The language corpus
The Faber language corpus is the language dictionary: one top-level
directory per keyword, operator group, or language type surface. It is the
development source for faber explain and the primary input for
multi-target compile matrices.
Location#
Corpus content is divided by shape:
| Content | Lives in |
|---|---|
| Single-file language exempla (keywords, operators, types, concepts) | radix/corpus/reference/ |
| Package-shaped corpus fixtures | radix/corpus/product/ |
| Complete simulation programs | radix/corpus/simulations/ |
| The Norma suite, one proof package per case | radix/corpus/norma/ |
| Application demos | examples/ (sibling dirs) |
The old examples/corpus/ path is a redirect stub. The generated corpus
pages on this site are built from radix/corpus/.
Simulations#
Complete programs live in corpus/simulations/, each with a README, a small
smoke and its provenance. Before 1.19 the directory was corpus-simulations/.
1.18 shipped four of them, and 1.19 adds four more:
| Program | What it does |
|---|---|
fusion-containment | Particle orbits from one Latin source, on the reference route, multicore CPU and Metal |
drone-mission | Branchy aircraft mission evaluation, where each GPU lane flies a whole mission |
swarm-decimal | A deterministic many-actor simulation in fixed-point decimals |
h264-encoder | A constrained-baseline H.264 encoder on native multicore CPU workers; no GPU acceleration is claimed |
a2-flood | A regional flood solver at corpus size: run-time extent stencil kernels and a device-resident clock, pinned against an independent binary64 reference |
chemical-diffusion | Coupled spatial chemistry, checked against an independent reference, with CPU worker and partition controls |
rocket-nozzle-flow | Quasi-one-dimensional compressible nozzle flow: geometry, a host solver and f32 device primitives |
structural-fea | Static three-dimensional linear elasticity on eight-node hexahedral elements, solved matrix-free with preconditioned conjugate gradients (paused spike) |
Device execution and the native Rust emitter are in the released faber binary,
not the open-source build. Each README gives its exact route requirements.
Size#
The corpus holds several hundred .fab exemplar files, one directory per
keyword, operator group, or concept. Each term has one owning file, and the
faber build embeds the corpus, so there is no separate index to maintain.
faber explain --list prints the current term list.
Layout#
Since 1.19 one corpus/ tree holds every suite, and each subtree says what its
cases are. A language exemplum lives at reference/<kind>/<term>/<file>.fab.
corpus/
reference/ # language exempla, one directory per term
keyword/ # keywords (for/, fn/, …)
type/ # types (list/, tensor/, …)
operator-group/ # glyph and operator groups
reject/ # programs the compiler must refuse
...
product/ # package-shaped fixtures, one faber.toml per case
simulations/ # complete programs with smoke packages
norma/ # the Norma suite
format/ # faber format inputs and their expected output
...A case's expected output lives in the file itself, in +++ text stdout (or
stderr, json, jsonl, base64) sections; a reject case names its
diagnostics in its header. There are no .expected sidecars.
File format#
Each .fab file begins with TOML frontmatter describing the term:
+++
term = "functio"
kind = "keyword"
category = "function"
canonical = false
summary = "Declares a named function or method."
syntax = "functio <name>(<params>) [modifiers] [→ <type>] [⇥ <error-type>] <block>"
related = ["→", "⇥", "redde", "sponte"]
locale = "en"
+++
fn saluta() {
print "Salve, Mundus!"
}Usage#
faber explain fn # show keyword reference
faber explain ≡ # show glyph reference
faber explain --search query # search across corpus
faber explain --list # list all termsCategories#
Terms are organised by category, for example function, control-flow,
type, collection, conversion and testing. List one category with
faber explain --category <CATEGORY>.