Skip to content

chore: pin the toolchain to topaz-1's own ref and regenerate vendored/ - #71

Merged
zxxma merged 1 commit into
feat/topaz-v2-renamefrom
chore/repin-gnoversion-topaz
Jul 20, 2026
Merged

zxxma merged 1 commit into
feat/topaz-v2-renamefrom
chore/repin-gnoversion-topaz

Conversation

@zxxma

@zxxma zxxma commented Jul 20, 2026

Copy link
Copy Markdown
Collaborator

Stacked on #70. Merge order: #69 → #70 → this.

The skew was real

GNOVERSION was 2c7f1abe (2026-06-01), six weeks older than the chain this is about to be published to immutably. topaz-1 runs fc4052651 (2026-07-17).

Roughly 15 vendored packages differed from what the chain hosts — including p/nt/avl/v0, whose Get is single-value on-chain and was two-value in the vendored copy. A green build genuinely did not prove publishability. The gap was held open only by a hand argument about which symbols gnodaokit happens to touch, re-argued on every change.

After regeneration the vendored p/nt/avl/v0 is single-value, matching the chain. The p/samcrew/avl fork is untouched and still exposes the two-value Get the data-structure role needs.

The other cost, which the identity review surfaced

The old toolchain has no sub-realm support at all. cur.Sub() fails preprocessing with missing field Sub in .uverse.realm, and the symbols are absent from the binary:

realmSubpathOf:             old(2c7f1abe)=0  new(fc4052651)=1
topmostCrossingFrameCurHIV: old(2c7f1abe)=0  new(fc4052651)=1
newSubRealmHIVPointer:      old(2c7f1abe)=0  new(fc4052651)=1

So nothing could exercise the sub-realm semantics #69's design reasons about, on the very compiler the freeze was being validated with.

What changed

15 vendored files changed content, one was added (p/moul/md/example_test.gno), and p/moul/md/filetests is gone upstream — its filetest became that example_test.gno. Every other vendored package exists at the new ref with an unchanged layout.

Verification

Run with the network unavailable, at the new pin: 10 packages ok · lint clean · gno fmt zero diff · zero gno: downloading.

The provenance comparison was run locally in both directions before pushing — 118 files checked, 0 drift, 0 deleted, 0 symlinks.

Deployer assert_gnodaokit_source_form topaz-1 passes against this tree.

Docs that asserted the old skew

vendored/README.md had a section headed "A green build here does not by itself prove publishable on topaz-1" and the provenance workflow's stdlib-avl gate carried the same rationale. Both corrected: the gate stays, but it is belt-and-braces now rather than load-bearing — with the vendored copy matching the chain, an ntavl.Get() two-value call fails the local build instead of surviving to AddPackage.

Supersedes

#63 proposes 5111dbc2, and #64's Makefile bump proposes ba9da8eb — both older than the chain. This takes the chain's own ref instead.

GNOVERSION was 2c7f1abe, six weeks older than the chain this is about to be
published to immutably. Now fc4052651, which is what topaz-1 runs.

The skew was not theoretical. Roughly 15 vendored packages differed from what
the chain hosts, including p/nt/avl/v0: its Get is single-value on-chain and was
two-value in the vendored copy. A green build did not prove publishability, and
the gap was held open by a hand argument about which symbols gnodaokit happened
to touch. Regenerating at the chain's ref closes it — the vendored stdlib avl is
now single-value, matching what AddPackage will resolve.

The old toolchain also had no sub-realm support at all: cur.Sub() failed
preprocessing, and realmSubpathOf / topmostCrossingFrameCurHIV are absent from
the binary. Nothing could exercise the sub-realm semantics the identity work
reasons about, on the compiler the freeze was being validated with.

15 files changed content, one was added, and p/moul/md/filetests is gone
upstream — its filetest became example_test.gno in the package. The fork at
p/samcrew/avl is untouched and still exposes the two-value Get.

Verified at the new pin with the network unavailable: 8 packages ok, lint clean,
gno fmt zero diff, zero "gno: downloading" lines, and the provenance comparison
run locally in both directions — 118 files checked, 0 drift, 0 deleted,
0 symlinks.

The stdlib-avl boundary gate stays, but it is belt-and-braces now rather than
load-bearing: a misuse fails the local build instead of surviving to AddPackage.
Its comment and vendored/README.md said the opposite and are corrected.
@zxxma
zxxma changed the base branch from freeze/exported-surface to feat/topaz-v2-rename July 20, 2026 18:50
@zxxma
zxxma merged commit 221c0c1 into feat/topaz-v2-rename Jul 20, 2026
5 checks passed
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.

1 participant