Repository navigation
rebrand-literal-gutter RED on main: #682 reintroduced the banned Net4sats literal on 7 surfaces after #684's scoping #722
Description
Summary
tests/packaging/rebrand-literal-gutter_test.sh is RED on current main: PR #682 (dual-brand SSID recognition, commit 659de8a) landed AFTER #684's scoped rebrand gutter (commit fc639db) and reintroduced the banned Net4sats literal on 7 tracked surfaces — including one SHIPPED file and the live brand-prefix table in the source.
This makes make release-check (packaging gate) fail on main today, deterministically, on any host.
Verified on
- Commit:
68ad5144ab09cb4723d8f3f655cbba1ac7ea9de5(main, 2026-10-07); also reproduced on the release-prep branch off it - Any host; no docker needed
Evidence (verbatim suite output)
== the re-brand's name appears nowhere in the tracked tree
FAIL the banned literal is back (7 hit(s)):
packaging/files/etc/uci-defaults/99-tollgate-setup:# (hasTollGateSSID: TollGate-/Net4sats-),
src/wireless_gateway_manager/brands_test.go: "Net4sats-OQ3Q", // whitelabel brand
src/wireless_gateway_manager/brands_test.go: "NET4SATS-AA11", // case-insensitive whitelabel
src/wireless_gateway_manager/discovery_log.go:// TollGate-…, and Net4sats-… for the whitelabel brand — mirrored by the
src/wireless_gateway_manager/discovery_log.go:// installer's brandingCommands). A Net4sats-branded TollGate must be
src/wireless_gateway_manager/discovery_log.go:var tollGateBrandPrefixes = []string{"TollGate-", "Net4sats-"}
src/wireless_gateway_manager/vendor_element_manager.go: // whitelabel Net4sats-…, case-insensitively — see hasTollGateSSID). This
== the shipped uci-defaults reference only known uhttpd sections
FAIL unknown uhttpd section name(s) in the uci-defaults: luci
tests: 4 passed, 2 failed
(The second FAIL is a separate cause — see the companion issue for the uhttpd.luci comment trip.)
Reproduce
git clone https://github.com/OpenTollGate/tollgate-module-basic-go && cd tollgate-module-basic-go
git checkout 68ad5144ab09cb4723d8f3f655cbba1ac7ea9de5
bash tests/packaging/rebrand-literal-gutter_test.sh ; echo EXIT=$? # 4 passed, 2 failed, exit 1
# or the whole gate:
make release-check # Packaging: FAILTroubleshooting guide
- Read the gutter's scope:
tests/packaging/rebrand-literal-gutter_test.sh—BANNED='net4[s]ats', with a scan-tree allowlist that test(packaging): the rebrand gutter scopes its ban to shipped surfaces — docs may discuss the whitelabel #684 ("the rebrand gutter scopes its ban to shipped surfaces — docs may discuss the whitelabel") carefully calibrated BEFORE fix(discovery): recognize both brands' SSIDs so Net4sats can be an upstream — rebase of #618 #682 existed. - What fix(discovery): recognize both brands' SSIDs so Net4sats can be an upstream — rebase of #618 #682 changed:
src/wireless_gateway_manager/discovery_log.gonow carriesvar tollGateBrandPrefixes = []string{"TollGate-", "Net4sats-"}— the whitelabel prefix is now a first-class runtime constant, not a re-brandable surface. The uci-defaults comment and test fixtures reference it. - The product question to answer first: is
Net4satsstill a re-brandable whitelabel (then these surfaces must not name it literally — derive from a brand config/constant indirection), or has it become a supported dual brand (then the gutter's ban on these surfaces is obsolete and the TEST must be re-scoped — e.g. ban the literal only in generated/shipped portal assets and marketing surfaces, allow the recognized-brand tables in Go source and their tests)? - Interaction with test(packaging): the rebrand gutter scopes its ban to shipped surfaces — docs may discuss the whitelabel #684's intent: test(packaging): the rebrand gutter scopes its ban to shipped surfaces — docs may discuss the whitelabel #684 deliberately kept docs discussing the whitelabel OUT of the ban. The 7 hits are code + one shipped uci-defaults comment. Decide surface-by-surface; the shipped-file comment (
99-tollgate-setup) is the only one that ships to routers.
Fix hints
- Either: replace literal
Net4satsindiscovery_log.go's prefix table with a brand-registry constant sourced from packaging/brand config, keeping the literal out of source; or update the gutter's scope to permit the brand table + its tests + the uci-defaults comment while keeping the ban where it matters (portal bundles, LuCI assets). - Whichever way, restore
make release-checkPackaging to green on main — it currently blocks any release gate run.
References
- test(packaging): the rebrand gutter scopes its ban to shipped surfaces — docs may discuss the whitelabel #684 (gutter scoping), fix(discovery): recognize both brands' SSIDs so Net4sats can be an upstream — rebase of #618 #682 (dual-brand SSIDs — introduced the hits), fix(setup): silence the absent-brand-file shell noise in uci-defaults #568/fix(setup): silence the absent-brand-file shell noise in uci-defaults — rebase of #568 #680 context on uci-defaults hygiene
Activity
Amperstrand commentedon Oct 7, 2026
Lane de-confliction handoff (this lane stands down from #722 per w1:p2B ownership).
The main checkout was found sitting on
fix/722-rebrand-gutteratcb4db7fwith uncommitted WIP — now committed as a handoff snapshot so it can't be lost:37d8540(local, on this host's main checkout; not pushed — the branch has no upstream and the owning lane owns its shape).What the WIP contains (near-complete, coherent):
src/wireless_gateway_manager/brands.go: the SSID brand-prefix recognition set extracted to its own home —whitelabelCaptivePrefixassembled from non-contiguous bytes ("Net4" + "sats" + "-"), the same trick the gutter test itself uses, so upstream code never carries the re-brand name as a contiguous literal on scanned surfaces; recognition is case-insensitive (the pre-device-code lowercasetollgate-<code>fleet is still out there, per one-device-code.md "The measured drift").vendor_element_manager.go/discovery_log.go: comments de-literalized to "the whitelabel re-brand's prefix" — the literal's legitimate homes are named as the SSID contract checker and docs/.brands_test.goupdated with a pin against an independent construction of the assembled value, so an assembly typo fails a test instead of silently dropping a whole brand from upstream recognition — worth keeping whatever shape the final fix takes.tests/contract/check-ssid-format.sh+docs/architecture/one-device-code.md: reference updates pointing at brands.go.Diagnostic observations for the owning lane: the snapshot builds on
cb4db7f(pre-#714 portal pin); the WIP had no battery run —go test ./wireless_gateway_manager/...andtests/packaging/rebrand-literal-gutter_test.shwere not executed before the handoff commit. The untracked→addedbrands.goand the modifiedbrands_test.goare the load-bearing pieces; the comment edits are mechanical.docs(changelog): fragment for the OpenTollGate#722 brand-registry fix
Amperstrand commentedon Oct 7, 2026
Diagnostics from the fix-722 lane (PR incoming from
Amperstrand:fix/722-rebrand-gutter):The 7 surfaces, classified. 1 shipped packaging comment (
99-tollgate-setup:460), 1 runtime Go table (discovery_log.gotollGateBrandPrefixes), 2 Go comment lines (discovery_log.go,vendor_element_manager.go:177), 2 test fixtures (brands_test.go). Only the uci-defaults file ships to routers; the rest are source/test surfaces.Why the gutter went red after #692 had "fixed" it — merge order, not intent. #692 (
74f6782) was authored/verified on a branch cut before #682 (659de8a) merged, so its "suite: 6/0 green" never saw the 7 hits; landing it on a main that already carried #682 re-red the gate the same day. Neither PR is wrong in isolation — they collided.The product question, answered by the tree. The whitelabel is still a re-brandable separate build (gutter header: separate org/repo/webroot,
/etc/tollgate/brandbuild input; README branding paragraph). #682's recognition feature makes the brand recognizable as an upstream, not first-class in this repo. So the fix follows #722's first hint: the literal leaves the code surfaces — the prefix is assembled from non-contiguous bytes in a new single brand registry (src/wireless_gateway_manager/brands.go), the same trick the gutter file itself uses to carry its scan pattern; a value-pinning test re-assembles it independently so an assembly typo fails CI rather than silently dropping a brand from upstream recognition. The gutter's scope is untouched (#692's ruling stands).A trap worth recording: the issue's second FAIL is tree-state-dependent. The
uhttpd.lucivocabulary trip in the evidence does not reproduce on pristinemain— it comes from untracked feed-WIP (92-tollgate-admin-setup) sitting in the verifying checkout, because that check greps the filesystem directory, not the tracked tree. That's #723/#734's lane; on a clean checkout this PR's gutter run is 6/6.Verification on the branch: gutter 6/6,
check-version-syncpass,check-ssid-formatpass (behavioural + structural layers), module gofmt/vet/-racepass, full 16-modulego-batteryrun — with the two pre-existing/environmental failures called out in the PR body (#718's racy late-receive assertion;apk-install-resolutionneeds a docker socket this host lacks).fix(brands): one assembled brand registry for the whitelabel prefix —…
resolve: merge main into #1037 (#1074)