[ConfigManager] Register Node Sections 4/4

[bdchatham]

Declares the keys at the top of the node's configuration file, with no table of their own: 11
keys. With this the binary declares 27 sections and 271 keys across both configuration files,
with no refused registration.

Why a schema rather than the node's own type

That type carries the root keys and every table, so declaring against it would declare each
table's keys a second time. The schema squashes the same base group that type squashes, so the
spellings still come from the node's own tags, and restates only the fields it holds beside that
group. Each of those is held against the type it came from by name, tag and type, with the
schema's own field count holding the set, so a further one appearing there fails rather than
going undeclared.

A root field carrying no mapstructure tag fails too. The reader falls back to the field's own
name, so such a field is a key an operator can write under that spelling which no section
declares — and the walk used to skip it while its doc said it fails.

What is left out, and why

home is the file's own location, which the command line carries. mode is the fact the file
states at the top under its own name; declaring a second spelling would let an operator write
one kind of node in the file and resolve another. A generated file does state the node kind, so
a reader handed a decoded file reports it beside an operator's typos, and exempting that one key
belongs to the reader rather than to this declaration.

Three more name settings the node removed: the address of an out-of-process application it no
longer runs, the transport to that application, and peer filtering through it. Nothing in the
tree reads any of them and the generated file writes none. Two state an affirmative value in
the node's own defaults, an address and a transport name, so declaring them put settings for a
transport the binary no longer has into the space a writer renders from.

One key that is not what a generated file carries

The node name. The command takes it as a required argument, so a generated file always carries
an operator's own, and the declaration answers with the hostname of whatever machine is asking.
It joins the peer seeds on the list of keys the generator fills, and the test checks what makes
each declaration wrong rather than comparing it to a value it cannot know.

Three checks that were measuring less than they claimed

The check on what varies by node kind matched keys by treating a section's name as a dotted
prefix, so it was blind to every key sitting at the root of the file — the claim that those
answer the same for every node kind was measured by nothing. It now reads the keys out of the
registry, which also stops it assuming a section's name and its prefix are the same string.

The pairing of a section to the structs carrying its deprecation markings was written by hand
and nothing held it against the section's own paths, so a row naming a wrong but existing struct
gave a marking set for a type the section does not declare and both checks reading it passed
measuring nothing. Confirmed against the unfixed file: pairing the peer-to-peer section with the
remote procedure call struct leaves the whole package green while twenty-one keys go unwatched.
Every path a section derives now has to name a field of a struct its row lists, and every listed
struct has to be named by one of those paths.

The root section was exempt from the key-count table, on the stated reason that a squashed group
counts as a single tagged field. That is false — the count walk recurses, giving sixteen, and
sixteen less five exclusions is the eleven the section registers — so the exemption cost the one
check that holds the registry's own derivation against an independent count, on the only section
that lacked it. The row is there now.

The collision check

The root-key collision check claimed to cover the whole registered set and could not: nothing
imports either registrar, so in that package's test binary the set was the ten sections it
registers itself. It moves to the package that links every registrar, with both registrars
linked there as named test imports, so a dropped import stops compiling rather than quietly
narrowing the union to one file.

Any reused name fails, including a cross-file pair that is harmless on its own. Telling the two
apart would need a section to state which file it belongs to, and it does not; rather than add
that for a case which has never arisen, the rule is the stronger one, that a name is a table's
or a root key's once across the whole key space.

Formatters, go vet and golangci-lint clean; race detector clean. Mutation-verified,
including against the unfixed file so each check is shown to have content: a root key made to
vary by node kind, a section paired with the wrong struct, a root field left untagged, and a
node root key pointed at a section name from the other file each fail with the reason that
names them.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 60.32%. Comparing base (4ae239d) to head (574d272).

Additional details and impacted files

@@                   Coverage Diff                    @@
##           plt-775-node-sections-3    #3995   +/-   ##
========================================================
  Coverage                    60.31%   60.32%           
========================================================
  Files                         2065     2065           
  Lines                       177011   177025   +14     
========================================================
+ Hits                        106773   106787   +14     
  Misses                       60473    60473           
  Partials                      9765     9765           

Flag	Coverage Δ
sei-chain-pr	54.05% <100.00%> (-45.95%)	⬇️
sei-db	69.80% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
config/tendermintbase/tendermintbase.go	100.00% <100.00%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[github-actions]

The latest Buf updates on your PR. Results from workflow Buf / buf (pull_request).

Build	Format	Lint	Breaking	Updated (UTC)
✅ passed	✅ passed	✅ passed	✅ passed	Aug 26, 2026, 11:54 PM

[cursor]

PR Summary

Medium Risk
Changes expand the declared configuration key space and default-resolution behavior for root settings; mistakes could mislead operators or config tooling, though impact is mostly metadata and tests rather than live node decoding.

Overview
Registers top-level node config.toml keys (section node_base) via a nodeRootSchema that squashes BaseConfig without re-declaring table keys, with explicit exclusions for home, mode, and removed ABCI-related settings. moniker joins filledByTheGenerator because init supplies it per operator, not from mode defaults.

Registration is refactored to declareSection / declareRootKeys with registeredHere so tests walk the same set the binary registers. Cross-file root-key vs table-name collision checking moves to cmd/seid with named imports of cosmosbase and tendermintbase, so dropped imports fail compile instead of shrinking coverage.

tendermintbase tests now read keys from the registry (including root keys), add markedIn / root schema and exclusion tests, and harden generatedFileCarries for the headerless root region and commented template lines.

^{Reviewed by Cursor Bugbot for commit 574d272. Bugbot is set up for automated code reviews on this repo. Configure here.}

[seidroid]

The new root-key section declares three deprecated, no-op node settings (proxy-app, abci, filter-peers) that the generated config.toml does not carry, contradicting both rootDefaults' own doc and the exclusion rule this file already applies to the removed consensus settings. The rest of the change is well-covered, though the new cross-package collision test cannot see any other package's sections in a package-local test binary.

Findings: 1 blocking | 3 non-blocking | 2 posted inline

Blockers

• None at the file/PR level.
• 1 blocking issue(s) flagged inline on specific lines.

Non-blocking

• [suggestion] Root keys sit outside the whatVariesByNodeKind safety net. ours() (tendermintbase_test.go:65) matches on name+".", and declaredSections() was not extended with RootSectionName, so TestWhatVariesByNodeKindIsTheRecordedSet skips all 14 root keys. rootDefaults goes through forMode(mode), so a future mode rule touching e.g. DBPath or LogLevel would start varying by node kind with nothing recording it — the exact failure that test exists to catch for the nine tables. Making ours() also return true for a key declared by a registered section with an empty prefix would close it.
• [suggestion] Excluding mode has a read-side consequence the PR description does not cover: mode is written at the top of every generated config.toml (sei-tendermint/config/toml.go:103), so once a caller feeds a real file into registry.Resolve, mode lands in Resolved.Unknown alongside genuine typos. Whatever consumes the registry will need an explicit exemption, not just the write-side mapping the description mentions.
• 1 suggestion(s)/nit(s) flagged inline on specific lines.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit ad5474c. Configure here.}

[bdchatham]

@seidroid review

[seidroid]

The root-key section for config.toml is well constructed and the two prior review findings are genuinely fixed, but the new cross-package collision test does the opposite of what its doc comment (and the reply on the previous thread) says: it hard-fails on a cross-file name match instead of reporting it, and it cannot name the file a section came from.

Findings: 1 blocking | 3 non-blocking | 4 posted inline

Blockers

• None at the file/PR level.
• 1 blocking issue(s) flagged inline on specific lines.

Non-blocking

• None at the file/PR level.
• 3 suggestion(s)/nit(s) flagged inline on specific lines.

[bdchatham]

@seidroid review

[seidroid]

The root-key section for config.toml is well constructed and all four unresolved findings from the previous review are genuinely fixed. Two non-blocking notes remain: the deprecation check covers only nine of the root section's eleven keys, and one doc line calls the new named imports "blank".

Findings: 0 blocking | 2 non-blocking | 2 posted inline

Blockers

• None at the file/PR level.

Non-blocking

• None at the file/PR level.
• 2 suggestion(s)/nit(s) flagged inline on specific lines.

Superseded: latest AI review found no blocking issues.

[bdchatham]

@seidroid review

[seidroid]

The root-key section itself is well built and both prior findings are genuinely fixed, but the change that was supposed to bring root keys under the node-kind check does not: declaredHere still walks declaredAgainst, so all 11 root keys remain unmeasured while the rewritten doc comment and the PR description say otherwise.

Findings: 1 blocking | 0 non-blocking | 1 posted inline

Blockers

• None at the file/PR level.
• 1 blocking issue(s) flagged inline on specific lines.

Non-blocking

• None at the file/PR level.

[bdchatham]

@seidroid review

[seidroid]

The root-key section for config.toml is well built and the previous review's blocker is genuinely fixed — declaredHere now walks registeredHere, so all eleven root keys enter the node-kind check. What remains are non-blocking notes about the checks' own docs: the stated reason for keeping the root section out of the key-count table is contradicted by taggedFields, declaredAgainst's doc no longer describes what reads it, and the new schema test skips an untagged root field.

Findings: 0 blocking | 5 non-blocking | 5 posted inline

Blockers

• None at the file/PR level.

Non-blocking

• None at the file/PR level.
• 5 suggestion(s)/nit(s) flagged inline on specific lines.

Superseded: latest AI review found no blocking issues.

[bdchatham]

Applied the godoc discipline from 3993 to what this branch adds, rather than waiting for it to be found here.

Removed: a comment attached to no declaration (the root-key exclusions had one sitting between the list it described and the next declaration's doc), three closing paragraphs that were commentary about the comment rather than its subject, a count that reads four on one branch and five on the next, and a paragraph narrating the reasoning an earlier revision used before this one changed it.

Kept, in the present tense: that an excluded path whose field stops being deprecated is a setting the node honours and the space refuses; that the state sync keys are a set an operator fills together with the scratch directory the one the template omits; that any reused name fails because no section states which file it belongs to; and that a section's markings come from a set of structs because the root schema carries none of its own.

[bdchatham]

@seidroid review

[seidroid]

The root-key section for the node's config.toml is well built — the schema derives its spellings from the node's own tags, the exclusions each carry a checked reason, and moving the cross-package collision check into the package that links both registrars makes its claim true. No blockers, but the five suggestions from the previous review are all still open at the current head, two of them independently re-found by Codex.

Findings: 0 blocking | 5 non-blocking | 5 posted inline

Blockers

• None at the file/PR level.

Non-blocking

• None at the file/PR level.
• 5 suggestion(s)/nit(s) flagged inline on specific lines.

[bdchatham]

Also closed a hole outside the five, found while checking them: the generated-file helper looked for a table header to decide whether it had measured anything, and that is true by construction for the keys above the first table. So the three root probes passed measuring nothing, which is the shape the earlier scoping fix removed for named tables and left open here. It counts written keys instead, which closes both cases with one rule.

Verified by mutation against the unfixed file in each case, so every check is shown to have content rather than only to pass. Race detector clean.

[bdchatham]

@seidroid review

[seidroid]

The root-key section for the node's config.toml is well built: the schema derives its spellings from the node's own tags, every exclusion carries a reason that is measured rather than asserted, and the struct-pairing check now fails on both a wrong row and a superfluous one. All five suggestions from the previous review are genuinely fixed at head, and I found nothing new worth blocking or flagging.

Findings: 0 blocking | 0 non-blocking | 0 posted inline

Blockers

• None at the file/PR level.

Non-blocking

• None at the file/PR level.