Skip to content

chore(ci): bump pinned gno toolchain (includes upstream #5357) - #63

Closed
zxxma wants to merge 1 commit into
mainfrom
chore/ci-bump-gno-toolchain
Closed

zxxma wants to merge 1 commit into
mainfrom
chore/ci-bump-gno-toolchain

Conversation

@zxxma

@zxxma zxxma commented May 11, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

Bumps GNOVERSION in the Makefile from 4e80c37e (gno @ 2025-09-29) to 5111dbc2 (gno master HEAD @ 2026-05-11).

Root cause

Every branch's CI was failing with:

amino: unrecognized concrete type full name vm.InvalidPackageError

This is fixed upstream by gno PR #5357 (commit 6d40b3207, 2026-03-25), which registers vm.InvalidPackageError / vm.InvalidFileError in the gno CLI's amino codec. Before that, the CLI couldn't decode the typed errors that vm.QueryFile started returning after #4410 (gnomod.toml v2 introduced typed errors in vm/qfile ABCI responses).

Last green CI on main was 2025-11-12 (run 19292629043/19292629044); no main commits since masked the rot.

Why this specific commit (and not v1.1.0)

I tried 96264d21d (v1.1.0 / chain/gnoland1.1, 2026-04-01) first — that's the gno commit running on gnoland1, so it mirrors deployed-chain behaviour. It did not work (CI run). Reason: the v1.1.0 release branch is for the chain runtime, not the CLI tooling. PR #5357 fixed a CLI/codec issue and was not backported to the v1.1 release line — only master has it.

5111dbc2 = gno master HEAD as of today (2026-05-11). Includes #5357 plus 6 weeks of stabilization. Subject: fix(tm2): add duplicate peer protection (#5319).

Trade-off vs picking a tagged release:

  • master HEAD isn't a release tag, so it could shift if we re-bumped naively. The pin is a 40-char SHA, so it's immutable in practice.
  • If/when gno ships a release line that backports #5357 (e.g. v1.1.1 or v1.2.0), switch to that for stability symmetry with the chain.

Diff

-GNOVERSION=4e80c37e8d1870aa5d3b01966ed4c5dfa18a7566
+GNOVERSION=5111dbc2287377990a1366cd369971cb299a1c43

One-line change.

Test plan

  • gno-lint job green on this PR
  • gno-test job green on this PR
  • After merge, rebase gnodaokit#62 (security audit v5 realm fixes) onto new main and confirm its CI goes green too

History

  • v1: 4e80c37e → 96264d21d (v1.1.0). Failed: lacks #5357.
  • v2 (current): 4e80c37e → 5111dbc2 (master HEAD with #5357). Force-pushed.

Notes

  • Marked Draft to mirror the Memba v7.1 Phase 1 drafts-only-first protocol. Pure CI/tooling change, no realm code touched — review surface is one byte string.

The previous pin (4e80c37e, gno @ 2025-09-29) can no longer unmarshal
current upstream package-server responses, causing every branch's CI to
fail with:

  amino: unrecognized concrete type full name vm.InvalidPackageError

Root cause: gno upstream PR #5357 (commit 6d40b3207, 2026-03-25) registers
`vm.InvalidPackageError` / `vm.InvalidFileError` in the gno CLI's amino
codec. Before that, the CLI couldn't decode the typed errors that
`vm.QueryFile` started returning after #4410 (gnomod.toml v2).

The v1.1.0 release branch (96264d21d, 2026-04-01) is the gnoland1.1 chain
release but does NOT include #5357 (release branches don't get CLI tooling
backports automatically). v1.1.0 is correct for chain runtime but wrong
for CLI tooling.

Master HEAD (5111dbc2 @ 2026-05-11) includes #5357 plus 6 weeks of
stabilization. This is the smallest pin that produces working CI today.

Trade-off vs picking a tagged release: master HEAD isn't a release tag,
so it could shift if we re-bumped naively. The pin is a 40-char SHA, so
it's immutable in practice. If/when gno ships a release line that
backports #5357, switch to that.

Unblocks gnodaokit#62 (security audit v5 realm fixes) and any future PR.
@zxxma
zxxma force-pushed the chore/ci-bump-gno-toolchain branch from ea6983c to 6bba257 Compare May 11, 2026 16:53
@zxxma zxxma changed the title chore(ci): bump pinned gno toolchain to v1.1.0 / gnoland1.1 chore(ci): bump pinned gno toolchain (includes upstream #5357) May 11, 2026
@zxxma

zxxma commented May 11, 2026

Copy link
Copy Markdown
Collaborator Author

v2 CI ran (gno-test, gno-lint) — error type changed but still fails. Progress confirmed:

Before this bump (4e80c37e, 2025-09-29):

amino: unrecognized concrete type full name vm.InvalidPackageError

The CLI couldn't decode the typed errors at all. PR #5357 fixed this.

After this bump (5111dbc2, master HEAD with #5357):

gno.land/p/samcrew/basedao:0: package has no files (code=gnoUnknownError)
gno.land/p/samcrew/daocond:0: package has no files
gno.land/p/samcrew/daokit:0: package has no files
...
package "gno.land/r/demo/profile" is not available
qfile failed: invalid package

The CLI now correctly decodes the error and surfaces TWO underlying issues:

  1. Local package discovery — gno test ./gno/... reports our own basedao / daocond / daokit source dirs as having "no files." Local discovery should not require an RPC round-trip; this looks like a schema or local-resolution change between the old pin and master. Possibly the gnomod.toml v2 format declared at gno = "0.9" is being interpreted differently now, or a build-tag / file-name convention shifted.
  2. Remote package availability — gno.land/r/demo/profile is reported as not available on whatever RPC the CLI defaults to. That's an ecosystem-side issue, separate from the CLI bump.

Both are deeper than this PR's scope (one-line SHA bump). Parking this investigation — the bump itself is the right direction (closes the amino decode bug, gets us a meaningful error message), but the gnomod.toml / local-discovery side needs a separate look.

Won't block gnodaokit#62 in the meantime — it's Draft, awaiting a secondary CODEOWNERS reviewer per Memba v7.1 Phase 1 protocol. CI red is acknowledged; we'll come back to it.

@zxxma

zxxma commented Jul 20, 2026

Copy link
Copy Markdown
Collaborator Author

Superseded by #71, which repins the toolchain to fc4052651 — topaz-1's own ref.

This proposes 5111dbc2, which is older than the chain gnodaokit is about to be published to. The freeze stack needed the vendored dependency closure to match what topaz-1 actually serves, so the pin went forward to the chain's ref rather than to an intermediate one.

@zxxma zxxma closed this Jul 20, 2026
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