Skip to content

feat(t27os): G3+G4 live mycelium — PTY tap fold, k-hop reading, batch fold, k-hop cache (Closes #12779) - #13031

Merged
gHashTag merged 5 commits into
masterfrom
lane/land-tap-ring
Oct 31, 2026
Merged

gHashTag merged 5 commits into
masterfrom
lane/land-tap-ring

Conversation

@gHashTag

Copy link
Copy Markdown
Collaborator

Closes #12779.

The lane's G3+G4 work as one batch: PTY panes folded into the work graph (pty_live_graph.t27: pty_state_to_kind, pty_parse_line, pty_fold_batch), the k-hop reading of the folded graph (graph_k_hop.t27: graph_fold_batch, k_hop_read, k_hop_read_cached keyed by (seed_pid, k_hops) and invalidated by graph_digest), the batch-fold optimization, and the two fixes the landing gates demanded (the t27b CSE-hoist in pty_parse_line; the undefined-read and never-hitting cache in graph_k_hop).

Gates on the merged tree (tri land --check, every gate passed): tests 6/6 and 8/8, 0 vacuous; t27b test --check clean on both specs (0 JIT/interpreter mismatches); seals MATCH; tri types ratchet CLEAN; dupe_scan failure set a subset of master's.

Refs #12988 — the T27F Live Actors Map loop; this is its gap "land the Queen band tap ring from lane/queen-band-graph".

Landed by hand because tri land's stage step cannot execute: a443044 puts two commands in one argv ( census pin --bless git add .trinity/seals -> clap exit 2), and the built binaries carry an older variant that fails with git exit 129. The step sequence above is tri land's own; the tool defect is filed separately.

…atch fold (Closes #12779)

G3 (petal C): specs/t27os/pty_live_graph.t27 — PTY pane events from
~/.t27os/pty/ folded into work graph via pty_state_to_kind, pty_parse_line,
pty_fold_batch (single pass, batched apply_event). pty_fold kept as thin
wrapper for backward compatibility.

G4: specs/t27os/graph_k_hop.t27 — k-hop traversal from seed pid over folded
work graph via graph_fold_batch (single pass, batched apply_event),
k_hop_read BFS with trace dump. graph_fold_ring kept as thin wrapper.

Both: batch fold optimization #1 — one parse pass, apply_event per pair.
Tests: G3 6/6, G4 5/5. Seals saved.

No cron, no seal drift, no new epics, no fixture/demo data.
phi^2 + phi^-2 = 3 | TRINITY
… graph_digest (Closes #12779)

Adds k_hop_read_cached over k_hop_read:
- key: (seed_pid, k_hops, graph_digest(g))
- cache hit: skip BFS, copy stored output
- cache miss: recompute, store result + digest

Tests: 8/8 (incl. cache hit returns same total, cache invalidates on digest change).

Batch fold (#1) untouched. No cron, no seal drift, no new epics, no fixture/demo.
phi^2 + phi^-2 = 3 | TRINITY
…2988)

The landing gate (t27b test --check, specs/tri/land/land.t27 step 13, #11225)
refused: t27c's optimizer hoists e - i, a subexpression over the stable locals
e and i, as a _cse temp above their declarations. The span and its newline are
computed once into consumed before any use; the trailing-space skip after the
state parse is dropped (its s was never read again). A comment line returns
the same count; at end of text the outer loop terminates either way.

Before: t27b exit 2, unsupported construct FnDecl(reference CSE hoist).
After: t27b exit 0, 6/6; t27c test-report 6/6.
…hable (Refs #12988)

Three defects the landing gate (t27b test --check, #11225) and its own tests
had failed to catch, found while landing:

1. The miss path snapshotted the whole caller buffer; k_hop_read writes only
   its lines, so the copy read bytes never written (interpreter fault; the JIT
   tolerated it). Every byte of the buffer is now defined first.
2. kcache_valid was never set: the cache never hit, and the on-hit test passed
   with two recomputations. The flag is set on a miss, a hit counter
   (kcache_hits) makes the claim countable, and the two tests now assert the
   hit and the invalidation by delta.
3. A hit copied kcache_out_len bytes regardless of the caller's buffer;
   the copy is now bounded by the smaller of the two.

Before: t27b exit 4, 2 JIT/interpreter mismatches. After: t27b exit 0, 8/8;
t27c test-report 8/8.
… fold, k-hop cache (Closes #12779)

Merge of master and the seals, by the tri land step sequence (specs/tri/land/land.t27), run by hand: the stage step of a443044 puts two commands in one argv and cannot execute. The gates ran on the merged tree: tests 6/6 and 8/8, t27b --check clean on both specs, seals MATCH, tri types ratchet CLEAN, dupe_scan failure set a subset of master's. Refs #12988.
@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-31 21:33:46 UTC

Summary

Status Count
Total Open PRs 38
PRs with Failing Checks 34
PRs with All Checks Green 4
READY 3
FAILING 34
PENDING 0
NO CHECKS YET 0

These columns do not partition: 3 + 34 + 0 + 0 = 37, and there are 38 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=e75b498e5e79 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

@gHashTag
gHashTag enabled auto-merge (squash) October 31, 2026 21:34
@gHashTag
gHashTag merged commit b5a49a7 into master Oct 31, 2026
38 checks passed
@gHashTag
gHashTag deleted the lane/land-tap-ring branch October 31, 2026 21:34
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[EPIC] T27F Live mycelium on all 27 T27 OS petals -- the swarm's work graph visible everywhere

1 participant