Tree:
b0a4c6ef29
master
v0.1a
${ noResults }
1541 Commits (b0a4c6ef29ead2fa9c76553f9b44c0efafa3cfda)
| Author | SHA1 | Message | Date |
|---|---|---|---|
 Julian Noble
|
b0a4c6ef29 |
buildsuites publish riders: librev default -> 2 (published-state record) + record path fix
librev defaults bumped 1->2 in both recipes with repaired help text (was "nothing published yet" while r1 had been published since 2026-07-30 - the manual-drift failure mode recorded in G-171's publish follow-on); the help now states the convention: the default IS the committed record of published state, bumped in the same work unit as each publish. Family-record comment fix (tcl90): the repairs-manifest pointer is artifact-relative - lib/tcllib<ver>/punkbin-upstream-repairs.toml in the mounted image (the attached zip roots at the staged base/, so the previous base/lib/... path named a location that does not exist to a reader inside the runtime). Probe-caught before publish; family kits rebuild before the replace. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
05638877d3 |
G-171 note: publish command owns artifact-revision derivation
Extends the dirty-bake provenance follow-on's publish-command endpoint with the -r<N> derivation duty (user question 2026-08-08: must -Dlibrev be manual?): the build recipe cannot safely self-derive (deliberate-publish discipline, no configure-time mutable-state inputs, rev-inside-artifact means comparison must build AT the published rev and sha1-compare); publish verdicts current / bump to max+1 / alpha in-situ replace, asserted on a checked claim. Records the recipe-default-as-published-state convention (familyrev precedent) and librev's observed drift as the manual failure mode - default+comment repair due at the r2 publish. goals_lint clean. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
988c7868fd |
G-171 amendment follow-up: GOALS.md Scope mirror sync + G-176 archive marker
goals_lint caught both: the index Scope mirror must match the amended detail Scope (index carries a mirror the amendment commit missed), and the new alternatives bullet's G-176 reference needed its achieved/archive marker. Lint now fully clean, no findings or warnings. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
e42587cd79 |
G-171 amendment: runtime-content masks (declared kit-level battery demotion)
Extends the proposed goal with Approach 11 + matching Goal/Scope/Acceptance clauses and a Context assembly finding (2026-08-08): zip-kit bake EXTRACTS the runtime's attached vfs into the per-kit merge tree, so runtime batteries are structurally maskable - but supersedes/replace run in the VFSPAYLOAD phase before extraction and never reach them, so a bi-runtime kit with a lower-versioned payload battery ships both copies and highest-wins picks the runtime's. The amendment: mapvfs.toml kit-entry runtime_excludes applied at merge time after extraction - the payload version wins by absence of competition. Guardrails (boot-critical refusal, smokerequire proof), provenance (mask = declared config, punkkit-stamp runtime_masks key, no-volatile-fields safe; runtime's own record never edited), zip-lane first. Alternatives recorded: require-time pinning via the managed ::package override, runtime-variant proliferation. goals_lint clean. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
8db289fd57 |
punk9wintk905.vfs: carry tdom 0.9.6 (punk kits need rest-class tcllib deps)
The 905-era vfs was created without tdom while its predecessors (punk9wintk903/90b2) and the bi vfs (punk9win_for_tkruntime) all carry it - so punk905/punk9_beta kits could not load punk::netbox-class libs and 'package require rest' failed on the missing dep before ever reaching the tcllib pkgIndex skew. User direction: bi kits AND plain punk executables both carry tdom (bi may stay conservative, punk kits may advance later). Copied verbatim from punk9win_for_tkruntime.vfs (slim variant, no HTML docs, dual-generation pkgIndex; its tcl9tdom096.dll is live-proven in the deployed punk9bi_beta today). Reaches kits at the next bake; CHANGELOG entry rides the pre-bake version bump. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
0c090fa50b |
buildsuites: gated-tree repairs manifest + provenance-record pointers
Closes the tcllib_checkout honesty gap: a gated tree's shipped content is its checkout PLUS the pkgindex_gate repairs, but no record travelled with the divergence. pkgindex_gate.tcl (both suites, byte-identical) now writes punkbin-upstream-repairs.toml at the gated tree root after both gate checks pass - the applied-repair enumeration ($repaired, sorted), the upstream ticket URL (promoted to data as UPSTREAM_REF), deterministic content only. Both artifact consumers carry the tree verbatim, so the enumeration ships inside every kit and lib zip adjacent to the divergence it explains; no manifest is written on gate failure (fixture-verified both directions). Records point rather than duplicate (single owner = the gate's allowlist; the manifest is written from the post-verification applied set, so it can never drift): library_artifacts.tcl [provenance] gains upstream_repairs = "punkbin-upstream-repairs.toml" when the package folder carries the manifest; build905.zig family records gain tcllib_gated = true with the resolved base/lib/tcllib<ver>/ manifest path (build86 emits no family records - covered via the shared tools). Verified: gate fixture with all five recorded defects + clean module (repairs + manifest exact); info-complete on both Tcl tools; zig ast-check + extracted-template compile render proving 18-arg format arity. Cross-suite tool copies byte-identical. No project bump (buildsuite-only). Effect materializes at the r2 rebuild chain - r2 records honest from birth. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
5c47f3b97f |
bootsupport promotion: G-176 coherent set + make.tcl libunknown guard follow-on
Release-step-(9) commit of the src/bootsupport refresh (user-run 'make.tcl
bootsupport'): commandstack 0.7.1->0.8.0, punk 0.2.8->0.3.0, ansi
0.2.0->0.2.1, args 0.18.0->0.25.2, moduledoc tclcore 0.4.0->0.4.2,
libunknown 0.2.4->0.3.1, ns 0.9.3->0.11.0 - all seven verified
byte-identical to the minted modules/ tree. make.tcl regains committed
provenance for the tooling it boots from; the staleness gate is satisfied
by committed content again.
With it, at the G-176 follow-on's own due point ("before or with the
bootsupport/kit promotion"): the libunknown skip-guard swapped from 'info
commands ::punk::libunknown::package' (a proc the 0.3.x
commandstack-managed install no longer creates) to a 'package provide
punk::libunknown' probe - satisfied by every vintage's boot-site source
(verified in 0.2.4 and 0.3.1; tclsh-driven boot smoke green). Disposition
flipped to landed in the archived goal; goals_lint clean. No project bump:
promoted content already versioned (0.63.0-0.65.0 arc), guard swap is
interface/output-unchanged make tooling internals.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
|
2 days ago |
|
Julian Noble
|
2ec08535a8 |
G-171 note: dirty-bake provenance indication - follow-on candidate
Records the 2026-08-08 assessment as a live-tier Notes item (becomes a Follow-on line at the flip, per goals/AGENTS.md - live goals carry no Follow-ons section): near-term tree_state field in the advisory kit sidecar surfaced by bakelist/BAKE SUMMARY/buildinfo; endpoint a publication-side content-addressed manifest (kit sha1 -> producing commit, absence IS the indication) sequenced after G-171's records-free-clone reproducibility lands; stamp/payload and filename homes rejected (determinism principle, name-keyed machinery). goals_lint clean. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
cc92078c44 |
make.tcl launch-site steering (red) + countdown to 4 (project 0.65.2)
The staleness fix block now ends with a red-highlighted steer back to the project root for the remaining steps (vfscommonupdate/bake are canonically root-launched on the refreshed bootsupport snapshot, which also re-arms the staleness check), and the src/-launch banner carries the same red canonical-mode steer plus a repair/testing-is-fine line - so the mode is self-describing however the user arrived in src/. Early-site colour via the ::punkboot::sgr gate (NO_COLOR/force/tty precedence, piped output ESC-free; verified forced-colour and NO_COLOR renders). Dirty-src proceed countdown extended to 4 seconds (user change). Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
c8b51bc49a |
G-171 note: staleness-message emitter seam for Approach 10
Records that the stale_bootsupport gate cited as the layout-store report's
shape template now emits via the shared punkboot::lib::bootsupport_stale_warning
helper (
|
2 days ago |
|
Julian Noble
|
8aa824ee88 |
layout store sync: propagate make.tcl staleness-gate message (0.65.1) to project layouts
sync_layouts propagation of the reworked bootsupport-staleness warning
(
|
2 days ago |
|
Julian Noble
|
b729ce07d7 |
make.tcl staleness-gate message: explain launch-site module precedence (project 0.65.1)
The recommended fix (rerun modules && bootsupport from src/) read as circular when the gated command was itself modules or bootsupport. The warning now names the launch directory, contrasts src/ vs elsewhere module precedence (the stale bootsupport copies are the code executing the current run), special-cases the repair commands with a NOTE, and prints the resolved src path as a copy-pasteable fix line. Shared emitter (punkboot::lib::bootsupport_stale_warning) for the pre-dispatch gate and the check report tail. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
71e3c96387 |
G-177 + G-178 drafted (proposed): punk::args scalability - cost decoupling and multi-provider safety
Drafted from a measured scalability review asking what breaks if every command - including third-party libraries - carries punk::args documentation. Baseline probes (punk91, 58 registered namespaces / 693 definitions) found three cost centres and one governance gap; the two goals split them by kind. G-177 (engine, measurable): the resolved-spec caches key on the whole raw definition text (~args:2152, the author's own todo), so ~4 full-text hashes ride every parse call - an isolation probe shows the ENTIRE per-call growth with documentation size sits in those keys (220 B definition 28.8us vs 9776 B definition of identical argument shape 56.8us; resolve 1.5 -> 29.6us). Registry membership is lists with a nested-loop ldiff behind a length-equality early-out that never fires in a stock session (registered=58 loaded=60 - argdoc namespaces enter loaded_packages un-registered and registration is not deduped), so update_definitions costs 9.3us at R=100 and 2515us at R=2000 while being called per ensemble subcommand. The scan phase is eager and global (15us per definition). Acceptance is six machine-independent ratio pins (1.15x doc-size independence, 2x registry-size independence, early-out liveness plus its false-positive, O(1) scan, 40% raw-text residency drop, byte-identical corpus renders) rather than absolute times, so the suite pins the complexity class and survives hardware changes. G-178 (governance, policy): flat global id namespace with silent collisions (the "last will silently win" todo at ~args:5137), non-idempotent registration into a foreign namespace variable, no definition format-level declaration, and third-party help text that is tstr-evaluated with -allowcommands at resolve time - documentation that runs code. Split from G-177 per the goals/AGENTS.md rule that a goal must be summarizable by title and Scope alone. Overlap survey (goals_xref paths + score): Related lines to G-175 (disjoint - it owns how many times a render resolves, G-177 owns what one resolve costs), G-044/G-050 (latency consumers), G-075 (shared id-lookup path), G-167 + G-056 (both deliberately change render output, which G-177 pins byte-identical - whichever lands second re-baselines the characterization), G-046 (archived predecessor whose display-field deferral is why the residual doc-size tax is purely the cache keys). No archived follow-on disposition required flipping. goals_lint clean (81 active-index goals / 97 archived). Claude-Session: https://claude.ai/code/session_01Y5vEJUZkD4p8jmEFEBVhLi Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
7787f0e06e |
G-176 archive: Follow-ons section brought to contract (single-line items, last section) - the two open items now enumerate in goals_followons
The achieved flip's step (c) was written non-conformant: the two deferred
items were multi-line prose bullets with trailing '=> open' markers, and
the '## Follow-ons' section sat before '## Notes'. The Follow-ons
contract (goals/AGENTS.md, goals-system v4) requires one item per line at
column 0 ('Follow-on: <item text> => open') in a '## Follow-ons' section
that is the LAST section of the archived detail file -
goals_followons.tcl enumerates only grammar-conformant lines, so the
items were invisible to the report (and goals_lint only validates lines
that BEGIN with 'Follow-on:', so nothing flagged them).
Both items (boot-site commandstack pre-sourcing; make.tcl skip-guard
modernization) now appear in the goals_followons OPEN section. Item
content unchanged in substance - compressed to single lines, with the
now-moot scope-amendment caveats dropped (an archived goal's follow-ons
are enacted via new goals or direct landing per the disposition
vocabulary).
Flip-guidance audit against goals/AGENTS.md prompted by the miss: Status
grammar, Progress evidence, pure-rename archive moves, index/archive
records and the live-tier reference sweep (no G-176 references exist in
goals/*.md or GOALS.md) all check out; the Follow-ons grammar was the
one non-conformance.
goals_lint clean (79 active-index / 97 archived); goals_followons
enumerates both G-176 items as open.
Claude-Session: https://claude.ai/code/session_01QgaxV27VZkmEec7oNbEVFc
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
|
2 days ago |
|
Julian Noble
|
8fa716b62b |
G-176 achieved: override hints + 'package epoch' documented via stack-tied doc block (punk::ns 0.11.0, punk::libunknown 0.3.1; project 0.65.0)
Increment 4 (the final increment) + the achieved flip.
punk::ns 0.11.0: cmdinfo gains an 'overridden' result key - a second axis
beside cmdtype/unavailable (the G-166 pattern) reporting the LIVE
commandstack override state of the base command: one {renamer <string>
docids <list>} dict per stack record (bottom-up), docids being the
punk::args ids of doc blocks the record attached via 'rename_command
-punkargs' (derived with punk::args::rawdef_id). Guarded live query - no
hard commandstack dependency, empty when commandstack is not loaded,
never cached so it is correct as stacks change. cmdhelp surfaces it: the
table/string renders lead with a plain "note: '<cmd>' is currently
overridden/extended via commandstack by: <renamers> (override-attached
docs: '<ids>')" line, -return text leads with an OVERRIDDEN: line (below
UNAVAILABLE when both apply), -return dict carries the key on all return
shapes incl the undocumented fallback.
punk::libunknown 0.3.1 (renamed from 0.3.0): the 'package epoch'
subcommand is now DOCUMENTED - a substitution-free doc block with the
space-form id {::package epoch} rides the override's commandstack record
via -punkargs: mirrored for lazy pickup when punk::args loads after init
(normal early boot), defined immediately when punk::args is present, and
removed with the record by any removal path. 'i package epoch' renders
real usage (all six grammar forms) - the doc gap that motivated the whole
arc, retired. init's require tightened to 'commandstack 0.8-': a
pre-mechanism stable copy fails pointedly at init instead of mid-call
with a usage error (found via shadowing.test, whose require-based
children needed 'package prefer latest' - the default stable preference
picks a stale stable bootsupport commandstack over the alpha-versioned
dev copy).
End-to-end verified: 'i package epoch' renders the epoch doc; 'i package'
leads with the note naming punk::libunknown and '::package epoch';
-return text/-return dict carry OVERRIDDEN:/the structured key;
restore_original ::package removes doc and note.
Tests: cmdhelp.test gains 2 hint pins (documented-command note incl docs
clause + dict axis; doc-less-proc autodef render leading with the note,
OVERRIDDEN: text lead, note clearing on unwind) and its two key-list pins
(cmdhelp_cmdinfo_result_shape, cmdhelp_return_dict_valid) flip
deliberately with the new key - exactly as they did for G-166's
'unavailable'. packageoverride.test gains the flagship pending-path pin
(epoch doc mirrored pre-punk::args in path-cleared children, lazily
defined on punk::args arrival, removed on unwind) and its pko_probe
injects ::PKO_SRCMODULES. shadowing.test children set 'package prefer
latest'. src/tests/modules/AGENTS.md punk/ns + punk/libunknown bullets
updated.
G-176 flipped ACHIEVED 2026-08-08: every acceptance clause met across the
four increments (characterise -> migrate -> -punkargs mechanism -> hint +
flagship docs). Detail file archived to
goals/archive/G-176-commandstack-doc-integration.md with its two
boot-site Follow-ons kept open (commandstack pre-sourcing at the
libunknown-style boot sites; make.tcl skip-guard modernization - due by
the 0.3.x/0.8.0 bootsupport promotion). GOALS.md entry removed;
GOALS-archive.md record added; the one tree reference
(packageoverride.test header) swept to the archive path.
Project 0.64.0 -> 0.65.0 + CHANGELOG.
Suites: libunknown+ns+commandstack+loadedlib 216/216; full modules tree
1354 total / 1343 pass / 11 constraint-skipped / 0 fail (zig-built
tclsh90s 9.0.5); testbody_lint 1707 clean; goals_lint clean (79
active-index / 97 archived); make.tcl projectversion consistency OK.
Claude-Session: https://claude.ai/code/session_01QgaxV27VZkmEec7oNbEVFc
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
|
2 days ago |
|
Julian Noble
|
ad0489b881 |
commandstack 0.8.0: rename_command -punkargs stack-tied doc blocks (G-176 increment 3; project 0.64.0)
rename_command accepts a new leading option '-punkargs
<list-of-definitionlists>' attaching punk::args doc blocks to the
rename's stack record - e.g documenting a subcommand the override adds,
with space-form ids such as {::package epoch}. Each element is one
definitionlist exactly as given to punk::args::define (the same shape as
a PUNKARGS variable element).
Lifecycle: docs attach only when the rename lands (aborted renames
return early) and are LIVE while the record is on the stack. The detach
hook sits at remove_rename's doomed-record site - pop_rename,
remove_renamer and restore_original all funnel through it - so every
removal path removes the docs with the record.
Works whether or not punk::args is loaded at rename time: deflists are
mirrored in a new ::commandstack::stackdocs namespace registered inert
into ::punk::args::register::NAMESPACES at module load (punk::args need
not be present - the early-boot renames precede it), so a later-loading
punk::args picks them up lazily; when punk::args is already present they
are ALSO defined immediately (a namespace already consumed into
loaded_packages never lazy-loads later appends; redefinition is
idempotent). Detach removes one mirror occurrence per deflist and
undefines via punk::args::undefine_deflist (deflist-keyed - commandstack
never parses ids; never-defined deflists skip silently). Supported
shape: one live record per doc id (same-id declarations in two records
shadow; removing either removes the doc).
The record carries the deflists under a trailing 'punkargs' key - the
append-only key-order contract is preserved (token idx 0-1, renamer idx
2-3). The leading-option parse is generalised: -renamer/-punkargs in
either order, both leading-only, non-dash word ends option scanning so
misplaced options raise the pointed error. stackdocs state follows the
0.7.1 reload contract (info-exists guarded; registration idempotent
across re-source). commandstack::help and the rename_command argdoc
document the mechanism; the In-tree users line gains punk::libunknown
(the increment-2 migration).
commandstack.test: 5 new pins (docs live with the record incl key-order
assertions, all four removal paths detach, the pre-punk::args pending
path in a bare child that loads punk::args afterwards and lazily
consumes the registered mirror, reload contract on mirror + single
registration, option positions in either order + pointed misplacement) +
2 existing G-160 message pins updated to the generalised
usage/misplacement wording (behaviour contracts unchanged). Suite 46/46.
src/tests/modules/AGENTS.md commandstack bullet updated; goal Progress
entry added. Project 0.63.0 -> 0.64.0 + CHANGELOG.
Suites: commandstack 46/46; full modules tree 1351 total / 1340 pass /
11 constraint-skipped / 0 fail (zig-built tclsh90s 9.0.5);
testbody_lint 1704 clean; goals_lint clean (80 active-index / 96
archived); make.tcl projectversion consistency OK.
Claude-Session: https://claude.ai/code/session_01QgaxV27VZkmEec7oNbEVFc
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
|
2 days ago |
|
Julian Noble
|
72725deecb |
tcllib pkgIndex gate in both buildsuites: upstream ifneeded/provide skew (rest 1.7 vs 1.8) repaired before it reaches kits
'package require rest' fails in every punk kit built from stock tcllib: modules/rest/pkgIndex.tcl indexes 1.7 while rest.tcl provides 1.8, so Tcl runs the 1.7 ifneeded script, the file provides 1.8, and the require dies with "attempt to provide package rest 1.7 failed: package rest 1.8 provided instead". No version request can satisfy it. Upstream regressed this in tcllib checkin [0b3dd048f8] (2025-07-16), which bumped the provide and left pkgIndex.tcl (and rest.man) at 1.7; still unfixed on trunk/master, so the pin cannot be moved past it. Reported as https://core.tcl-lang.org/tcllib/tktview/523f7e055647611328efa2539717023d18b99ccd Four siblings share the class in the shipping tree: doctools::toc (2 vs 2.1), math::filters (0.3 vs 0.4), nettool (index sources a nonexistent nexttool.tcl), and tcl::chan::std (1.0.2 vs 1.2 - latent only, punkshell's own lib/virtchannel_base copy shadows it). tools/pkgindex_gate.tcl (byte-identical copies in both suites) sits between the tcllib installer output and the install tree. The index follows the file: the provide is what executes, so rewriting a file's provide down to its indexed version would misrepresent the code. It is a gate, not a silent fixer - an unrecorded defect fails the build (new upstream breakage and a genuine version bump are indistinguishable here, and the difference is a human decision), and an ALLOWLIST record matching nothing also fails, meaning upstream fixed it and the record must go. build905.zig had TWO consumers of the raw installer output - the install prefix and the kit-family base/lib/tcllib<ver> copy. Both now take the gated tree; repointing only the install would have left every kit broken while the fix looked applied. Verified: 440 indexed loaders checked, 5 repaired, exactly 5 single-line diffs and nothing else touched; an independent re-audit of the repaired tree reports 0 mismatches; rest 1.8, doctools::toc 2.1, math::filters 0.4 and nettool 0.5.4 all require cleanly against it; both gate failure arms exercised; the tool runs identically under Tcl 8.6.18 and 9.x; both zig recipes compile and construct their graphs under the pinned zig 0.16.0. No full suite rebuild was run - the fix reaches kits only after a rebuild, lib-tier r2 emission, publish, libfetch and bake. No project-version bump: buildsuites are a non-shipped surface. Claude-Session: https://claude.ai/code/session_01DeQ2aPnbsinpFABUTYK34k Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
a7bc62c3fb |
punk::libunknown 0.3.0: ::package override installs via commandstack (G-176 increment 2; project 0.63.0)
The ::package override (the 'epoch' subcommand and controlled 'forget') now installs via commandstack::rename_command -renamer punk::libunknown instead of the former 'rename ::package :📦:' + namespace-import mechanism. Both live ::package overrides (this and punk::packagepreference's) now share one managed stack whatever the install order - stack removals and third-party renames re-link correctly instead of operating on the wrong command (previously commandstack's ::package stack believed the packagepreference wrapper was live while libunknown's unmanaged import actually was; restore_original would have stranded :📦:). Migration deltas (exactly the set enumerated in the goal Context): - the override body runs at :: - the four bare epoch_incr_pkg/epoch_incr_tm calls are fully qualified (the old namespace-import execution context is gone) - the dead 'variable has_package_files' declaration dropped (its only use was already commented out; the module-load probe stays) - the two :📦: delegation sites (forget arm, default arm) now delegate via COMMANDSTACKNEXT; :📦: no longer exists (nothing in the tree called it) - init does 'package require commandstack' (a no-op when already provided - e.g whenever punk::packagepreference::install ran first, as in punkshell boot; resolvable via tm paths in every other real init context - bootsupport and the common vfs both carry commandstack-0.7.1) - re-entry guard is now a stack query for the punk::libunknown record (same 'already done' notice; the old ::punk::libunknown::package proc-existence check tested a proc that no longer exists) - init argdoc reworded; file renamed libunknown-0.2.4.tm -> libunknown-0.3.0.tm (git mv + fossil mv) with Meta/manpage/provide-block version updates and a changelog entry per the in-file bump procedure. Behaviour of every subcommand arm is unchanged: all increment-1 BEHAVIOUR pins in packageoverride.test passed byte-identically; only its four MARKED-TO-FLIP install-shape pins flipped as fenced (origin ::package, no :📦: residue, managed stack records - 1 solo, 2 in both packagepreference interop orders). discovery.test and packageoverride.test child interps now pre-source the source-tree commandstack by path (their tm paths are cleared); libsearch/shadowing children resolve commandstack via inherited parent paths and needed no changes. Two out-of-scope items recorded as open Follow-ons in the goal (Scope amendment needed if pursued): boot-site commandstack pre-sourcing (punkboot core.tcl / make.tcl / punk::repl / punk::lib child-seeding - to preserve the minimal-untracked epoch intent exactly) and make.tcl's 'info commands ::punk::libunknown::package' skip-guard modernization (old kits still satisfy it; switch to a package-provide check before or with the 0.3.0 bootsupport/kit promotion). Goal Scope mirrors updated to the new filename (rename anticipated in the Scope text). Project 0.62.2 -> 0.63.0 + CHANGELOG. Suites: libunknown tree + loadedlib 42/42; full modules tree 1346 total / 1335 pass / 11 constraint-skipped / 0 fail (zig-built tclsh90s 9.0.5); testbody_lint 1699 clean; goals_lint clean (80 active-index / 96 archived); make.tcl projectversion consistency OK (and that run itself exercised the bootsupport-0.2.4 transition path live). Claude-Session: https://claude.ai/code/session_01QgaxV27VZkmEec7oNbEVFc Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
642fb265ea |
G-176 increment 1 + active flip: ::package override characterisation - the pre-migration safety net (packageoverride.test, 11 pins)
G-176 marked active (user-directed). New src/tests/modules/punk/libunknown/testsuites/discovery/packageoverride.test pins the CURRENT punk::libunknown ::package override ahead of the increment-2 migration onto commandstack::rename_command. BEHAVIOUR pins (must stay green byte-identically through the migration): 'package epoch tm|pkg' single-arg forms, indexed '<which> <index> ?key?' queries, the four error arms, the filtered forget arm (real-ifneeded forgotten, no-ifneeded filtered out, tcl/Tcl/tcl::oo/tk denied, mixed lists partially applied), abbreviation acceptance (ep/epo/epoc + fo routing, with the parked builtin's rejection of 'ep' pinned as the override's deliberate delta), delegation pass-through, and the re-init no-op contract (stderr notice captured via a child ::puts shim). MARKED-TO-FLIP pins (comment-fenced to the increment-2 commit and no other): the install shape (::package origin = the ::punk::libunknown::package namespace import; :📦: parked and functional) and the two-override packagepreference interop children in BOTH real-world install orders (punkshell's pkgpref-first: stack count 1 + import origin; make.tcl's libunknown-first: stack count 1 + ::package proc origin). Children are fresh interps sourcing the source-tree libunknown/commandstack/packagepreference by path via a shared highest-version locator (discovery.test pattern). All 11 pins green on the first run against the live implementation. Tests AGENTS.md index updated; goal file gains its Progress entry. testbody_lint 1699 clean; goals_lint clean (80 active-index / 96 archived); modules tree 1346 total / 1335 pass / 11 constraint-skipped / 0 fail (zig-built tclsh90s 9.0.5). Claude-Session: https://claude.ai/code/session_01QgaxV27VZkmEec7oNbEVFc Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
b10abcbae0 |
G-176 drafted (proposed): commandstack doc integration - stack-tied overrider argdocs, override hints, libunknown package-override migration
Four-increment arc from the 2026-08-08 feasibility investigation:
(1) characterise the untested arms of the current punk::libunknown
::package override (epoch single-arg/indexed forms + error arms, the
entire forget arm, abbreviation acceptance, delegation, re-init guard,
packagepreference interop in both install orders; install-shape pins
marked to flip), (2) migrate that override onto
commandstack::rename_command (one managed stack for both live ::package
overrides - cures the restore_original/wrong-command latent hazard of
the current unmanaged mix), (3) rename_command -punkargs stack-tied doc
blocks (attach at the single rename site, detach at the remove_rename
choke point every removal path funnels through; stackdocs registered
namespace covers renames performed before punk::args loads), (4)
cmdhelp/cmdinfo override-in-place hint (G-166-style second axis, live
get_stack query, no hard dependency) + the flagship {::package epoch}
doc attachment and the commandstack::help how-to.
Investigation evidence embedded in the goal Context: probe-verified
space-form doc define/resolve/undefine/fallback cycle ('i package
epoch' resolves despite the parent's restricted choices - G-051
space-form jump), record-key append contract, undefine_deflist
primitive, no external :📦: callers, the dead
has_package_files declaration, both real-world install orders.
goals_lint clean (80 active-index goals / 96 archived).
Claude-Session: https://claude.ai/code/session_01QgaxV27VZkmEec7oNbEVFc
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
|
2 days ago |
|
Julian Noble
|
b82827d837 |
punk::args 0.25.2 + tclcore moduledoc 0.4.2: qualify ensemble names into ensemble_subcommands_definition - 'i ::tcl::prefix' unqualified-ns warnings gone; G-175 drafted (project 0.62.2)
'i ::tcl::prefix' emitted 12x "warning: update_definitions received unqualified ns: tcl" on stderr (6x for subcommand renders). Root cause: the tclcore moduledoc's @dynamic ::tcl::prefix definition passed the ensemble name to punk::args::ensemble_subcommands_definition UNQUALIFIED ('tcl::prefix'). The generator's space-form id_checks derive namespaces via 'namespace qualifiers' - for a relative multi-component name that yields a non-empty UNQUALIFIED namespace ('tcl') which the empty-qualifier guard ('' -> ::) never catches - and update_definitions warned once per subcommand per resolve. @dynamic definitions re-expand on every resolve and the cmdhelp pipeline resolves the id four times (parse_status spec fetch + its internal parse, cmdhelp's get_spec, arg_error/synopsis), so 4 resolves x 3 subcommands = the reported 12. tcl::prefix was the ONLY affected ensemble: all tclcore ensemble docs pass unqualified names, but the single-component ones (info/dict/file/ namespace/array/encoding/zipfs) derive an empty qualifier which was already mapped to ::. Both fixes: - tclcore call site now passes ::tcl::prefix. - ensemble_subcommands_definition normalizes its ensemble argument to fully-qualified in the CALLER's context (uplevel namespace which, :: prepend fallback) before deriving anything - covering any future relative multi-component caller. Verified: zero warnings at both depths; rendered help byte-identical (6301/5585 chars); generator snippet identical for qualified vs unqualified input. New pin ensembledef.test ensembledef_unqualified_ensemble_name_no_warnings (fixture two-component ensemble called relatively under stderr capture; capture_stderr helper added to the file). src/tests/modules/AGENTS.md index updated (ensembledef.test now listed). G-175 drafted (proposed): goals/G-175-punkargs-dynamic-resolve-multiplicity.md - the warning was the only signal of the 4x-per-render @dynamic resolve multiplicity; the goal records the verdict question (render-scoped single resolution vs pinned accept), the attributed resolve sites, the measured cost floor (redundant builder calls ~3ms on 'i ::tcl::prefix', ~22ms on 'i ::dict' steady-state; the full argdata re-processing share is unmeasured - first investigation step), and candidate mechanisms smallest-first. GOALS.md indexed. punk::args 0.25.1 -> 0.25.2, punk::args::moduledoc::tclcore 0.4.1 -> 0.4.2, project 0.62.1 -> 0.62.2 + CHANGELOG (user-visible repl stderr noise fix). Suites: testbody_lint 1688 clean; goals_lint clean (79 active / 96 archived); modules tree 1335 total / 1324 pass / 11 constraint-skipped / 0 fail (zig-built tclsh90s 9.0.5); make.tcl projectversion consistency + staleness OK. Claude-Session: https://claude.ai/code/session_01QgaxV27VZkmEec7oNbEVFc Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
661418b732 |
punk::ansi 0.2.1 + punk::args 0.25.1: add missing -- to 'regexp -about' sites - dash-led grepstr patterns and -type regexp values work again (project 0.62.1)
grepstr regression report (user, 2026-08-08): 'grepstr -h underline
-group test-group' errored 'bad option "-group": must be -all, -about,
-indices, ...' - regexp's own option list - with or without the --
end-of-options marker, where an old punk901t.exe kit worked.
Diagnosis: the punk::args parse layer was never at fault. A live probe
seats the dash-led word as the pattern value in both invocation forms
(without a marker via the starved-values/valmin reservation rule). The
failure was the body's capture-group probe 'regexp -about $pattern' in
punk::ansi::grepstr - no -- guard, so any dash-led PATTERN VALUE was
read as a regexp option. Present since the proc's introduction
(
|
2 days ago |
|
Julian Noble
|
df74c3b2db |
punk::args 0.25.0: bounded -multiple ranges participate in leader/value allocation (directed work post-G-053; project 0.62.0)
Bounded -multiple occurrence ranges now PARTICIPATE in positional allocation
for leaders and values instead of being post-loop enforcement only.
Previously greedy collection overran a bounded max unless a type screen
happened to stop it - untyped {3 3} followed by a -multiple tail collected 4
words and occurrencecount then reported the overrun the allocation itself
created (and typed {3 3} with four consecutive ints failed identically) -
and a later REQUIRED ranged argument was starved to a single reserved
clause.
Three cooperating sites:
(a) get_dict_can_assign_value caps a satisfied -multiple argument at its
bounded max: the word is yielded (no-consume) so the caller's retreat
advances to the next argument - one proc serving both the leaders and
values loops. The yield carries a G-082 rejection record of new kind
'occurrence', and both loops' overflow selection sites render it as the
pointed occurrence-limit report with the G-053 occurrencecount errorcode
(count = the attempted occurrence) instead of the generic toomanyarguments,
so genuine over-supply keeps its pointed class - now describing an overflow
the allocator refused rather than one it created.
(b) Reservations are min-occurrence aware: the allocator's tail_needs and
the derived valmin floor (leader/value split + option-scan reservation)
reserve min-occurrences x min-clause-length for a required ranged argument
- a following required -multiple {3 3} reserves 3 clauses, {2 2} pair
clauses reserve 4 words - so earlier greed cannot starve it. An explicit
'@values -min' still overrides the derived floor; legacy boolean -multiple
(no MULTIPLE_RANGES entry) reserves exactly as before.
(c) The greedy leader scan caps a bounded -multiple last leader at max
occurrences (via the leader_posn_names_assigned counter), so remaining
words flow to the opts/values sections. (The split is a separate pre-loop
scan - capping only the loops left over-provided leader words raising
toomanyarguments.)
First occurrences are never capped (resolve guarantees range max >= 1);
{0 1} at-most-once scalars are unaffected (no collection); bounded-variable
ranges take greedy-up-to-max deterministically; legacy boolean greed is
byte-unchanged (pinned). define -help -multiple documents the allocation
semantics.
multipleranges.test gains 7 allocation pins: untyped/typed {3 3} cap incl
the cap-beats-type-screen four-ints case, greedy-up-to-max {1 2},
required-range and pair-clause reservations, the leaders-side scan cap +
split floor, the pointed over-supply report, and a legacy-greed-unchanged
guard. src/tests/modules/AGENTS.md index updated.
Project 0.61.0 -> 0.62.0 + CHANGELOG (allocation behaviour is user-visible
shell parsing). punk::args 0.24.0 -> 0.25.0.
Suites: punk/args 408/0; modules tree 1320 pass / 11 constraint-skipped /
0 fail (zig-built tclsh90s 9.0.5); testbody_lint clean; make.tcl
projectversion consistency OK.
Claude-Session: https://claude.ai/code/session_01QgaxV27VZkmEec7oNbEVFc
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
|
2 days ago |
|
Julian Noble
|
39e0ec2ba7 |
G-084 achieved: settled cross-member -multiple rejection + value-parsekey integrity + storage-key attribution (punk::args 0.24.0, project 0.61.0)
G-084 increment 2 + achieved flip - parsekey completeness closed: Cross-member -multiple on a shared-parsekey option group is now a DEFINE-TIME error unless ALL members belong to one @opts group marked -parsekeymode error (G-083 mutual exclusivity makes per-member collection well-defined; a single -multiple member with its own parsekey is unrestricted). This settles the G-084 cross-member accumulation decision as rejection rather than accumulate-in-received-order: collection on one storage key is ambiguous (collect-vs-replace undefined), and rejection eliminates the silent loss without new collection semantics. The check runs AFTER the G-053 canonicalisation so range forms verdict correctly ({0 1} replace-shape allowed; a pre-canonicalisation placement would raise a raw non-boolean error for any range form). Value -parsekey integrity settled at define time: duplicate value parsekeys rejected (trailing values consume positionally - the shared slot silently overwrote), dash-led value parsekeys rejected (they landed in the options namespace of result/received and aborted parse with an internal error in the option-shaped validation paths; every option storage key is dash-led, so this one rule also removes all cross-class collisions). parse_status/parse_report storage-key attribution: parse_status_build now resolves a storage key claimed by exactly ONE argument (value -parsekey, declared option -parsekey, '='-suffixed optset tail) back to that argument, so argstatus/Source report received with the value-in-effect instead of default/0. Shared-parsekey groups stay unresolved - the remaining recorded display gap (engine-level per-member tracking; archived as the goal's open Follow-on). undefine/undefine_deflist now return "" (previously the whole internal cache dict to any caller capturing the value). define -help documents -parsekey for the first time (rename semantics for options and values, leader rejection, shared groups and -parsekeymode, last-defined-member defaults precedence, the -multiple rule, value-key uniqueness/non-dash). Moduledoc tclcore comment corrected (leader -parsekey settled as rejection, not pending). Internal sweep: dead values-loop identifiers removed (val_ident/val_ident_is_parsekey/values-side seen_pks), leader reverse map removed (VAL_NAMES-driven build), dead parsekey-from--default derivation removed, stale pre-G-084 comments corrected. Tests: parsekey_shared_key_multiple_rejected strengthened (message pins, {0 1}/{0 4} range coverage, the -parsekeymode error escape hatch incl collection and optionconflict); new parsekey_value_parsekey_integrity, parsereport_storagekey_attribution, parsestatus_storagekey_attribution; inc-1 comment damage repaired (duplicated block, eaten G-083 provenance lines, stale GAP-era comment); 2026-08-08 future-dating normalized to 2026-08-07 in touched files (CHANGELOG 0.59.0/0.60.0 headers corrected likewise; the underlying commits landed 2026-08-07 AEST). Goal flip (user-approved): the Goal line's accumulate wording amended to the settled define-time-rejection design; G-084 flipped achieved 2026-08-07 and archived (detail -> goals/archive/, index -> GOALS-archive.md, reference sweep applied - G-167 survey note gains achieved markers for G-053/G-083/G-084; Follow-on recorded: shared-group per-member received attribution => open). Detail file gains ## Progress (inc 1 + inc 2 + flip evidence). src/tests/modules/AGENTS.md punk/args index updated to the settled-pin state. Project 0.60.0 -> 0.61.0 + CHANGELOG (value -parsekey support and the define-time integrity rules are user-visible shell behaviour; G-053 precedent). punk::args 0.23.0 -> 0.24.0. Suites: punk/args 401/0; modules tree 1313 pass / 11 constraint-skipped / 0 fail (zig-built tclsh90s 9.0.5); testbody_lint clean; goals_lint clean; punkboot version-parsing 81/81; make.tcl projectversion consistency OK. Claude-Session: https://claude.ai/code/session_01QgaxV27VZkmEec7oNbEVFc Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
2 days ago |
|
Julian Noble
|
5037a9b51d |
.test authoring guard-rails: testbody_lint.tcl, runtests empty-actual hint, sanction 'return $result'; fix vacuous args.test test
- scriptlib/developer/testbody_lint.tcl (new): tcltest .test lint per
src/tests/AGENTS.md Verification. E-level: a -body whose final command
always returns empty vs a -result that cannot match empty under its
-match mode; test invocations with no -body at all (vacuous pass);
unbalanced-brace parse kills. W-level: uninterpretable shapes.
Hand-rolled command-word splitter (the corpus '{desc}\' style is
command-legal but list-illegal), namespace-eval recursion, literal
{*}{...} splices, -selftest (8 fixtures) + -stats. 142 files /
1674 tests parse clean.
- src/tests/runtests.tcl: FAILED reports with an empty actual and a
non-empty expected append a hint= (compact) / hint : (markdown) line
naming the trailing 'set result'/'return $result' convention, so the
failure carries its own diagnosis. JSON reports unchanged.
- src/tests/modules/punk/args/testsuites/args/args.test: the
literalprefix sibling of parse_withdef_value_leading_multiple_not_greedy
was passing VACUOUSLY - a missing description made the braced
description word swallow -setup/-body/-cleanup/-result, so tcltest ran
an empty body against the default -result "". Description restored,
orphan closing brace removed; the de-vacuized test executes and passes
(file 34/34).
- AGENTS.md (root), src/tests/AGENTS.md, .agents/.claude tcl-runtests
SKILL.md (byte-identical copies): the body-ender convention now
sanctions 'return $result' alongside 'set result' (verified on 8.6 and
9.0: default -returnCodes {ok return} accepts the return form,
-cleanup still runs, an explicit -returnCodes list omitting 'return'
rejects it loudly); linter and hint documented.
Verification: testbody_lint -selftest PASS on tclsh90s and tclsh86ts;
corpus lint clean; args.test 34/34; full suite in both runner modes with
runtests_parity.tcl PARITY: ok (1677 tests, 1631 pass). Residual 23
failures are pre-existing/environment, not this change: 12 maketclhelp
(auto_execok tclsh Bash-env trap; pass under PowerShell), 8
maketcllibfetch + 2 maketclbakelist (reproduce identically with this
runtests.tcl change stashed), core exec-14.3 (pins the pre-'-encoding'
exec error message vs tcl 9.0.5).
Claude-Session: https://claude.ai/code/session_016Bk571eG3prsJWYnewboW2
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
|
3 days ago |
|
Julian Noble
|
abd8616f3c |
G-084 inc 1: value -parsekey support + leader -parsekey define-time rejection + defaults precedence
G-084 (punk::args 0.23.0) increment 1 - parsekey completeness part 1:
Value -parsekey is now supported (was: accepted at define time but aborted
parse). A value record declaring -parsekey renames its result `values` slot
and its `received` key to the parsekey (e.g `@values {v1 -parsekey renamed
-type string}` parses `hello` to `values={renamed hello}`, `received` keyed
by `renamed`), and -multiple value members collect under the parsekey (the
`variable` setvalues form `?name value...?` pairs collect under `name_value`
- the motivating caller that carried the `#todo - fix -parsekey for leaders
and values` moduledoc comment).
Implementation: VAL_DEFAULTS and the values_dict init/storage are keyed by
parsekey (matching the already-parsekey-keyed VAL_REQUIRED); a new
api_valnames_received list tracks the parsekey (the internal valnames_received
stays argname-keyed for the -multiple first/continuation gate and G-053
occurrence enforcement); api_valnames_received drives the
required-satisfaction check and the received dict; a per-form
val_pk2name/leader_pk2name reverse map lets the valmin clause-length fallback
and the post-parse validation loop resolve a parsekey back to its argname
for ARG_INFO/arg_checks lookups.
-parsekey on a leader is now REJECTED at define time with a clear message
(was: silently ignored) - no live caller uses a leader -parsekey, so the
goal's no-silent-ignore contract is met by rejection rather than parallel
hot-path surgery; the dead req_name derivation in the leaders resolve branch
is removed. The @values directive line still rejects -parsekey (a
group-default parsekey is not a feature; per-arg value -parsekey is).
The defaulted-members precedence rule for a shared-parsekey group (none
received) is now documented as last-defined-member-wins (deterministic in
definition order) and the in-code `? review` is removed.
parsekey.test: the value GAP flipped to parsekey_value_result_key_settled,
the leader GAP flipped to parsekey_leader_parsekey_rejected, the @values-line
GAP split to parsekey_values_line_rejects_parsekey; the defaults GAP flipped
to a settled pin (parsekey_shared_key_member_defaults).
dev/parsekey-knownbugs.test: the value and leader disabled pins retired
(settled in parsekey.test); the cross-member -multiple accumulation pin
remains (increment 3, pending the design decision).
Full punk/args suite 399/0 (1 skipped), punk/ns 125/125, broader punk sweep
989/0. Cross-member -multiple collection (inc 3) and the parse_report/
parse_status received-attribution display gap (inc 2) remain.
Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=huggingface.co
|
3 days ago |
|
Julian Noble
|
c1f64ca0ab |
G-174 drafted (proposed): cmd_traverse word accounting - the doc-lookup walk never drops argument words
Root-caused from 'i a? web green' presenting the loose sample form (ordinal 5) instead of web (ordinal 3): punk::args form narrowing is sound (parse_status on the full tail reports the web/sample ambiguity correctly); punk::ns::cmd_traverse drops all but the last trailing word for commands whose definition's first form is zero-arg (a?'s sgr_overview) - the dropped words land in neither args_resolved nor args_remaining, so cmdhelp's advisory parse narrows on a one-word tail and cleanly selects the wrong form with no failure signal. The goal records the probe evidence, the four-step drop mechanism (first-form-only inspection, ensemble-shape test, the '#todo - return?' empty-leader break, the code-10 last-snapshot fallthrough), the consumer sensitivity split (docid-only vs args_remaining-sensitive), and the coverage determination: cmdflow/cmdhelp pin the choice-walk parity and post-resolution narrowing arcs, but nothing asserts the conservation invariant, no fixture has the zero-arg-first-form shape, and the code-7/code-10 lanes are unwitnessed. Approach: characterise first (defect witnesses pinned, flipped by the fix commit), minimal fix at the annotated sites, consumer verification, with first-form-only inspection and the ansi-side sample-form tightening held as recorded decision items. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
3 days ago |
|
Julian Noble
|
2aad4d1777 |
G-053 achieved: punk::args range-valued -multiple occurrence arity (punk::args 0.22.0, project 0.60.0)
-multiple now accepts a {min max} range (max -1 = unbounded) alongside the
legacy booleans 0/1, so a definition can declare at most once ({0 1}, a second
occurrence is a parse error), bounded repetition ({2 4}), or one-or-more
({1 -1}) instead of choosing between silent last-wins (0) and unbounded
collection (1).
The spec compiler canonicalises once at resolve: the stored -multiple becomes
the computed boolean (list-shape collect: true for max>1 or -1, false for
legacy 0 and max==1) so every existing collect-vs-replace / scalar-vs-list /
leader-value-single-multiple hot-path truth-test stays correct, and the range
companions (min/max/maxbounded) live in a separate per-form MULTIPLE_RANGES
dict - NOT in ARG_INFO, so they do not ride along when ARG_INFO is round-tripped
as a spec via resolved_def copyfrom. Legacy 0/1 and boolean strings
(true/false/yes/no) are coerced to the boolean and stay byte-unchanged (no
MULTIPLE_RANGES entry for unlimited cases, so {0 -1} is equivalent to legacy 1).
Resolve validation: max positive or -1, min <= max, and the -optional/range-min
reconciliation (non-zero min forces presence, contradicts -optional -> reject
with a clear message; declare -optional 0). The value-shape rule: max==1 forms
stay scalar, max>1 or -1 yield the occurrence list.
Parse enforcement: a new PUNKARGS VALIDATION occurrencecount failure class
(payload <argname> count <n> min <m> | max <m>) fires in a single post-loop pass
per section (opts/leaders/values) via a private::multiple_range_enforce
helper. Over-max is a hard contradiction (fires in both normal and
viability-probe modes; parse_status_classify maps it to invalid); under-min is
pure end-of-input exhaustion (SUPPRESSED in the G-152 viability probe via the
viabilitycheck arg, classified incomplete so a viable form reports incomplete
not invalid). Legacy -multiple 1 required with 0 occurrences still reports
trailingvaluecount (byte-unchanged, not occurrencecount).
Rendering: the usage-table Multi column reflects the range (0-1 / 2-4 / 1+ for
unbounded-with-floor; the greencheck stays for legacy 1), the string renderer
emits MULTI:0-1 etc., and the synopsis distinguishes at-most-once (?arg?, no
ellipsis) from repeating (arg...). -multipleunique/-multipleuniqueset compose
with max>1 ranges unchanged. define -help documents the range form and the
-optional/range-min rule.
New testsuite multipleranges.test (28 tests: 13 define-time
canonicalisation+validation, 12 parse-time enforcement incl parse_status
verdicts and a legacy-required-still-trailingvaluecount guard, 3 rendering).
Legacy untouched by default confirmed: full punk/args suite 399/0 (3 skipped),
punk/ns 125/125. All G-053 acceptance criteria met; goal flipped to achieved
2026-08-08 and archived (detail -> goals/archive/, index -> GOALS-archive.md,
reference sweep applied to G-072/G-084, deferred G-084 cross-member -multiple
collection decision recorded as a Follow-on => goal G-084).
Project version 0.59.0 -> 0.60.0 (minor: backward-compatible behaviour
addition). punk::args module 0.21.0 -> 0.22.0.
Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=huggingface.co
|
3 days ago |
|
Julian Noble
|
35008f7338 |
G-084 notes: record the shared-parsekey received-attribution display gap
parse_report's Source column and parse_status's per-arg `received` count both
report default/0 for an explicitly-supplied shared-parsekey member (e.g
parse_report {-exact ...} withid ::lsearch shows every MATCHSTYLE member as
(default) though the value is correctly bridged) - parse_status_build keys
the positions dict by the storage key, so no per-member received attribution
survives the fold. Pre-existing (G-151 bridge, G-049 attribution loop), not
introduced by G-083. Flip when the engine-level storage-key model G-084 owns
lands (the same follow-on that retires the value-bridge).
Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=huggingface.co
|
3 days ago |
|
Julian Noble
|
9125d56237 |
G-083 increment 3 achieved: punk::args argument-relations usage rendering + lsearch moduledoc adoption (punk::args 0.21.0, project 0.59.0)
The -conflicts and -parsekeymode error vocabulary (inc 1, 0.19.0) and its parse-time optionconflict enforcement (inc 2, 0.20.0) now surface in usage/arg_error and synopsis output, and the lsearch tclcore moduledoc models its documented option incompatibilities with the new vocabulary. Rendering: - A per-arg -conflicts list appends a "conflicts with: <targets>" hint to the argument's help text in the usage table (targets resolved to display names via lookup_optset, so a parsekey target shows its member flag, not the raw parsekey). - A named @opts group marked -parsekeymode error is annotated "mutually exclusive (distinct members may not be combined)" in its group header; override (default) groups carry no such annotation. - The synopsis one-line form carries no conflict detail. lsearch moduledoc (tclcore): - -sorted gains -conflicts {-glob -regexp} and -bisect gains -conflicts {-all -not} (per-arg conflicts; the whole group stays -parsekeymode override so -glob/-regexp remain last-wins). - The "(documentation incomplete - punk::args fixes required for grouped mutually exclusive options and prefix calculation)" caveat is dropped from the @cmd -help. clock clicks left unchanged - its active positional-choice definition already models exclusivity (exactly-one-of) and the acceptance clause names lsearch only. relations.test extended with 8 increment-3 pins: 4 rendering (conflicts_usage_hint, parsekeymode_error_usage_group_header, parsekeymode_override_no_header_annotation, conflicts_synopsis_no_hint) and 4 lsearch adoption (lsearch_sorted_conflicts_glob + lsearch_bisect_conflicts_all raise optionconflict via errorcode substring match, lsearch_glob_regexp_last_wins stays last-wins, lsearch_caveat_dropped). Legacy untouched by default (no -conflicts/parsekeymode-error definition carries no new rendering). Full punk/args suite 371/0 (3 skipped); tclcoreparity 10/10; punk/ns 125/125. All G-083 acceptance criteria met; goal flipped to achieved 2026-08-08 and archived (detail -> goals/archive/, index -> GOALS-archive.md, reference sweep applied to G-053/G-084, deferred -requires/-implies dependency relations recorded as an open Follow-on). Project version 0.58.0 -> 0.59.0 (minor: backward-compatible behaviour additions, following the G-173 punk::path precedent). punk::args module 0.20.0 -> 0.21.0. Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=huggingface.co |
3 days ago |
|
Julian Noble
|
43f009ed80 |
G-083 increment 2: punk::args optionconflict parse-time enforcement (0.20.0)
The -conflicts and -parsekeymode error vocabulary declared at define time
(0.19.0) is now enforced at parse. A new optionconflict failure class joins the
PUNKARGS VALIDATION errorcode vocabulary, mirroring optionmissing's shape:
{optionconflict <arg_a> <arg_b> received <receivednames>} - it names both
offending received arguments, for both per-arg -conflicts violations (any
pair, cross-group) and -parsekeymode error group co-occurrence (distinct
members of a shared-parsekey group).
The check runs in a single post-resolution site in get_dict_form (after the
optionmissing/valuemissing block), so the ordinary option path and the mash
(short-flag bundling) path share it - both raise identically. Checked against
RECEIVED arguments only (defaults never conflict), after prefix/abbreviation
resolution, on optset identity (a new optsets_received tracker for options,
since flagsreceived collapses shared-parsekey members onto one api_opt). Runs
unconditionally - a received conflict is a hard contradiction in any mode, not
end-of-input exhaustion, so parse_status_classify maps optionconflict to invalid
(not incomplete) and the candidacy/viability probe reports it.
Define-time check added: a group marked -parsekeymode error must also declare a
non-empty -parsekey (the strict mode only applies to a shared-parsekey group).
relations.test extended with 7 enforcement pins (raise cases for -conflicts
and -parsekeymode error, defaults-never-conflict, cross-group conflicts,
one-received-ok, override-legacy last-wins, parse_status invalid). buildversion
0.19.0 -> 0.20.0.
Bug found and fixed during the increment: the group-co-occurrence dedup check
initially used `ni` (not-in) where `in` was meant - `ni` returns true on an
empty list, so it skipped every member and the check never fired. Corrected
to `in`.
Legacy untouched by default: -parsekeymode override is the default and absent
-conflicts means no check runs, so the full existing suite (including the
pinned parsekey_repeat_ordering last-wins / prepend-defaults idiom) passes
unchanged.
Verification: full punk/args suite via canonical tclsh90s - 363 total / 360
passed / 3 skipped (punkargsKnownBug) / 0 failed; punk/ns suite 125/125 clean.
goals_lint clean. No punkproject.toml bump (module API addition, not shell-
level user-visible behaviour yet - the lsearch moduledoc adoption that ships
user-visible behaviour lands in increment 3).
Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=huggingface.co
|
3 days ago |
|
Julian Noble
|
055f4bbf2f |
G-083 increment 1: punk::args argument-relations define-time vocabulary (0.19.0)
Spec compiler (resolve) accepts and stores the new relation keys: - per-argument -conflicts <list>: parsekeys or flag names that must not be RECEIVED together with this argument; checked against received args only at parse time (defaults never conflict) - parse-time optionconflict enforcement lands in a later G-083 increment. - @opts-level -parsekeymode override|error: per named OPT_GROUPS group; error = distinct-member co-occurrence within a shared-parsekey group raises optionconflict at parse; override = legacy last-wins, the default. Both keys are cross-validated at resolve: -conflicts targets must name a defined argname or declared -parsekey; -parsekeymode requires -group and a value of override|error. Define-time integrity hole closed: a -parsekey colliding with a distinct defined argument's name (one not sharing that -parsekey) is now a resolve error instead of silently forming an implicit shared-key group (parsekey_collides_with_defined_optname_GAP flipped in parsekey.test to a define-time error pin, plus redundant-self-name and deliberate-shared-group guards). Design settled (recorded in the goal detail): -parsekeymode is per named group (OPT_GROUPS entry, keyed by the @opts line's -group), not form-wide. Rationale: lsearch uses per-arg -conflicts only (partial conflicts, groups stay default override); clock clicks uses -parsekeymode error on its single named group (full pairwise exclusivity). Unnamed shared-parsekey groups default override. Legacy untouched by default: -parsekeymode override is the default and absent -conflicts means no check runs; the pinned parsekey_repeat_ordering last-wins / prepend-defaults idiom and the full existing suite pass unchanged. New testsuite relations.test pins the define-time vocabulary (7 tests, matching the longopts.test/mashopts.test error-pin style). buildversion 0.18.0 -> 0.19.0. Verification: full punk/args suite via canonical tclsh90s - 356 total / 353 passed / 3 skipped (punkargsKnownBug) / 0 failed. goals_lint clean. No punkproject.toml bump (define-time only; no shell-level user-visible behaviour yet - the module change, not the shell product). Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=huggingface.co |
3 days ago |
|
Julian Noble
|
6b4aeac95f |
punk::args: characterise longopt style distinctions + mash edges (new longopts.test)
Gap-fill from a coverage review of long-option and mashopt parsing: the module's comment blocks specify style distinctions the suite never pinned. All behaviour was live-probed before pinning (characterisation, not aspiration). New testsuites/args/longopts.test (6 tests): - mixed optset -f|--file|--file= accepts inline/spaced/short-alias forms incl unique-prefix abbreviation of the longopt itself (--fi=abc / --fi abc) - gnu-only --flag= splits at the first = only (--filename=a=b -> a=b), accepts an empty inline value, rejects spaced and solo usage (badoptionformat) - plain --flag accepts spaced only; inline =value is invalidoption - the two =-member resolve-time definition errors (-type none with an = member; optional ?type? member without one) surface at first parse - single-dash -flag=value is never split at the = (invalidoption) - @opts -any 1 adhoc lane: undefined --flag=value splits to --flag + value, but single-dash -zz=5 stays a WHOLE adhoc flagname expecting a following value mashopts.test additions (2 tests): a mash containing an undefined letter is rejected whole (invalidoption naming the mashable set); under -any 1 -mash 1 defined flags still mash while undefined mash-lookalike tokens (-xy, -az) are adhoc flags taking the next word as value - never mash attempts. Error-shape pins use the first three -errorcode elements (PUNKARGS VALIDATION + failclass) rather than the ANSI-rendered messages; the two resolve errors are pinned by distinctive message substrings (plain text, no ANSI). Verified: runtests -include-paths modules/punk/args/testsuites/args longopts.test mashopts.test via tclsh90s - 13/13 pass, no warnings. src/tests/modules/AGENTS.md punk/args index entry updated (DOX pass). Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
3 days ago |
|
Julian Noble
|
f48bf78900 |
G-083: activate (proposed -> active) - punk::args argument relations
Activation-freshness overlap survey (goals_xref score G-083) confirmed the existing Notes already record the real relationships (G-053, G-072, G-084 - all linked). G-167's back-pointer explicitly declares a deliberate non-overlap with no sequencing constraint; remaining unlinked pairs are low-score term-only siblings with no path overlap. No new Related: lines warranted. This work is a prerequisite for G-084 (parsekey completeness), whose cross-member -multiple combination decision should be made with this goal's relations vocabulary in hand. G-053 (occurrence arity) is slotted before G-084 in the same sequence at the user's direction. Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=unknown |
3 days ago |
|
Julian Noble
|
3450a9d3cb |
Refresh stale punkbin commit-hash pins in archived goals (punkbin shellspy->punkshell sweep)
The punkbin artifact repo carried the wrong project token ("shellspy" - the
experimental subcommand, not the project name per root AGENTS.md) in 10 commit
messages (subjects AND bodies) back to the repo's earliest scaffolding. A
message-only `git filter-branch --msg-filter 'sed s/shellspy/punkshell/g'` over
the whole shellspy-bearing history rewrote all 10 (20 commits replayed, trees
byte-identical - --msg-filter cannot alter content), force-pushed to
origin/master (now 63f5e558). punkbin is git-only (no fossil).
The rewrite changed every commit hash in the swept range, so punkbin commit-hash
pins in archived goal detail files went stale. Refreshed all of them to the new
hashes so the records stay honest pointers rather than dangling references:
- G-172 (3 pins): ab5611d -> a73d862e; 59e7475 -> 63f5e558 (this goal's two
in-situ republish commits - my session's defect; the earlier 2-commit amend
had already left these pins stale and this refresh clears them).
- G-128 (3 pins): 1b471fc -> 2d2bf48 (punkres stamper publication).
- G-139 (1 pin): ee571ed -> e5c86eb (lib tier r1 publication).
- G-123 (2 pins): 1f10390 -> 0d49aa7 (retroactive schema-v2 sidecars).
- G-165 (1 pin): a22e16b -> a8303c4 (punkzip root-strip/exclude build options).
Per goals/AGENTS.md archived files are frozen except Follow-on dispositions;
this refresh is a sanctioned exception under explicit user permission to make
archived goals honest regarding hashes. Pure hash-token substitutions - no
prose, Scope, Goal or Acceptance text changed. The dated achievement-time
qualifiers around the pins (e.g "LOCAL ONLY - not pushed", "STAGED", "REMAINS
for acceptance") are point-in-time history and stay untouched. goals_lint clean.
Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=huggingface.co
|
3 days ago |
|
Julian Noble
|
2954d08f84 |
G-172 achieved: flip to achieved + archive (distributed-binary CPU floor)
Rebuild + in-situ republish (Approach 6) completed the acceptance. Full `zig build bootstrap` (pinned 0.16.0) rebuilt the r2 family WITH the G-172 cpu_floor field now in the embedded [provenance] record (re-stage picked up the build905.zig floor-field edit; zig-cached C compiles, only the wrap + emit re-ran). family_check passed for all three members (the new embed_cpu_floor/embed_cpu_model assertions green). Audit of the rebuilt r2 exes: all three PASS at the baseline floor (zmm=0 ymm=0 out-of-floor=0). Published in-situ to punkbin (git 59e7475, pushed origin/master): the three r2 exes + three r2 tomls replaced under existing names/revisions; each toml [provenance] now carries cpu_floor="baseline" + cpu_model="x86_64"; the [tests] tclcore line refreshed to the shipped-codegen gate evidence (passed=56041 failed=8, the Aug-06 baseline-rebuild test-gate - was 56039/9); library [tests] lines unchanged; sha1sums.txt: the 6 r2 lines updated. build_id unchanged per artifact (the floor field does not enter the identity digest); sha1/size/built updated. Every acceptance conjunct verified on this Zen 5 / v4 host: audit zmm=0 on the republished artifacts; -Dcpu=native still host-tuned (probe: native zmm=123 FAIL at baseline, PASS at v4 - the floor is a default, not a restriction); the republished tomls state the floor; punk-runtime surfaces it (info/list/use/fetch); use/run produce a named 0xC000001D diagnosis on a below-floor recorded cpu_model (verified by mocked-host tests - not end-to-end demonstrable here: every published artifact is baseline/v1 which every host meets, and no x86-64 level exceeds v4); fetch stays ungated. DOX closeout: detail file -> goals/archive/ (pure rename; Status achieved 2026-08-07; verification evidence + ## Follow-ons recorded). Index entry -> GOALS-archive.md (removed from GOALS.md). Reference sweep: only the GOALS.md index entry mentioned G-172 in the live tier (no pending-tense rewrites); design- decision Notes pointers added to G-105 (inherits the settled default_target cpu_model shape) and G-130 (consumes the baseline floor decision on its i686 axis). goals_lint clean (80 active, 93 archived); goals_followons parses the 2 open Follow-ons (r1-generation disposition; bin/ punk9* kit rebake). Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=huggingface.co |
3 days ago |
|
Julian Noble
|
a58d0117f4 |
Add punkpath agent skill: PATH-executable resolution + Tcl-command clash diagnosis
A new .agents/skills/punkpath/SKILL.md (canonical) with a byte-identical
.claude/skills/punkpath/SKILL.md copy for Claude Code discovery (verified
sync with a plain diff). Models the tcl-whatis/tcl-nslist shape: a single
~0.7s side-effect-free punkshell invocation answering two questions agents
guess wrong:
1. Which PATH executable a bare command name actually runs, and what
shadows it (overshadowed/overshadowed_by entry-idx fields). This is
the diagnostic companion to the tcl-runtests warning that bare tclsh
from a Bash tool often resolves to the MSYS tclsh - the skill gets
the machine-readable verdict on which copy wins instead of guessing.
2. Whether a PATH executable clashes with a Tcl command (tcl_conflicts
command:match=exact|nocase) - the 'invalid command name X' where X is
X.exe on PATH (dir/sort/find/clip).
Documents the -return json/text/dict forms, the -pathglob/-conflicts
filters, and the -context lever (G-173: pass it for a stable conflict
verdict through a wrapper/one-liner; the default-caller namespace is for
interactive use). Scope kept tight to PATH-resolution and Tcl-command
clash; explicitly disclaims being a general 'what is installed' tool.
No root AGENTS.md index change (the .agents/skills/ entry is generic and
does not list individual skills, matching the tcl-whatis/tcl-nslist/
tcl-runtests handling).
Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=huggingface.co
|
3 days ago |
|
Julian Noble
|
d4d70d3808 |
G-173 (achieved 2026-08-07): flip punk::path machine-parsable returns to achieved + archive
All acceptance criteria verified on tclsh9.0.5-punk (Tcl 9) and tclsh86ts (Tcl 8.6): existing path.test 67/67 green (table output byte-unchanged), pathcmd.test 7/7 (dict/json shape, ansi-freeness, round-trip parity, -context stability), full runner/testsuites/parser suite 23/23. Detail file moves to goals/archive/; GOALS.md entry moves to GOALS-archive.md (newest-first). Reference sweep: G-064 (proposed, the sibling lib.search -return dict/json goal) carried an open json-serialization-mechanism question that G-173's tcllib json::write decision resolves - added a pointer to G-064's Notes. Follow-on recorded (open): apply linecont_lint to the 46 remaining pre-existing line-continuation findings in punk-999999.0a1.0.tm (tech debt predating G-173, outside its edit sites). Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=huggingface.co |
3 days ago |
|
Julian Noble
|
f945bb12ca |
G-173 closeout: remove proc-body line-continuations; add linecont_lint + mandate -& doc blocks
The G-173 structured-renderer procs (path_conflicts_for_exe, path_entries, path_render_text, path_render_json, and the punk::path dispatch tail) were authored with \ line-continuation backslashes in proc bodies, violating src/modules/AGENTS.md (debug line-number matching). Refactor every G-173 helper to the dict-create + dict-set accumulator and json::write accumulator idioms so no proc body uses a trailing \. A stray continuation introduced at line 6752 (return [punk::path_machine_return ...]) is replaced with an lappend accumulator + expand. The user's doc-block edits convert the punk::path punk::args::define block from \ continuations to -& record- continuation markers with manually wrapped help text (table-width safe). Add scriptlib/developer/linecont_lint.tcl, a linter that flags \<newline> continuations outside punk::args doc blocks. It resolves a real Tcl parser (the C tclparser library, or punk::lib's pure-Tcl punk::tclparser fallback) and locates doc-block spans structurally (punk::args::define brace arg, lappend PUNKARGS command range) so the exemption is not a regex guess; comment-line trailing \ is also exempt. Sourceable linecont_lint_run proc with a script-mode guard; run as a script it exits 0 clean / 1 findings / 2 usage. 5 pinning tests in runner/testsuites/parser/linecontlint.test cover proc-body flagging, punk::args::define exemption, lappend PUNKARGS exemption, comment-line exemption, and a clean expand-style proc (pass Tcl 9 + 8.6). src/modules/AGENTS.md: the -& record-continuation marker is now REQUIRED for new and edited punk::args doc blocks (existing blocks may keep \ until touched; editing a \ block requires converting to -& in the same change); the closeout trailing-\ search is cross-referenced to the linter. Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=huggingface.co |
3 days ago |
|
Julian Noble
|
50f20f1cab |
AGENTS.md: add huggingface to the Assisted-by api-location table
huggingface.co is a public provider endpoint (acts as a proxy to backing providers). Add it so the Assisted-by trailer api-location derivation resolves the huggingface provider prefix to huggingface.co instead of unknown. Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=huggingface.co |
3 days ago |
|
Julian Noble
|
147f5104e7 |
G-173 (active): punk::path machine-parsable returns -return text/dict/json + -context + -conflicts
Activated per user direction; activated goal flipped proposed -> active (overlap survey re-run at activation; new Related: G-016 - sibling json round-trip consumer. G-042/G-019/G-061/G-110/G-167/G-171 deliberate non-overlap). ::punk::path gains three options: - -return text|dict|json (default table, byte-unchanged): the structured modes build an ordered entry list from the shared collect dicts (d_path_info/d_bin_info/d_index_executables) and serialise without ANSI/textblock. text is a fixed-key plain-ASCII layout; dict returns a Tcl dict; json is emitted via tcllib json::write (the preferred encoder settled by the runtests json_emit hardening) so booleans/counts are bare JSON numbers/1-0 and the decoded form equals the dict form byte-for-byte. - -context <namespace> (default the caller's namespace): makes the TCL-context conflict verdict stable for a machine consumer invoked through a wrapper. - -conflicts 0|1 (default 1): skip conflict computation for a small perf win. The table render is kept verbatim (byte-identical default output). The structured modes use a clean exact|nocase conflict classification (path_conflicts_for_exe) independent of the table render's ANSI path. A duplicate-path entry now reports the original's exe_count with an empty executables list (mirroring the table column). PUNKARGS argdoc updated; new options documented (whatis -doc punk::path shows the synopsis + per-option help + choices). Tests: src/tests/modules/punk/pathcmd/testsuites/tests/pathcmd.test (7 tests: dict shape, duplicate/overshadowed fields, json ansi-freeness + round-trip parity, text ansi-freeness, -context stability, -conflicts 0, default-table still ANSI). Passes on Tcl 9.0.5 and Tcl 8.6. Existing punk::path module tests (path.test, 67) unchanged/green. Versioning: punk module buildversion 0.2.8 -> 0.3.0 (minor - new user-visible options on ::punk::path); punkproject.toml 0.57.1 -> 0.58.0 (minor - new user-visible shell behaviour) + CHANGELOG entry. goals_lint clean; projectversion OK. Assisted-by: harness=pi; primary-model=zai-org/GLM-5.2; api-location=huggingface.co |
3 days ago |
|
Julian Noble
|
359a160f2c |
G-173 (proposed): punk::path machine-parsable returns (text/dict/json) with stable conflict verdict
Drafted per the proposal-first goals workflow (overlap survey run this session:
goals_xref.tcl paths over src/modules/punk-999999.0a1.0.tm,
src/tests/modules/punk/path, src/tests/testsupport/json_emit.tcl; cross-refs
recorded to G-064 sibling surface, G-049 established the -return dict pattern).
Status proposed; not activated.
The goal adopts tcllib json::write as the preferred JSON encoder for the
punk::path -return json mode - the decision surfaced by the runtests json_emit
hardening committed just prior (
|
3 days ago |
|
Julian Noble
|
f70aa63245 |
runtests JSON emitter: harden number predicate + C0 control-char escaping; add pinning suite
The hand-rolled -report json emitter in runtests.tcl classified values as JSON numbers via `string is entier -strict`, which (a) is deprecated in Tcl 9.0 per TIP 514 (string is integer became the canonical unlimited-range form) and (b) accepted several forms that are NOT valid JSON numbers, emitting them bare: 007 (leading zeros), +5 (leading plus), 0x10/0o17/0b101 (hex/octal/binary). The G-093 work already patched one symptom (a "0\n" result embedded a raw newline) but left the rest of the class open. The string escaper (runtests_json_string) also left most C0 control chars (0x00-0x1F) raw, which RFC 8259 forbids in strings - a test whose result_was carries a bell or NUL corrupted the report the same way. Hardening (src/tests/testsupport/json_emit.tcl, extracted from runtests.tcl so the new suite can unit-test it without a child run): - runtests_json_int? replaces `string is entier -strict` with an RFC 8259 integer regex (^-?(0|[1-9][0-9]*)$) - version-independent, eliminates entier entirely (no Tcl-9 deprecation exposure), subsumes the G-093 whitespace guard via ^...$ anchoring. - runtests_json_string now escapes the whole C0 range: 0x08 -> \b, 0x0C -> \f, others -> \u00XX. The existing \n \r \t \ \" escapes are preserved byte-for-byte (the multilinebanner.test end-to-end suite probes exact substrings such as "result_was":"x\n"). Pinning suite (src/tests/runner/testsuites/parser/jsonemit.test, 8 tests): RFC 8259 number syntax (007/+5/0x10/0o17/0b101 emitted as quoted strings; valid integers bare), C0 control-char escaping (no raw 0x00-0x1F survives; \b \f short escapes; \u00XX for the rest), entier-deprecation hygiene (asserts the loaded predicate proc body uses regexp and mentions no entier), named-escape byte regression, and tcllib-json round-trip of a representative testdict array. DOX: src/tests/AGENTS.md testsupport bullet records json_emit.tcl; src/tests/runner/AGENTS.md parser description records jsonemit.test. Verification: jsonemit.test 8/8 on Tcl 9.0.5 and Tcl 8.6; multilinebanner.test 2/2 (no regression to the existing emitter output contract); full runner/ subtree 40/40; mode parity runtests_parity.tcl PARITY: ok on a broad real-data subset (modules/punk/args, 35 files / 341 tests, singleproc vs multiproc). Tests-only change - no punkproject.toml bump per the versioning policy. Assisted-by: harness=pi; primary-model=zai-org/GLM-5.2; api-location=huggingface.co |
3 days ago |
|
Julian Noble
|
8f446690c1 |
G-172 (active): record + surface + audit the distributed-binary CPU floor
Approach 3 (record floor field): build905.zig declares cpu_floor="baseline" beside the default_target pin and writes cpu_floor + cpu_model into each family member's embedded record [provenance] (= target.result.cpu.model.name, so "x86_64" for the default build, "znver5" for -Dcpu=native - the floor is a default, not a restriction). family_artifacts.tcl derives both from the embedded record (G-117 single-source-of-truth) into the sidecar [provenance], failing if absent; family_check.tcl asserts them non-empty. Approach 4 (audit tool): scriptlib/developer/cpufloor_audit.tcl disassembles a binary's .text via a located objdump (llvm-objdump preferred, GNU fallback) or audits a pre-disassembled excerpt, classifies by AVX register width (zmm=v4, ymm=v3; baseline/v2 forbid both, v3 forbids zmm, v4 allows all), reports out-of-floor count + first RVA + sample mnemonics, exit 0 PASS / 1 FAIL. Bundled fixtures in cpufloor_fixtures/ are real disassembly excerpts (avx512, avx2, clean); selftest 12/12 cases pass. Verified on real binaries: rebuilt bin/punkzip.exe PASS (zmm=0); a -Dcpu=native probe FAILs at baseline (123 zmm) and PASSes at v4. Approach 5 (punk-runtime surfacing): both .ps1 and .bash twins updated and re-wrapped. Host CPU level detection (PS via IsProcessorFeaturePresent PF SSE2=10/SSE4.2=38/AVX2=40/AVX512F=41; bash via /proc/cpuinfo flags) feeds a meets/below/unknown/norecord verdict. fetch prints cpu floor (UNGATED - a cross-platform fetch for another machine stays free); list annotates rows the host cannot run (!CPU-FLOOR:v<req>, local rows only); use/run gate on below with a named diagnosis instead of a silent STATUS_ILLEGAL_INSTRUCTION (0xC000001D) - the 2026-08-06 AVX-512 regression failure mode; info adds the fields. bin/punk-runtime.cmd re-wrapped (deterministic; scriptwrap_runtime_cmd_roundtrip_no_drift test PASSES). Activation: overlap survey re-run at activation (goals_xref score G-172) - no goals drafted in the interval; existing Related: notes cover every pair. Project version 0.57.0 -> 0.57.1 (patch - user-visible punk-runtime behaviour shipped in kits). Remaining for acceptance: rebuild + republish in-situ (per Approach 6's alpha dispensation) and audit the republished artifacts to confirm zmm=0 - a publication-time, user-owned step. Assisted-by: harness=pi; primary-model=huggingface/zai-org/GLM-5.2; api-location=huggingface.co |
3 days ago |
|
Julian Noble
|
4b221ed960 |
goals-system v5: two-path activation rule - explicit user instruction IS the confirmation; agent-initiated activation forbidden
Assisted-by: harness=opencode; primary-model=huggingface/moonshotai/Kimi-K3; api-location=unknown |
3 days ago |
|
Julian Noble
|
ca51898470 |
G-035: update goal detail to reflect resolved mechanism investigation
The avoid-mixing rule is no longer folklore: both co-causes are now identified and pinned. Co-cause 1 is stock Tcl's $satisfied early return suppressing the library scan (silent older-version selection under prefer latest, order-dependent indexing). Co-cause 2 is the libunknown 0.2.2 namespace-context bug (fixed 0.2.3, pinned by pkgindex_relative_namespace_is_global). The rule is substantiated for stock Tcl and neutralised under the punkshell chain. Updates: - Goal: replaced open-ended 'substantiate or retire' framing with the resolved finding; remaining work is the crossform characterization suite and AGENTS.md guidance - Context: removed 'folklore' characterization; presented the mechanism as established knowledge with both co-causes - Restructured Questions into Resolved (identified-mechanism findings with reproduction table) and Still Open (same-version collisions, package forget crossing forms, packagepreference interaction) - Approach: references the new crossform suite complementing existing shadowing/ and discovery/ suites; guidance deliverables named - Notes: promoted the 0.2.3 bug fix from baseline note to co-cause alongside the $satisfied mechanism; both explain the original folklore - Identified mechanism section: restructured as two co-causes with explicit consequences replacing the either/or framing - Acceptance: tightened to enumerate the five characterization cases and name the specific guidance deliverables - Fixed typo: lllength -> llength in quoted stock Tcl source Assisted-by: harness=pi; primary-model=anthropic/claude-sonnet-4-20250514; api-location=anthropic.com |
3 days ago |
|
Julian Noble
|
d949274f2b |
G-172: correct the zig-gate rebuild-blocker claim; Approach 6 -> in-situ; r2 family + tools republished at baseline floor
The drafting session recorded the recipes' required_zig gate as blocking any rebuild on this box (a prerelease sorts below 0.16.0; "newest zig otherwise present is 0.15.2"). Wrong: only the stray dev copy is rejected - the zigpin toolchain bin/tools/zig-x86_64-windows-0.16.0/zig.exe is a released 0.16.0, present and passing the gate (the zig-x86_64-windows-* naming family in bin/tools was missed and a removal back-inferred from the artifact records). Notes bullet and Progress corrected; nothing was removed and nothing blocks. Approach 6 rewritten per user decision: republication is in-situ replacement under existing artifact names/revisions - punkbin immutability deliberately violable during alpha (all consumers in-house, zero third-party users). Progress: Approach 6 executed for the r2 generation + tools (punkbin ab5611d pushed) - full bootstrap rebuild with the pinned zig, zmm=0 ymm=0 audits on all five binaries, family_check PASS, two consecutive core-gate PASS runs with identical totals (interp-36.7 bgerror-race jitter and a 4h httpProxy.test ThreadLevel-1 hang documented). Record-honesty residual noted: published tomls carry emission-time [tests] summaries - refresh folded into Approach 3. Remaining: floor field, audit tool, punk-runtime surfacing; r1 disposition + bin/punk9* kit rebake behind those. goals_lint: clean (81 active, 91 archived). Claude-Session: https://claude.ai/code/session_01WXribLxUjk3dZUsN3r6rKb Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
3 days ago |
|
Julian Noble
|
946a32da19 |
G-172 drafted (proposed) + the four recipe pins: distributed binaries get a baseline CPU floor
The published tclsh9.0.5 runtime family is AVX-512-only. A flagless
bin/punk-runtime.cmd fetch hands a fresh checkout the curated win32-x86_64
default (tclsh9.0.5-punk-r2.exe), which on a CPU without AVX-512 raises #UD and
is killed with STATUS_ILLEGAL_INSTRUCTION (0xC000001D, errorlevel -1073741795)
before writing a byte - no message, no prompt, no error. First reported as a
Windows-version difference (25H2 works, 23H2 does not); it is not one.
Cause: the recipes called b.standardTargetOptions(.{}) and no build path passed
-Dtarget/-Dcpu, so zig resolved cpu_model .determined_by_arch_os by NATIVE
detection and ReleaseFast auto-vectorised to the Zen 5 build host. Measured: all
six published tclsh9.0.5-* artifacts (punk + plain, r1 + r2, both -bi- variants)
carry 2756 zmm-operand instructions each - real EVEX encodings, first site at the
front of .text, and Tcl has no runtime CPU dispatch - as do bin/punk905.exe,
punk9-dev.exe, punk9_beta.exe, punk9bi_beta.exe, punkdeclare.exe and
punkfiledemo.exe. bin/punkzip.exe carries 1724 and bin/punkres.exe 1423, so a
bake on such a machine fails too. An older generation (punk901*, punk9.exe,
xcritcl.exe, tclsh90s*) is AVX2-only from an earlier host: same defect, milder
floor. Confirmed across three machines - Zen 2 dies, Zen 4 and Zen 5 run.
Pins applied to all four in-repo recipes (suite_tcl90, suite_tcl86, punkzip,
punkres): standardTargetOptions gains .default_target = .{ .cpu_model =
.baseline }, so a flagless build is portable while -Dcpu=native remains the
developer opt-in and -Dtarget=<triple> is unaffected. Floor chosen as baseline
(x86-64 v1) - for a Tcl interpreter the codegen delta against v2/v3 is
negligible set against the default runtime failing to start. This has to be a
recipe DEFAULT rather than a build-command convention: "remember to pass -Dcpu=
when publishing" is precisely what failed silently for six artifacts.
Verified by zig ast-check on all four recipes, and by reproducing the mechanism
with the same toolchain and optimize level on an equivalent recipe (flagless ->
x86_64 baseline, zmm=0; -Dcpu=native -> znver5, zmm=805). No suite build was
run: all four recipes gate on required_zig = "0.16.0" with a .lt SemVer compare,
which rejects the bundled prerelease bin/tools/zig-0.16.0-dev.254 - a separate
pre-existing defect, recorded in the goal's Notes and deliberately not fixed
here.
Nothing is rebuilt or republished, and no shipped surface changes, so no project
version bump. Recording the floor in the artifact metadata record, an audit tool,
punk-runtime surfacing it on use/run (fetch stays ungated - fetching for another
platform is a normal workflow), and the rebuild/republish with the defaults.txt
repoint all remain open as G-172.
Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com
|
4 days ago |
|
Julian Noble
|
f103b90c3a |
G-166 achieved: cmdinfo/cmdhelp surface recognised-but-unavailable landings distinctly
Activated at user direction and achieved in the same session. On Tcl 8.6,
'i string is dict' (a class the runtime lacks) and 'i string is true' (a real
class) were indistinguishable to both eye and machine: cmdinfo reported
cmdtype 'doconly' with identical key sets, and cmdhelp's advisory parse of an
argument tail succeeded against the virtual docid, so 'i string is dict 5'
rendered exactly like a usable command line. Documentation REACHABILITY for
unavailable words is G-073's intent; the INDISTINGUISHABILITY was emergent -
nothing structural told a consumer the documented word is rejected here.
punk::ns 0.10.0
(a) cmdinfo gains an 'unavailable' key on EVERY result - empty, or the
canonical -choiceunavailable name the resolution addressed. It is a SECOND
AXIS beside cmdtype, never a cmdtype value: cmdtype answers what kind of
executable thing resolution landed on (both 'string is dict' and
'string is true' stay 'doconly'), availability answers whether the resolved
word is selectable in this runtime/context. Overloading cmdtype would have
broken every consumer switching on cmdtype eq "doconly", and the two only
co-occur incidentally. cmd_traverse's yielded/returned records carry a sixth
element for the attribution; cmdinfo keeps the final landing's value.
Population is nearly free - the level's choices-bearing first leader is
resolved once per level via the shared punk::args::choiceword_match, and both
the exact space-form landing and the prefix retry consult that one result
(the G-051 branch was restructured, not duplicated).
(b) PREFIX-LANDING DECIDED, enacting the G-073 doc-walk follow-on: a unique
prefix landing on an unavailable name now RESOLVES that name's virtual docid,
attributed - the conservative non-resolution was not kept. Encoding
availability as resolve-vs-don't-resolve conflated it with addressing and
made addressing inconsistent: 'i string is tr' resolved but 'i string is dic'
did not, while exact 'true' and 'dict' both did. Now every form of address
that names a documented word resolves it, and the availability key plus the
render marking carry the rejection story parse reports for the same word.
The choice traverse's G-040 parse-parity claim is untouched - an unavailable
word still identifies no SUBCOMMAND there, it is only attributed (the path
taken when the name has no space-form virtual docid).
(c) cmdhelp marks such subjects in every return form from one shared wording
(_cmdhelp_unavailable_marking, mirroring punk::args'
private::unavailable_choice_msg so a user meeting both reads one story):
-return dict carries the key, table/string render the message below the usage
under the non-info scheme, -return text leads with an 'UNAVAILABLE:' line.
The marking takes precedence over BOTH the ok-parse info render and any parse
failure message - the subject not existing here is the more fundamental
report, and it is the one a parse against a virtual id cannot make. Parse
status still passes through, so argument positions keep their marking.
punk::args::moduledoc::tclcore 0.4.1 - on a runtime lacking a curated forward
class, that class's per-class virtual id LEADS its help with the unavailability
statement (runtime patchlevel + the rejected call shape), then the unchanged
static description. The static version note existed but sat at the END of the
prose, so a reader met the full class documentation before learning the class
does not exist here. Runtimes where the class is live generate the id unchanged.
Tests: cmdflow.test gains the flowunavail / flowunavail_nodoc fixtures and 5
pins (key always present; exact landing attributed with cmdtype unchanged; the
prefix decision pinned against parse's choiceunavailable rejection;
choice-traverse attribution without addressing; attribution surviving an
argument tail). cmdhelp.test gains the helpstrua fixture and 5 pins (3 local
render, 2 tclcore-gated real-world) plus the two DELIBERATE key-list flips
(cmdhelp_cmdinfo_result_shape, cmdhelp_return_dict_valid). tclcoreparity.test
gains the virtual-id leading-note pin. All new pins are live-derived, so the
modern arm asserts the same shapes with an empty key.
Verified: cmdflow 25/25 + cmdhelp 50/50 on native 8.6.10 (THE AFFECTED ARM,
unavailable={dict}), 8.7a6 and 9.0.5; full punk/args subtree 341 total / 338
passed / 3 intended skips / 0 failed on all three; full punk/ns 125 green on
9.0.5 and green on 8.6/8.7 apart from two nsprimitives failures verified
PRE-EXISTING on both (confirmed with the module changes stashed); broad
src/tests/modules/*** sweep on 9.0.5 - 1257 tests, 108 files, 0 failed.
Consumer audit: every in-tree cmdinfo consumer reads named keys via dict get,
so the added key is inert for all of them. goals_lint and architecture_lint
clean.
Project 0.56.2 -> 0.57.0 (user-visible shell behaviour) with CHANGELOG entry.
Goal flipped to achieved and archived; four follow-ons recorded (to G-050,
G-044, G-055, G-167) with reference-sweep pointers pushed into those goals'
Notes, and the G-167 back-pointer added at activation - G-166 landing first
settles the marking-vocabulary ownership the two goals share. Also corrected
two pre-existing stale claims in src/tests/modules/AGENTS.md that this work
falsified: tclcoreparity's "version-note labels conditional on class presence"
(stale since G-073) and its "under 8.6 run the file directly via a plain
tclkit" note (stale since native 8.6 became a supported runner 2026-07-21).
Staged deliberately as this change-set only: the concurrent G-172 zig
CPU-floor lane (GOALS.md entry, goals/G-172-*.md, the four build.zig files)
is left uncommitted in the tree.
Claude-Session: https://claude.ai/code/session_01Y5vEJUZkD4p8jmEFEBVhLi
Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com
|
4 days ago |
|
Julian Noble
|
5bf9c83103 |
G-170 follow-on landed: argsid formhint authoring note in punkboot moduledoc (0.3.2)
The registration definition's 'argsid' key text now notes that a multi-form project definition may carry its own '@cmd -formhint' template of the shape ' <punkexe> help <name> %formname%' - the launcher narrows 'help <name> <formword>' (exact form name or decimal ordinal) to that form's usage, so the hinted line names real navigation. Authored with NO trailing ellipsis per the display-hint honesty rule: words after the form word are not a continuation of the narrowed render - on a parse=1 registration they dry-run through argsid instead. Doc-only patch bump 0.3.1 -> 0.3.2; follow-on disposition flipped to landed in goals/archive/G-170-launcher-help-depth.md. Verified: live render probe via bin/punk91 src script (definition parses, note renders in the argsid column), goals_lint clean, goals_followons reports the item landed, make.tcl modules mint green (punkboot-0.3.2.tm). No project-version bump - the doc-only source edit ships at the user-owned promotion+bake, where the project-version gate covers the pending bump. Claude-Session: https://claude.ai/code/session_01SCqtCXT1q45sjjMFfwARRj Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com |
4 days ago |