after four rounds and assumed
the candidates left even when the last round had still been dropping some.
A non-inductive guess became a fact, and a false ASSERT was proved
(houdini-cap.pseudo). Inference now runs up to eight rounds and keeps its
guesses only when a whole round leaves them unchanged.
- RAND(X) assumed 0 <= R < X for every X. For X <= 0 that fact is false
(RAND(0.0) is 0.0, RAND(-1.0) is in (-1, 0]), so the checker held a
contradiction and proved anything, even ASSERT 1 == 2 (rand-zero,
rand-negative). The fact now covers all three signs of X.
FILL and RANDINT assumed facts about their arguments that the interpreter
enforces by stopping the program, without asking whether the arguments are
allowed. A crashing call was reported as fully proved. They now raise an
obligation of a new kind, 'argument' ("Argument allowed"), mapped to the
runtime 'value' error, and their facts are guarded.
tests/prover-soundness.test.mjs is the gate: each regression program's
MUST-NOT-PROVE obligation stays unproved and is really broken by a run, and
across every exercise, corpus program, lesson solution and example, nothing
marked proved is broken by the fuzzer (kind, line, column) or by running the
program (kind, line).
In the app, a proved obligation that a generated input breaks at the same
place is now shown as broken rather than proved.
Proofs over INTEGER are about whole numbers of any size; a unit that does
integer arithmetic now carries an 'int-model' note saying so, with the lab's
2^53 - 1 limit and Mojo Int's wraparound past 2^63 - 1 (from the stored
godbolt answer for probes/int-overflow-int).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FgvbRdoEheAHqsRmr2WuKZ