Title broadened in both tiers; Goal/Acceptance gain the -exclude-paths
clauses: expose treefilenames' existing capability on runtests (or an
equivalent post-discovery globmatchpath filter), accumulate semantics
matching -include-paths, exclusion winning over inclusion, a
subtree-exclusion run discovering exactly default-discovery minus that
subtree, and both options pinned by the same regression-suite
combination matrix. Notes record the bundling and the middle-ground
motivation examples.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
User-directed from the G-092 measurement findings: a single
-include-paths value with four deep patterns matched only one subtree,
and repeated -include-paths flags matched zero files, both under the
default *.test tail glob - contradicting the documented accumulate
semantics. Context records the full repro matrix (multi-pattern forms
with explicit tails worked; standalone globmatchpath matched the same
patterns, implicating the treefilenames walk/prune or option-value
shape), the candidate root causes, and the tail-glob workaround. Notes
record that treefilenames already implements -exclude-paths unexposed
by runtests - the planned middle-ground option should be pinned by the
same regression suite.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
multishell.test (10 tests, ~105s solo child - the -jobs floor) split five
ways along shared-fixture boundaries, every test moved verbatim with its
provenance lines: multishell.test keeps wrap + structure/LF-only + exec
smoke (incl WSL); multishell_wrapverify.test (fresh-wrap checkfile) and
multishell_wrapdeterminism.test (byte-identical re-wrap) each re-pay the
wrap fixture; runtimecmd_checkfile.test and runtimecmd_roundtrip.test
guard the committed bin/runtime.cmd and its source round-trip with no
wrap fixture at all. argparsingtest.test (17 tests, ~32s) split into
timeit/discover/contract/first_call (11) + argparsingtest_compare.test
(compare_*, 6). Assessed and left unsplit: dtplite.test (24.6s solo),
scriptexec.test (12s); exec.test excluded as imported core material.
File headers point at the src/tests/modules/AGENTS.md punk/mix bullet
for the sibling map; the runtests longest-first weight table carries
measured post-split weights.
runtests_parity.tcl gains a -names mode: multiset comparison of
'- testname:' lines from '-report markdown -show-passes 1' captures,
file attribution ignored - verbatim moves compare identical, lost/
duplicated/renamed tests are reported.
Verified (native tclsh 9.0.3): repeated -jobs 8 full-suite runs 1m26.3s
/ 1m26.8s (was 2m16s at G-091, ~5m40s sequential - 4.07x vs the
pre-parallel baseline, 36.6% vs the G-091 reference); slowest child
30.7s / 31.3s (bound 45s); NAMES identical (989 occurrences) vs the
pre-split sequential capture; tallies identical in every run (989/973/
15/1, exec-14.3 sole failure, files 87 -> 92); PARITY ok -jobs 8 vs
-jobs 1 and between repeats; repo clean after runs. Notable finding
recorded in the goal file: child-wall contention inflation (~45% with
50s+ dominant children present) collapsed to ~15% once the work
distribution was flattened - splitting reduced both the critical path
and the contention itself.
G-092 flipped to achieved 2026-07-19 and archived (GOALS-archive.md
record, detail file with full verification evidence to goals/archive/).
goals_lint clean. Also noted in the goal file, out of scope: runtests
-include-paths multi-pattern/repeated-flag matching misbehaved during
measurement (single value with four deep patterns matched one; repeated
flags matched zero) - contradicts documented semantics, needs its own
investigation.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
User-approved draft from the G-091 verification data: multishell.test's
~2m child wall is the -jobs floor (full-suite -jobs 8 = 2m16s); splitting
the long suites (multishell, dtplite, argparsingtest, scriptexec if
worthwhile; exec.test excluded as imported core material) along
shared-fixture boundaries targets no child wall above 45s and at least a
25% full-suite improvement, with name-level test conservation, provenance
lines travelling with moved tests, weight-table updates, and parity
gating. Context records the fixture-duplication tradeoff (total CPU up,
wall down) and the complementary lean-children candidate.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
-jobs N (implies -singleproc 0; explicit -singleproc 1 conflicts) runs
test files through a tpool of N concurrent child processes with
per-child capture files and tcltest -tmpdir subdirs, longest-first
static-weight submission, and a -serial-paths serial tail (default: the
console-state suites modules/punk/console/** + modules/opunk/console/**;
piped-child-driven suites proved parallel-safe and each serial file
costs its full child boot strictly after the parallel phase). Cached
results feed the unchanged processing loop in discovery order, so
reports are identical in shape and order to sequential runs. Watch mode
emits file-collected events (completion order, output relayed there);
file-end (tallies) stays discovery-order. The runner now reports jobs
phase times and the ten slowest child WALL times (child wall includes
boot + file-level fixture cost that per-test usec misses).
Two Tcl-core-level behaviours found and worked around in the child
spawn (see the verification record in the archived goal file):
- concurrent BLOCKING execs from worker threads convoy on Windows (all
return at the longest concurrent child's exit; plain sleepers 2-12s
from 8 workers all walled at 12.0s) -> spawn via 'exec ... &' and
poll per-pid tcl::process status; blocking fallback retained for
runtimes without tcl::process (tcl 8.6).
- Tcl_ReapDetachedProcs runs on every exec and purges finished
background children process-wide, racing the poll (nondeterministic
'key <pid> not known in dictionary') -> tcl::process autopurge false
+ explicit per-pid purge after the status read.
Verified on native tclsh 9.0.3: repeated -jobs 8 full-suite runs
2m16.2s / 2m15.6s with PARITY ok vs -jobs 1, between repeats, and vs
the pre-parallel baseline (987-test identity, exec-14.3 sole expected
failure); 2.58x vs the ~350s sequential baseline (2.41x vs one
cooler-machine 5m28.8s sequential run - recorded in the goal file);
repo clean after every run.
G-091 flipped to achieved 2026-07-19 and archived (entry to
GOALS-archive.md, detail file to goals/archive/ with the verification
record; reference sweep updated runtests.tcl comments and the
src/tests/AGENTS.md marker). goals_lint clean.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Both user-approved 2026-07-18 from the runx -teelog / runtests watch-mode
work (a65f9e5b) and the -singleproc 0 completion (218fbb1c).
G-090 records the probe evidence deliberately not addressed in that work:
float-placement starvation of a tee under the float-locked capture
junction, insert_transform re-instantiation churn, parked-worker adoption
never re-applying settings (silent cross-target log misrouting), the
per-file teehandle worker-thread leak in unwatched singleproc runs, and
the tcludp Tcl 9 receive-fileevent first-datagram-only behaviour.
G-091 is the phase-2 parallel test plan over the phase-1 child spawn:
tpool -jobs N, per-child tmpdirs, serial group for console-sensitive
suites, parity + >=2.5x acceptance (floor ~90s, multishell.test bound).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
New goal (user-approved wording): make.tcl bakes a curated scriptlib into
kit vfs app/scriptlib so standalone binaries have working lib: scripts, with
precedence following the packagemode mental model - kit-internal-wins only
for the default 'internal' launch, src/dev modes putting the live project
scriptlib first, and proj: (G-033) adding the visited project's scriptlib.
Context records the 2026-07-18 findings: kit-internal resolver branch never
exercised (no build path exists), not-overridable policy scoped to default
mode per user direction, missing boot->runtime packagemode hint, cookfs
mount-root gap, and the out-of-scope app-shellspy mapping. goals_lint clean.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Goal for embedding a short marker in the usage/error table's bottom frame edge
(bottom-right @examples indicator, counterpart of the top 'Usage' title). Two
severable deliverables: frame-parity -subtitle/-subtitlealign on
textblock::class::table (must work with or without footer rows), and footer
rows with column spans independent of the body via configure_footer mirroring
the header API (the -show_footer/-ansibase_footer/-ansiborder_footer styling
knobs already exist but no footer-row methods do). arg_error's interim
'Example:' header row retires when this lands.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
New goal from the 2026-07-16 layout-system review: thin layouts with
generation-time infrastructure injection, fauxlink layout refs, and a working
overlay merge primitive. Detail file carries the review findings (unimplemented
merge, dead module-pathtype refs, stale fat snapshots, latent bugs), the staged
approach (hygiene first), and the hygiene-stage completion record.
Activated at user direction 2026-07-17.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The punk9win.vfs kit was split into punk9wintk90b2.vfs and
punk9wintk903.vfs; the treectrl/Img/twapi/cffi payloads G-020 relies on
live in punk9wintk903.vfs/lib_tcl9 (the actively mapped tclsfe/punk91
kit). Path-only update, user-approved; goal contract otherwise
unchanged. goals_lint clean.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
User-approved from the serversecretsync tm-path-shadowing incident
(8.6 require-before-path-setup silently selected a stale tomlish from the
machine-default tm paths over the project's vendored copy). Default
mechanism per user: vendor punk::libunknown + punk::packagepreference as
the light layout's bootstrap minimum. Related: G-012/G-027/G-029/G-035.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
User-approved wording (drafted from the 2026-07-14 cmdtrace investigation):
an attribution option (live|sourcefile|auto - sourcefile mode doubling as a
practical workaround for the upstream nested-switch mis-attribution, tcl
tktview 5d5b1052280c976ea3d4) and a documented machine dict return for
future hotspot-analysis/repl consumers (implies no interactive pause).
Context records the unexplored regime boundary (module .tm procs trace
relative, plain sourced procs absolute) as goal groundwork. goals_lint clean.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
A word rejected by an OPTIONAL argument's allocation screen (G-071
restricted-choice screen, or the basic int/double/bool/number/dict type
screen) that subsequently overflows the argument list now reports the
specific rejection instead of "Received more values than can be assigned":
- choiceviolation naming the word and allowed choices, with -badarg/-badval
and the final-validation message template (Leading/Trailing argument
attribution), for choice-screen rejections
- typemismatch naming the declared (normalized) type, with -badarg/-badval,
for basic-type screen rejections
Mechanism: private::get_dict_can_assign_value returns a 'rejection' record
(single-member clauses only; literal-category mismatches deliberately not
recorded); get_dict_form collects first-rejection-per-word from the three
allocation call sites (leader/value split trial, leaders loop, values loop)
into a per-form store; the two toomanyarguments raise sites prefer a
recorded rejection for the unassignable word. Genuinely-surplus words still
report toomanyarguments; words consumed by later slots parse unchanged;
multiform selection (rank_form_failures) untouched.
errorselection.test _GAP pins flipped to the pointed expectations (pre-fix
behaviour pinned in the previous commit). punk::args buildversion
0.12.0 -> 0.12.1.
Closeout: G-082 flipped achieved + archived (index -> GOALS-archive.md,
detail -> goals/archive/ with mechanism/verification/residue recorded);
G-073 note updated to point at the new rejection-record seam;
src/modules/AGENTS.md authoring-ergonomics bullet re-documented (-optional 0
workaround no longer needed on >= 0.12.1; make.tcl keeps it - it runs
against the bootsupport 0.12.0 snapshot). Project version 0.12.28 +
CHANGELOG.
Verified: full punk/args suite 228/224/0-failed (4 intended skips) and
punk/ns 57/57 under Tcl 9.0.3; errorselection.test 7/7 under Tcl 8.7;
make.tcl frobnicate still pointed; goals_lint clean.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Set G-082 active (user-directed). New errorselection.test pins current
single-form error selection before the fix:
- _GAP pins: an optional choice-restricted leader (make.tcl dispatch shape),
an optional choice-restricted value, and an optional typed (int) value each
reject a word that then overflows - all report generic
{toomanyarguments N index I} with no -badarg/-badval attribution,
discarding the specific rejection (regression vs punk::args 0.5.0)
- reference pins: the -optional 0 equivalents report pointed choiceviolation
with -badarg/-badval and Leading/Trailing argument messages
- guards: genuine surplus stays toomanyarguments; a rejected word consumed by
a later slot parses cleanly; multiform failures still route through
rank_form_failures (noformmatch)
Verified: 7/7 pass against the unfixed module under Tcl 9.0.3.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The detail files carried richer scope text than the index (user-directed
resolution: enrich the index rather than trim the files). goals_lint now clean.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
New proposed goal G-082: single-form parse error selection should prefer an
optional argument's specific choice/type rejection over the generic
"more values than can be assigned" overflow (regression vs punk::args 0.5.0
found while dogfooding in G-030; the -optional 0 workaround sites in make.tcl
and the AGENTS authoring-ergonomics bullet are flagged for revisiting when it
ships).
Sibling notes parked in existing proposed goals:
- G-028: make.tcl's name-only process-sweep misidentification (unrelated
same-named executables) - full-path matching rides the same
process-identification infrastructure as the locker report.
- G-025: punk-exe-hosted make.tcl runs the kit's pre-loaded module copies, not
bootsupport snapshots - a self-report surface for running-interp module
provenance would make the mixing observable.
- G-073: reciprocal relation to G-082 (both want the "surface the specific
rejection" seam in validation error reporting).
- G-075: synopsis renders the definition id rather than @cmd -name - consider
a fallback for non-command ids when reworking id rendering.
goals_lint: unchanged (only the pre-existing G-078/G-081 index-mirror findings).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
All 15 make.tcl subcommands are punk::args-declared ((script)::punkboot +
(script)::punkboot::<sub>, built from single-source summary/help/opt dicts with
shared option fragments and explicit @form -synopsis). Bare make.tcl / -help,
'help ?subcommand?' and '<subcommand> -help' render tabled usage; invalid
flags/subcommands raise punk::args usage errors (exit 1) with unambiguous
prefix resolution; 'shell' is a declared passthrough.
Prompt policy: new -confirm 0|1 (default 1) on every confirmation point
(bootsupport minor-staleness gate, vfscommonupdate REPLACE, kit-type mismatch).
Default prompts only on an interactive terminal and aborts fast with guidance
on non-interactive stdin (::punkboot::lib::stdin_is_interactive; Tcl 8.6 lacks
the -inputmode probe and is assumed interactive). -confirm 0 proceeds
(staleness, REPLACE) or skips the mismatched kit. The piped-y agent workflow is
retired; the old vfscommonupdate decline half-abort (printed done, exit 0) now
exits 1.
Degrade rule: punk::args/punk::ansi/textblock boot requires guarded; definition
load caught with fallback to a self-contained scan (-k/-dirty-abort/-confirm
supported) + plain punkboot_gethelp (availability NOTE factored into
punkboot_availability_note, shared by both paths); rendering degrades
independently of parsing (errorstyle minimal / plain help when the ANSI stack
is unhealthy); PUNKBOOT_PLAIN=1 forces degraded mode. punk::args joins the
bootstrap-tracked _runtime_deps staleness set (parsing contract only - the
rendering stack degrades instead; contract docs updated in
src/bootsupport/AGENTS.md + src/modules/AGENTS.md, workflow docs in
src/AGENTS.md + src/tests/shell/AGENTS.md).
Verified: usage/help/error paths, vfscommonupdate fail-fast and -confirm 0 runs
with closed stdin, staleness flag route live (args 0.5.0->0.12.0 tripped the
gate), degraded-mode help/check/info, full 'project -k -confirm 0' kit build.
Dogfooding finding recorded in the goal file: punk::args 0.12.0 reports generic
overflow where 0.5.0 reported choiceviolation for optional-choice mismatches
(worked around with required leaders; candidate error-selection improvement).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Formalize the follow-on work identified during the punk::args documentation
trials and ansi->html prototyping (2026-07-13/14):
- G-078 punk::ansi ANSI->HTML rendering api: promote the prototype ansi2html
converter + headless ansicat flatten entry into a documented punk::ansi
api (palette option, art mode with bold-as-bright and block sealing,
-assets embed|link|none using the vendored punkdoc-mono; docs-screenshot
enabler)
- G-079 default-colour semantics: renderspace replay output bakes explicit
SGR 40 for never-set backgrounds; preserve default fg/bg semantically so
html can theme-match any page background (current workaround: black pre)
- G-080 deterministic cell-grid rendering mode: inline-block cell emulation
to eliminate the residual sub-pixel seams the span-mode recipe accepts
- G-081 argdoc build pipeline: dev doc.* generation of doctools + punk-native
html + markdown from punk::args definitions against dev modules, incl.
per-form table rendering, deeper-namespace pages, @package surfacing
(G-075) and shared doc-set font assets
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
New punk::args::formcheck <id> ?-return dict|summary?: reports the form pairs
of a multiform definition that some argument list could cleanly match
simultaneously (the multipleformmatches inputs). Static pairwise pass over the
resolved FORMS enumerates each form's positional word-slot chains (leaders then
values; -optional and ?-wrapped clause-member branching; -multiple capped;
options excluded - order-free), screens aligned equal-length chains per
position (discriminator words via choiceword_match, type witnesses from a
small table), and CONFIRMS every candidate witness arglist with a real
single-form parse against both forms (parse_status - no raise, no
user-supplied words, no define/resolve cost). Findings are therefore witnessed
- discriminated form pairs cannot false-alarm; misses (exotic types,
option-requiring forms, enumeration caps) are the documented conservative
direction. Classes: type_weakness (discriminator aligned with a permissive
non-validating type - any/none/string/ansistring/globstring/expr/script) vs
structural (forms genuinely share an argument shape).
New @form key -overlapallowed <formname-list> sanctions a KNOWN overlap on
either pair member: the finding reports with sanctioned 1 and leaves the
result's unsanctioned list (the actionable/gate subset). Parse behaviour is
never affected; unknown form names are rejected at definition resolve
(end-of-forms cycle). @form directive doc updated.
tclcore moduledoc 0.3.4: the documented after cancelid/cancelscript overlap
(runtime-liveness ambiguity, 0.3.0 record) sanctioned on the cancelid form -
::after now reports zero unsanctioned findings while 'after cancel <id-shaped>'
still raises multipleformmatches. ::lseq deliberately unsanctioned: formcheck
reports range/start_count (witness {1 count 1}) AND range/count (witness
{1 by 1}) - both real, both rooted in the expr-typed end slot, kept visible
pending an expr syntax-validating type (G-069/G-070).
Tests: new args/formcheck.test (7) - no-finding cases (parse withid/withdef
pair, afterish/sharedform fixtures), class+sanction fixture, unknown-form
rejection, -return summary, ::lseq/::after pins with witnesses re-verified,
sanction parse-neutrality. punk::args suite 210/210, punk::ns 57/57.
G-074 achieved and archived (acceptance review in the detail file); G-055
verification gate gains the formcheck step (unsanctioned must be empty for
regenerated multiform commands).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Proposed goal, wording user-approved 2026-07-13: make the perl/python/node -e
reflex work on punk kits (top-level and script subcommand forms) instead of
erroring, without ever reproducing stock-tclsh argv-swallow/stdin misparse
behaviour. Companion to the tclsh -e guardrails commit (AGENTS.md entry +
claude deny hook).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
New goal G-076 (active): adopt the upstream tcl9 dead-console fix (ticket
f10d91c2d3, root-caused in G-039) via a single version gate shared by the
'help tcl' warning and the repl watchdog arming.
- punk::lib 0.4.3: has_tclbug_console_deadspin + pure classifier
tclbug_console_deadspin_applies, gate variable
check::tclbug_console_deadspin_fixed_in (empty = all Tcl 9 windows
runtimes affected). Buginfo dicts gain a mitigated/mitigation axis
orthogonal to level; the deadspin check reports mitigated when
punk::repl >= 0.5.0 is available to the runtime.
- punk::repl 0.5.1: repl::start arms the dead-console watchdog only when
the check reports the runtime affected - recording a verified fixed
release once flips both consumers.
- punk 0.2.6: 'help tcl' renders mitigated warnings subdued (term-grey)
with a "(mitigated)" annotation and the mitigation text; severity
classification unchanged.
- checkbugs.test: classifier truth table, live-check consistency, and
mitigated-axis validation in the generic every-check sweep.
The fixed_in flip is deliberately deferred until a released Tcl passes
the G-039 kill-procedure re-verification (watchdog disabled, clean
script-visible eof exit) - recorded as the goal's remaining acceptance
item.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Acceptance reviewed clause-by-clause in the detail file: auto-selection
on a unique clean match (forms.test GAPs flipped), candidate-naming
noformmatch/multipleformmatches errors, -form list restriction across
get_dict/parse/parse_status/arg_error, doc surface presenting the
best-matching form(s) ('i after cancel <id>' presents the cancel form;
's after cancel someid' underlines the matching synopsis lines),
explicit single -form unchanged, full punk::args/punk::ns suites pass
(full tree 822/836 with only the exec-14.3 baseline failure).
Flip edits: index entry to GOALS-archive.md (achieved 2026-07-13),
detail Status flipped with acceptance review and live-verification
record, reference sweep updates G-044 (candidacy API now available -
parse_status formstatus), G-055 backlink, G-072 backlink. Detail-file
move to goals/archive/ follows as a pure rename; goals_lint transiently
unclean between the two commits.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
User-approved draft from the 8.6-vs-9 string is dict forward-compat
requirement: a new definition key (candidate -choiceunavailable) for
choices that are recognised but not available in the current runtime -
displayed among the choices with their -choicelabels notes, participating
in prefix disambiguation via the shared choiceword_match resolver (so
abbreviations colliding with forward-version choices are ambiguous
everywhere), and rejected at parse with a tailored recognised-but-
unavailable message. Adoption by the tclcore string is model gives 8.6
users the dict class annotated and reserved (deliberately stricter than
real 8.6 for the di prefix - recorded as a user-sanctioned parity
exemption; full-word behaviour stays parity-true), plus an
i-string-is-dict documentation id generated from the static description.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Space-form docid prefix parity: cmd_traverse's space-delimited child
docid jump, on an exact-word miss, now resolves the word against the
current level's choices-bearing first leader via
punk::args::choiceword_match (the shared G-040 resolver - -choiceprefix,
-nocase, -choicealiases, denylist, reservelist honoured) and retries
with the canonical word. 'i string is tr' resolves to the
"::tcl::string::is true" documentation exactly when 'string is tr 1'
executes; 'string is bool' -> boolean; ambiguous ('d'), unknown ('zz')
and denied words stay at the parent exactly when parse rejects them.
resolvedargs records the canonical, mirroring parse normalization. No
second matching rule exists.
Truthful cmdtype: cmdinfo classifies cmdwhich-notfound + non-empty docid
as 'doconly' (was 'notfound') - documentation-only levels below real
commands (per-class ids like "::tcl::string::is true", documented TclOO
method docids; the method case adopts doconly per the recorded design
decision, refinable by G-052). Consumers audited: cmdhelp/synopsis/eg
are docid-driven, cmdtrace tests only for 'proc', punk::help topic
fallback uses cmdwhich; punk::lib script analysis updated in step
(doconly accepted alongside notfound, bucketing unchanged).
Tests: the four G-051 GAP pins flipped and renamed
(cmdhelp_pseudo_command_cmdtype_doconly,
cmdhelp_spaceform_docid_prefix_honoured + ambiguous/unknown guard,
cmdhelp_string_is_true_pseudo_doconly,
cmdhelp_string_is_prefix_honoured). punk::ns 53/53 (cmdflow + non-GAP
cmdhelp unchanged), punk::args 193 pass / 1 pre-existing skip,
punk::lib 35/35; full source-tree suite result in the detail file.
Live verification (punk902z src): 'i string is tr' renders
byte-identical to 'i string is true' except the failure banner echoing
the typed words.
Flip edits: G-051 index entry to GOALS-archive.md (achieved 2026-07-13),
detail Status flipped with acceptance review and verification record,
reference sweep updates G-052 Notes. Detail-file move to goals/archive/
follows as a pure rename; goals_lint transiently unclean between the two
commits.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Root cause: get_dict_can_assign_value blanket-satisfied the member check
for any argument carrying choices, so an optional restricted-choice value
(the lseq '..'/'to' noise word) consumed any word whenever arity
permitted, starving later elements/clauses and failing validation with a
confusing trailing-choices error ('lseq 0 10 2' shape).
Fix: an allocation choice screen using choiceword_match (the shared G-040
implementation - exact/alias/prefix/nocase semantics; -choicemultiple
words screened per list member within min/max), applied only where
allocation has an alternative: -optional arguments and further
occurrences of -multiple arguments. Required arguments are deliberately
not screened - the word must fill them and validation's choiceviolation
reporting stays informative (the existing parsestatus/cmdhelp/choices
pins forced and now document both refinements). -choicerestricted 0
behaviour unchanged.
Results: the allocation.test GAPs flipped ('1 2 3' and '1 2 by 3' parse
with correct clause shapes {{} 3}/{by 3}; '1 2 3 4' gets the plain
excess-values rejection instead of blaming the unrelated optional);
prefix normalization survives the screen; all noise-word variants and
if-shape guards unchanged. lseq moduledoc matrix parses per lseq.n under
-form range, and 'i lseq 0 10 2' works end-to-end. Remaining lseq
divergences trace to G-041 form selection or the -type expr laxity
recorded in G-055 - not allocation.
Also corrected an increment-1 mis-premise: parse_status ALREADY accepts
-form before the withid/withdef tail per its documented synopsis - the
probe had passed it trailing. GAP replaced by
parsestatus_form_option_order (correct-order acceptance incl. the form
key, trailing rejection). No code change was needed for that acceptance
clause.
Verification: punk::args suite 193 pass / 1 pre-existing skip / 0 fail;
punk::ns 53/53; full source-tree suite 813 pass / 13 skip / 1 fail =
exec-14.3 only (known baseline) - zero regressions (tclsh 9.0.3; lseq
verification in punk902z src, Tcl 9.0.2).
Flip edits: G-071 index entry moved to GOALS-archive.md (achieved
2026-07-12), detail Status flipped with the acceptance review and
before/after record; reference sweep updates G-041 (prerequisite fixed,
parse_status correction) and G-072 Notes. The detail-file move to
goals/archive/ follows as a pure rename; goals_lint transiently unclean
between the two commits.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
G-071 flipped proposed -> active (user direction 2026-07-12).
New allocation.test (7 tests) with reduced fixtures isolating the failing
shape from the tclcore moduledoc: the lseq-range shape (start ?sep? end
?by-step clause?) reproduces the moduledoc failures exactly. Current
mis-allocation pinned as GAP tests ("1 2 3" -> invalid blaming sep,
"1 2 by 3" -> incomplete blaming end, "1 2 3 4" correctly invalid but
misblaming sep), noise-word variants and the if-shape noise-word clauses
pinned as regression guards, parse_status -form absence pinned as GAP.
Also silenced four more unconditional debug puts on get_dict allocation
retreat paths ((111)/(222)/(333)/(444)) - these printed to stderr during
NORMAL successful parses at every optional-skip retreat, polluting
interactive help/parse output for noise-word-modelled commands. Behaviour
unchanged; the retreat logic they marked is exactly this goal fix target.
Full punk::args suite 193 pass / 1 pre-existing skip / 0 fail (tclsh
9.0.3).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Both user-approved drafts from the G-041 prework probe findings:
- G-071: value-allocation correctness for optional standalone values +
trailing optional-member clauses (the lseq "0 10 2" in-form failure),
error-blame quality, and parse_status -form support. A G-041
prerequisite: no lseq form accepts those arglists under the current
allocator, so form auto-selection alone cannot fix them.
- G-072: named compound clause types with per-alternate element typing
and arity for heterogeneous repeating clauses (try on/trap handlers),
building on the design sketches embedded in the ::try moduledoc notes.
Includes a recorded examination of Elixir gradual set-theoretic types
(user direction): punk::args -type | and literal() are already a
fragment of the set algebra; negation would subsume reserved-word
deny-lists; per-alternate arity falls out of union-of-shapes.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Both user-approved drafts arising from the G-045 tclparser assessment:
- G-069: dev-time lint cross-checking punk::args record-splitter boundaries
against tclparser parse command where the binary is available -
dialect-aware comparison (-& pre-join, ANSI strip), capability-gated,
zero new runtime dependencies for punk::args.
- G-070: pure-Tcl implementation of the tclparser parse-command API scoped
to what punk::lib consumes, parity-verified against the C library over a
probe corpus; replaces the tclparser_tcl error stub with
prefer-C/fallback-pure dispatch. Reference source to be user-provided
under TEMP_REFERENCE with recorded provenance.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Increment 5 (final): the container quoting rules pinned by defquoting.test
now appear in the punk::args::define -help documentation - braced values
fully literal (backslash sequences survive as typed); double-quoted values
get Tcl backslash semantics at record parse while $ and [] stay literal
with no substitution outside tstr placeholders; the backslash-escaped
placeholder idiom renders a literal placeholder. Rendered section verified
in punk902z src (meta-escapes display correctly). punk::args suite 186
pass / 1 pre-existing skip / 0 fail.
Acceptance review recorded in the detail file: all criteria satisfied
across increments 1-5 (@cmd -unindentedfields 0.6.1, i help alignment
punk 0.2.4, -& record continuation 0.7.0, @normalize + define_docs
consumer proof 0.8.0/punk 0.2.5, quoting documentation 0.8.1; suite grown
128 -> 187 tests, GAPs flipped, none weakened, full source-tree suite zero
regressions).
Flip edits: G-045 index entry moved to GOALS-archive.md (achieved
2026-07-12), detail file Status flipped with evidence in Progress,
reference sweep updates G-053 notes. The detail file move to
goals/archive/ follows in the next commit as a pure rename (per the
goals/AGENTS.md archive rules) - goals_lint is transiently unclean between
the two.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
arg_error's cmd-help display transform (undent " "+help, max 4 - the
'#unindentedfields ?' todo site) is now gated by "-help" membership in the
@cmd line's -unindentedfields list, mirroring the existing per-argument
gate. The single transform site feeds both the table and string renderers,
so left-margin-authored @cmd help renders its first line flush with
continuations in both. @cmd -summary carries no indent transform in any
renderer, so -unindentedfields for -summary is vacuously honoured. No
in-tree definitions set -unindentedfields on @cmd - existing rendering is
unchanged. define doc for -unindentedfields now states where the option is
valid.
Tests: rendering_unindentedfields_cmd_help_GAP flipped to
rendering_unindentedfields_cmd_help (aligned expectation, both renderers
measured). Full punk::args suite 175 pass / 1 pre-existing skip / 0 fail;
punk::ns suite 53/53 (arg_error consumer); make.tcl modules builds clean
(tclsh 9.0.3).
Also: G-045 flipped proposed -> active (user direction 2026-07-12);
progress and remaining-work recorded in the detail file.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
An interactive shell orphaned by its hosting console dying (killed
conhost/terminal) previously spun ~2 cores forever: the Tcl 9 windows
console driver never delivers the dead-console state to the script as a
fileevent (tclWinConsole.c ConsoleEventProc only notifies on buffered
data) while its reader thread busy-loops on the persistent channel error,
so the script level is completely blind and nothing ever closes the
channel.
Fix: repl::console_watchdog - a self-rescheduling liveness poll (default
5s, repl::console_watchdog_ms; read-only GetConsoleMode probe via
chan configure -inputmode) armed by repl::start only for a tcl9 console
input channel serving the process-default console on windows. On probe
failure it closes the input channel (stopping the driver's reader
thread) and finishes the repl via the normal eof done-path; app-punkshell
then finds no console reopenable and exits cleanly. Piped, foreign-console
and tcl 8.6 inputs never arm it. repl::start's post-vwait reader
deregistration now tolerates a watchdog-closed channel.
Verification (2026-07-12): kill procedure post-fix exits the orphan in
1.5-2.0s with no traceback (src mode and rebuilt kits punkbi/punk91/
punk902z); 25s live-console soak shows no spurious trigger, CPU delta 0;
piped stdin unaffected; repl consolebackends 3/3 and punk::console suites
88 pass / 1 skip; root cause verified against plain tclsh 9.0.3.
Upstream ticket filed: https://core.tcl-lang.org/tcl/tktview/f10d91c2d3
Included:
- goals: G-039 flipped achieved 2026-07-12 and archived (index entry to
GOALS-archive.md, detail file to goals/archive/ with full root-cause,
repro and verification record); reference sweep adds the eof done-value
producer note to G-038
- src/modules/punk/AGENTS.md: watchdog contract documented
- project version 0.12.1 -> 0.12.2 + CHANGELOG entry
- vfs payload sync: _vfscommon.vfs punk/repl 0.4.0 -> 0.5.0
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
New goals (user-approved drafts, all proposed) covering the declarative
vendoring direction:
- G-065 declarative vendoring: toml-declared external packages with
version/commit pinning, retrieval provenance, binary-scan gating with
punkproject.toml override; manual drop-ins stay supported (surfaced as
undeclared); agent-free by design
- G-066 pkgIndex.tcl-to-.tm repackaging: lib.copyasmodule expansion with
embedded LICENSE/metadata datafile and distribution-unit tracking for
multi-package upstreams (tcllib-style)
- G-067 module artifact channel: publish/retrieve prepared .tm modules
against configurable artifact servers, sharing the G-006 consent-gating
and source-override pattern
- G-068 agent-assisted moduledoc generation workflow for vendored
third-party libraries (G-055 pattern generalized), decoupled from basic
vendoring
G-063 contract extended (user-approved): vendored-package license
indications additionally carry verification provenance (method, verifier,
date), reported by the audit surface.
Proving case recorded in G-065 notes: the hand-trimmed tcl_oauth2_library
copy was load-tested then removed before ever being committed, so the
first vendoring of it runs end-to-end through the declaration; the
TEMP_REFERENCE clone remains the upstream stand-in.
goals_lint: clean.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Finalises the G-062 achieved flip staged by the prior session (punkshell
0.12.1): LICENSE.txt (BSD-2-Clause, Julian Marcel Noble 2023-2026) at the
repo root, README/punkproject.toml/AGENTS.md license declarations, index
record to GOALS-archive.md and detail file to goals/archive/.
Archive-loss prevention, from the G-015 incident (its achieved-status
verification parenthetical was silently dropped by the 2026-07-11 two-tier
restructure's Status-line normalisation; the text is restored to the G-015
archive file's Notes with provenance):
- root AGENTS.md 'Doc Restructures': deleted-line conservation check before
committing bulk doc migrations, structured-field-normalisation hazard
(relocate prose to body sections first), moves separated from edits
- goals/AGENTS.md: lintable header field grammar (one line per field, bare
Status values), flip-time verification evidence recorded in the body, the
archive move content-identical, and a new archive-time reference sweep:
pending-tense mentions of the archived goal updated in the live tier, and
actionable archived notes pointed to from the affected live goal's Notes
- GOALS.md maintenance rules: achieved-flip clause extended to require the
evidence write-up and the reference sweep
Retroactive reference sweep over all 12 archived goals (25 live detail
files): pending-tense fixes (G-002 G-008 G-013 G-016 G-017 G-021 G-022
G-031 G-042 G-044 G-050 - G-017's Goal/Acceptance annotated as
gate-satisfied, contract meaning unchanged), forward pointers pushed from
archived notes (G-062 license-seeding follow-on -> G-012/G-027, G-037
vendorlib/vfs drift semantics -> G-024, G-058 static-baseline mechanism ->
G-005), and one 'Archived-goal references' Notes bullet per remaining
referencing file so every live mention of an archived goal is resolvable
without consulting the archive index.
scriptlib/developer/goals_lint.tcl (new, plain tclsh): validates detail-file
header grammar, strict status values in both tiers, orphan rules in both
directions, Status/Scope mirror consistency, plus an informational (never
fatal) warning for live files referencing an archived goal with no
achieved/archive marker - baseline now zero warnings. Root AGENTS.md
scriptlib/developer ownership note widened to cover developer utilities.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Acceptance verified (consolebackends.test child-process drivers): repls run
interactively through ssh-socket and tk-widget ::opunk::Console subclasses
with size/at_eof/can_respond answered by the overrides, base class and
punk::console unchanged. Detail file records increment 2 and the non-gating
residue (per-run output batching, punk/0-only junction wiring, raw-mode
untested on foreign consoles, process-global console state -> G-008).
Project version 0.10.3 -> 0.11.0 with changelog entry (new user-visible
shell capability: repl::init -console).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
G-062: canonical project license - root BSD-2-Clause LICENSE.txt with
SPDX-identified references in README.md, root AGENTS.md and a
machine-readable license field in punkproject.toml.
G-063: per-package license tracking - SPDX-normalized Meta license
indications via a friendly-name mapping module (authors need not know
SPDX ids), copyleft/viral-license flagging audit surface, and fixing
the literal %license% template placeholder seeding.
G-064: dev lib.search machine-parsable returns - ansi-free -return
dict/json with documented structure (G-049 pattern) plus an option
surfacing per-package license indications in table and machine returns.
All three proposed with user-approved wording (2026-07-11).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
- src/tests/AGENTS.md: agent-added tests carry a one-line provenance comment `#added <YYYY-MM-DD> (agent[, G-<id>])[ - <short note>]` - immutable facts only (never "G-NNN will flip this" expectations, which live in _GAP_ test names and goal files); user-added tests optional; no bulk retrofit of existing suites. Rationale: provenance must survive where VCS history does not travel - packaged test modules (G-029), git+fossil dual history, file regeneration/moves
- GOALS.md maintenance rules: marking a goal superseded/abandoned includes a test sweep - grep G-<old id> (at minimum src/tests) plus the tests named in the goal's detail-file Acceptance, and record each test's disposition (transfers to the superseding goal / stands down to plain characterization / lapses) in the appropriate detail file; the sweep never deletes, skips, or weakens tests on its own
- the provenance lines are what keep the supersession sweep's grep reliable over time; closes the dangling-GAP-pin risk analyzed for superseded goals
- no project version bump: governance docs only
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
- GOALS.md collapsed to summary entries (ID/status/title/Scope/detail pointer): 104.6KB -> 23KB so a full read stays cheap; Scope lines kept verbatim so path-grep goal discovery still works
- canonicality split: index canonical for ID/status/title/Scope, detail file canonical for Goal:/Acceptance:; all 61 detail-file headers reconciled from the index (fixes stale copies, e.g. G-036 detail acceptance predated the achieved rework), bodies preserved untouched
- detail files created for the three index-only goals G-017, G-034, G-054; every goal now has one
- 10 achieved goals (G-007 G-015 G-036 G-037 G-040 G-046 G-049 G-054 G-058 G-059) archived: summary records appended to GOALS-archive.md, detail files moved to goals/archive/; archive-on-flip replaces the ~30-achieved-entries trigger
- GOALS.md preamble/rules reworked: read workflow (read index in full, then detail files of scope-intersecting goals before editing), contract spans both tiers with proposal-first unchanged, next-free-ID counts the archive, achieved flip judged against the detail-file Acceptance and includes archiving
- progress convention codified from G-001 increment-1 practice: index entries carry status only, never progress; incremental progress lives in the detail file's ## Progress section (landed/remaining); achieved flip requires the remaining-work list resolved - a partial increment never flips a goal
- goals/AGENTS.md rewritten for the split (required 4-field header, one detail file per goal, orphan rules covering archive/); root AGENTS.md Child DOX Index + Repo-wide Notes updated; GOALS-archive.md preamble updated
- stale detail-file paths updated to goals/archive/ in src/lib/app-punkscript/punkscript.tcl (comment), runtimebash_wsl.test, wslprobe.tcl and the G-042 detail file; propagated copies (root lib/, src/vfs/_vfscommon.vfs, src/_build) left to refresh via normal build steps
- no project version bump: governance docs and source comments only, no user-visible shell behaviour
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
- three new backend modules under src/modules/opunk/console/, base ::opunk::Console and punk::console UNCHANGED (the G-001 acceptance constraint - verified diff-clean):
* opunk::console::test / ::opunk::TestConsole - deterministic channel-pair test double: fixed size via -columns/-rows, is_console_or_tty/can_respond 1 (settled value wins), at_eof = plain chan eof with NO probe (pending bytes never consumed - pinned). The console seam identified by the G-044 repl characterization work
* opunk::console::ssh / ::opunk::SshConsole - socket-carried terminal sessions: construction-time capability (detection is the wrong tool - per goal detail), chan-eof without byte-consuming probes, size via the registered ANSI size-query provider over the connection. Flagship test: a scripted remote terminal answers CSI 6n over a socket pair and size resolves 100x30 through punk::console machinery querying the socket
* opunk::console::tk / ::opunk::TkConsole - Tk text widget as terminal: widget path in the inherited in/out slots (documented non-channel reuse), terminal_class tk-text, size from widget char dims, at_eof via backend marker (opunk::console::tk::set_eof) or widget destruction; no Tk require at module load. Verified live under punk91 src (the tk-loading experiment kit)
- voo -extends subclassing pattern recorded in opunk/AGENTS.md: children inherit public accessors + field INDEX variables (not parent-private my.* accessors) - constructors initialise inherited private fields via index vars, method bodies use parent public accessor methods; virtual dispatch via the slot-0 tag needs no base changes
- tests: modules/opunk/console backends.test (8 tests: dispatch, spec_resolve acceptance of subclass values, probe-free eof, settled-capability precedence, ssh capability/eof/size-over-socket/settled-0-no-emission, tk gated behind env PUNK_TEST_TK=1) - 7 green + tk skip on tcl 9.0.3 and 8.7; tk case verified standalone under punk91
- goal detail records progress + remaining work (repl -console launch wiring, output-channel parameterization, interactive acceptance verification); goal stays active
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
- the ::tcl::string::is definition harvests its class set at define time from the bad-class error message of a deliberately invalid probe (safe, side-effect-free) - accept/reject parity with the loading interpreter by construction: 8.6.13 = 21 classes (no dict - the previous static 9.0 list wrongly ACCEPTED string is dict there), 8.7a6 = 23 (+dict +unicode), 9.0.3 = 22 (unicode removed). Fallback to the 9.0 set if the message format ever changes
- hand-written man-page descriptions (kept verbatim, tstr-processed as before so the A_WARN highlights are unchanged) apply only to classes the runtime accepts; accepted-but-undescribed future classes get a generic label; static version notes added to dict (not in 8.6) and the new unicode entry (unreleased 8.7 only - removed in tcl 9). The per-class virtual docids (::tcl::string::is <class>) follow the harvested set automatically
- new tclcoreparity.test (4 tests, gated on have_tclcoredocs): choices equal the live-harvested set and every documented choice is really accepted; per-class docids exist for every class; error-vs-ok outcome agrees between real string is and parse_status across the 23-shape probe matrix (missing args, trailing flag-like str, option/class prefixes + ambiguity, unknown option/class, -failindex var consumption, per-version dict/unicode presence, divergent-classification shapes); version-note labels conditional on presence. Expectations derived from the LIVE interpreter, never version arithmetic - green on 9.0.3, 8.7a6 and (via a direct tcltest driver, since runtests infrastructure does not run under the plain 8.6 kit) 8.6.13
- full punk/args + punk/ns trees green (219 pass + 1 pre-existing skip) incl the have_tclcoredocs cmdhelp pins
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
- G-016 goal+acceptance: -return option (default table unchanged; dict = canonical machine form with documented per-checkout records incl the VCS field; json = faithful rendering for out-of-process consumers, round-trips the dict data)
- G-017 acceptance: the documented agent invocation standardizes on -return json as the parse target with its record fields (replaces the vaguer "expected output shape")
- G-016 detail records the design decisions: dict-first assembly (table rendered from it), suggested record shape, json via tcllib json::write, markdown deliberately not offered as a machine form (json for parsing, table for humans)
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
- ConPTY (windows) + pty (unix) expect-like harness: spawn a built punkshell on a size-controlled pseudo-terminal, send keys incl control/arrow sequences, await patterns in the rendered ANSI byte stream with timeouts and forced teardown, full transcript capture
- the durable byte-accurate tier of the repl verification model recorded in goals/G-044 detail: pure layer (commandcomplete.test, done), console-seam unit tier (G-001 as enabling refactor), rendered tiers (G-020 screen capture near-term windows bridge; this goal headless + cross-platform)
- first verification targets = preserve-list behaviours: the in-proc {set x "{*}{"} continuation-hint sequence, raw-mode tab-marker rendering/deletion, up-arrow recall + edit of a multiline history entry
- detail file records open implementation questions (driving ConPTY from tcl: twapi surface vs helper exe; harness/shell synchronisation incl the inter-subshell beep-protocol idea vs prompt sentinels) and the G-059 WSL / G-060 QEMU-guest execution extensions
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
- new src/tests/testsupport/wslprobe.tcl (::punktest::wsl): memoized capability probe for the wsllinux constraint - default distro launches and unames as Linux, tools present (bash/mktemp/wslpath + sha1 tool), native tempdir file round trip; any failure -> skip with reason, so wsl.exe-present-but-unusable installs cannot false-positive. Staging helpers (staging_create/copy_in/run_in/cleanup) enforce the design: execution on the distro NATIVE filesystem only, shared /mnt used for one-way copy-in/out, the windows checkout never operated on from inside WSL (DrvFs slowness + cross-boundary stat differences that make git re-hash its index and fossil see phantom changes). Probe uses wsl -e invocations only (wsl --status/-l emit UTF-16)
- multishell.test: scriptwrap_multishell_exec_wsl_sh - the polyglot sh payload produces its marker under the distro bash from native staging (the direct isunix twin still skips on windows)
- new shell/testsuites/binscripts/runtimebash_wsl.test (8 tests): runtime.bash first-ever execution on real unix - multi-candidate run errors listing candidates, use/list/run round trip with argument passing, PUNK_ACTIVE_RUNTIME override, unknown/unselectable use errors, stale-active guidance, single-candidate fallback, offline list -remote (crafted cached sha1sums.txt + PUNKBIN_URL forced fetch failure), and the checkout-untouched guard (git status --porcelain identical before/after per acceptance)
- green on tcl 9.0.3 + 8.7 against Ubuntu-24.04/WSL2; suites skip cleanly when the probe fails; enablement/limitations in src/tests/AGENTS.md (known limitation: a present-but-HANGING wsl can stall the probe)
- GOALS.md G-059 flipped to achieved 2026-07-11; detail file records the outcome incl the UTF-16 discovery and the 512-byte label regression this work surfaced (caught by the guardrail suite as designed)
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
- G-059: capability-probed WSL constraint for driving unix-side tests from Windows (multishell sh-payload execution, runtime.bash behaviour). Native-filesystem execution is a design constraint: per-test artifacts staged into a WSL-native tempdir with one-way copy-in/out, no repo checkout inside WSL for these cases, future full-suite mode via a separate native clone - the Windows checkout is never operated on via /mnt (DrvFs slowness + cross-boundary stat differences force git index re-hashing and fossil phantom changes; two VCS clients must not share a working tree). Acceptance includes verifying the Windows checkout git/fossil state is untouched by a WSL-gated run.
- G-060: QEMU guests as strictly external tooling behind a hypervisor-agnostic guest-driving contract (provision/stage/run/collect, push-based - same pattern as G-059; guests never share a working tree with the host). GPL-safe posture recorded: external-process invocation only, no QEMU binaries or derived code in punkshell or punkbin, guest OS images not redistributed - fetched/built per machine. First target FreeBSD x86_64; config home deferred to the buildsuites toml direction.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
- punk_main.tcl captures the runtime static baseline at kit boot: empty-filename [info loaded] prefixes probe-loaded in a throwaway interp (configurable denylist tk*/vfs/mk4tcl/vlerq/tdbc* for side-effecting/composite inits), recording only packages a load actually PROVIDES (diff on provides, not package names - probe loads can trigger index scans) into ::punkboot::static_packages/static_prefixes; the boot interp is seeded with ifneeded {load {} <prefix>} mappings
- punk::lib interp_sync_package_paths/snapshot_package_paths propagate the baseline and seed fabricated interps/threads; punk::repl codethread init (new %staticprefixes%/%staticpackages% scriptmap) and shellthread worker init do the same - the repl code interp on a static runtime resolves Thread again (punk91 = tclsfe-x64 + punk9win.vfs failed with "can't find package Thread"; the appended vfs replaces the stub's //zipfs:/app mount taking its static pkgIndex files with it, and the tclsfe PostInit hook seeds per-interp auto_path only)
- punk::packagepreference static-vs-bundled policy: requiring a baseline package ensures the static mapping and triggers the package unknown scan BEFORE resolution so bundled copies register and the highest version wins (newer bundled reachable, static not masked by older bundled); static registrations excluded from the loaded-shared-object same-version pinning
- composite statics defer correctly: static twapi provides nothing under the probe (script layer lived in the replaced stub zip) so the bundled complete twapi is the resolution; punk91 vfs/vfs::zip warning re-diagnosed as kit content (tclsfe bundles no tclvfs) - recorded in the goal detail, not a seeding defect
- tests: staticseed.test (4) + staticpolicy.test (4) un-gated simulated-baseline units (green 9.0.3 + 8.7); staticruntime.test (4) kit integration behind the capability-probed statickitavailable constraint (probes ::punkboot baseline incl Thread; env PUNK_STATICKIT_TEST_EXE override, default bin/punk91.exe) - 4/4 against the rebuilt punk91; punk902z/full shell tree/full source-tree suite baselines unchanged (exec-14.3 only)
- verification runtime tclsfe-x64.exe pinned in the punkbin artifact repo (separate repo, commit b2dbbe6) with sha1sums + new AGENTS.md scaffolding
- punkshell 0.6.0: CHANGELOG entry; AGENTS updates (src/vfs capture contract, tests indexes incl new punk/lib + punk/packagepreference entries, shell staticruntime contract)
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
- diagnosis from the punk91 tk-loading experiment (tclsfe-x64 + punk9win.vfs): plain tclsfe stub worker threads resolve static Thread 3.0.5 fine, but punkshell boot replaces package search paths wholesale and discards the runtime seeding - the repl code interp cannot find Thread ("failed to load punk::console"); punk902z unaffected only because its vfs bundles a thread dll
- fix shape: capture the static baseline (empty-filename info loaded entries + versions) before path replacement, seed fabricated interps/threads with 'load {} <pkg>' ifneeded scripts, and make punk::packagepreference resolve static-vs-bundled by a documented version-aware policy (punk91 loaded the OLDER bundled twapi 5.0.2 dll over the newer static Twapi)
- testing: un-gated unit tests against simulated baselines + kit-boot integration tests behind a capability-probed constraint (static-entries-incl-Thread kit present) - skips cleanly, extends to zig-era static runtimes (G-005)
- artifacts: verification runtimes (tclsfe-x64.exe at minimum) to be pinned in the punkbin artifact repo (win32-x86_64 + sha1sums.txt) so the constraint is satisfiable off this machine; .vfs folders already git-tracked
- detail file records probe evidence, the runtests-testinterp precedent (same bug class, directory flavour), the suspected quiet thread-over-builtin instance on punk902z, and the vfs/vfs::zip re-diagnosis follow-up
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
- default icon src/runtime/punk1.ico, per-kit override from the kit's .vfs; mechanism per tcl-sfe (TEMP_REFERENCE/tcl-sfe, Ashok P. Nadkarni): RT_ICON/RT_GROUP_ICON replacement via twapi resource-update APIs, ordered so the appended vfs payload stays intact (icon stub before append, or sfe-style split/update/reattach)
- detail file records the _vfscommon.vfs override-detection wrinkle (common punk1.ico merges into every built vfs - consult the custom pre-merge folder or an explicit mapvfs.config element), idempotence via delete-then-write, twapi/non-Windows skip path, and resource-inspection acceptance
- follow-on candidate flagged (detail file only): RT_VERSION stamping - copyright defaulting to the input runtime's own version-resource copyright, Product Name indicating the core Tcl version by default, both configurable via punkproject.toml and/or per-vfs config
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
