Run against a local dev node, both specs failed for reasons in the specs
themselves:
- CreateAccount looked for an "Add account" button that is now "Account",
and that button only shows when accounts can be stored, which the spec
never turned on. The page has to be wrapped in KeyringCtxRoot and start
on its route. The button is awaited for up to 20s: the page draws once
the api and keyring are ready, which takes longer than the 1s default
on a busy machine.
- Bounties started outside its route and used a fixed title, so a node
that already held a bounty from an earlier run had two matches.
- Neither closed its connections, so the test process never exited.
Both stay skipped until the nightly run starts a dev node.
The nightly run was cancelled at its 120-minute limit every night until
it was disabled. The cause was the Elapsed clock, which kept the test
process alive and is fixed (a909c07): test:all now completes in about
80 seconds. A 20-minute limit leaves a wide margin and turns a future
hang into a failure within the hour rather than two runner-hours.
The three suites were skipped since the table rework. What changed under them:
- An account or contact spans three rows (main, total balance, details),
marked from isFirst; the helpers paired rows by two and filtered on a
row class the rows no longer carry. A collapsed details row is hidden,
so rows are read including hidden ones.
- Accounts render one table per account group; the page object reads the
rows of every group table in order, and the sorting fixture keeps its
three accounts in one group, since sorting applies within a group.
- The expand toggle is the row's expand column, not a row-toggle element.
- useAccountInfo reads ownership from the keyring context, which the test
page did not provide, so saving a name did nothing; the page now renders
the real KeyringCtxRoot. The mock api gains consts.system.ss58Prefix.
- Badge popups (react-tooltip v6) render only while hovered; the specs
hover the badge first. The multisig badge needs the isMultisig flag.
- Addresses are shown in full and shortened by css; the fixtures with
shortened addresses are gone.
- The total balance, transferable and locked cards are always shown, as
placeholders without accounts; the specs check that instead of their
absence.
- Checks that needed matchers the runner lacks (not.toHaveLength,
toHaveProperty, toThrow with a contained message) are written against
the DOM directly.
@pezkuwi/dev 0.86.6 passes --experimental-test-module-mocks through to
node and gives the browser env jsdom's own events, the DOM node
hierarchy and the layout observers. The test scripts pass the flag, and
the five integration specs that were excluded while they depended on
jest.mock (Accounts, Create, Contacts, Bounties, AccountSidebar) run
again.
The account, address, sidebar and bounties specs replaced hooks with
jest.mock, which the node:test runner does not have, so all five failed
to load and were excluded from yarn test. They now register mocks with
mock.module from node:test and import the page under test afterwards,
since Node only swaps a module that has not been loaded yet.
- test-support/mocks: mockPageHooks() replaces the hooks a Page reads.
An ES module mock replaces every export, so each mock lists them all
(useBestNumber's useBestNumberRelay included). The real useAccountInfo
is wrapped, typed now, with the fixture identity merged in a typed
helper.
- The page-specific mocks live with their page: useMultisigApprovals in
page-accounts/test, useAddresses in page-addresses/test.
- jest-dom matchers were never registered for this runner; specs and
page elements use assertHasClass, assertTextContent and assertVisible
from test-support, with the same meaning.
- Page's mock api gains apiIdentity, which useAccountInfo reads.
- The bounties voting tooltip renders hidden and fades in, so the spec
waits for it to become visible.
- jest/expect-expect counts the assert* helpers as assertions; the
disable comments that only existed for that are removed.
Create and Bounties run green. Accounts, Addresses and Sidebar load and
stay skipped as upstream left them (their FIXMEs are about the reworked
tables). Needs @pezkuwi/dev 0.86.6 for the module mocks flag and the
fuller browser env.
Each test wrapped its whole sequence in waitFor() and slept 1s inside
every step, so waitFor's 1s timeout always expired ("Timed out in
waitFor"). Steps are now awaited in order, "open" is findBy* and
"closed" a single-condition waitFor, with an explicit cleanup between
tests. Both now pass (4 and 3) and exit, so they leave the yarn test
exclusions.
tick() was called when the module loaded and re-armed itself every
100ms forever, on every page whether or not any Elapsed was shown. In a
test process it is the timer that never let Node exit: the Modal and
Popup specs finished in 3s and then hung until killed (the only live
timer left was Elapsed.tsx:32, traced by recording timer creation).
The first Elapsed to mount starts the clock; it stops on the next tick
after the last one unmounts.
`yarn test` excluded every ^page- and ^react- spec, and test:all (the
nightly) is disabled, so 21 of the 25 specs ran nowhere and had broken
unnoticed. The exclusions now name only the integration specs still
built on jest.mock (Accounts, Create, Contacts, Bounties, the account
sidebar) and the two that hang (Modal, Popup), plus the .slow. ones;
13 spec files run instead of 1 (137 tests). @pezkuwi/dev 0.86.5 brings
the dev-test fix toEqual(expect.arrayContaining(...)) needs.
- The static metadata in @pezkuwi/types-support predates the rename and
only knows Pallet* names, while the app (and the test-support creation
helpers) use PezpalletBountiesBountyStatus, PezpalletAllianceCid, ...
createAugmentedApi now also registers the @pezkuwi/types-augment
lookup definitions, and createLookupRegistry gives specs a plain
registry that knows them.
- calculateBountyBond: the description had been rewritten from
"Kusama" (24 bytes) to "Dicle" (23 bytes) with the expected bond left
for 24; now "Pezkuwi network UI Bounty" (25 bytes) and its value.
- page-alliance had no tsconfig.spec.json, so its spec was never
type-checked; added, and specs are excluded from its build config.
react-tooltip 4 brought uuid 7 (open advisory, fixed only in 11) and
had not been maintained since 2022. Tooltip now wraps the v6 component
(no more cast); triggers use data-tooltip-id instead of data-for and
data-tip (Icon, Badge, Tag, claims payload). v6 renders the content in
.react-tooltip-content-wrapper and draws the arrow as its own element
that inherits the background, so the wrapper's child selectors point
there and the ::after arrow colours for accounts-badge are gone.
The accounts-badge block in the global styles was never closed (as
upstream); it is closed now with the same nesting, so its link colours
still apply only inside badge tooltips.
Checked in a headless browser on Zagros, the "available on all
networks" account badge: with v4 the tooltip was one 1525px line
starting off-screen at x=-260; with v6 it is 300px wide (the max-width
already in the styles), wrapped, on screen, arrow on the badge, and in
the light accounts-badge style the styles define.
- api/constants: only PEZKUWI_GENESIS and ZAGROS_GENESIS remain. The
Zagros value was a placeholder from before its launch; it is now the
live chain's genesis (0x1b0b4727…2cd5, measured).
- Inflation, proposal thresholds and governance tracks keep the Pezkuwi
entries; the Dicle, Kulupu, Cere, Vara, Joystream, zkVerify, Dock,
Neatcoin and NFTMart ones go, as do the dicle tracks.
- getTeleportWeight was used nowhere: removed.
- page-society: the Dicle-only "design samples" button and its canary
drawing go, with their table column.
- page-assets: the next-id rule keyed on Polkadot's and Kusama's asset
hub genesis never matched a Pezkuwi chain; removed, no change there.
- ParaLink: the Subscan link went to scan.pezkuwichain.io, which does
not resolve; removed (coretime keeps the homepage link).
- typesBundle generator: no unused sort-keys directive when empty.
- 87 chain and all 185 node logos were referenced by nothing once the
foreign endpoints went; the images are removed and imgConvert no longer
generates a nodes set. Pezkuwi and Zagros logos remain.
- External links: none of the 14 services serves a Pezkuwi chain. Most
are third-party sites whose chain keys had been renamed by the rebrand
(on Zagros they linked to commonwealth, subsquare and others for
another chain's data); stats., statescan. and scan.pezkuwichain.io do
not resolve, explorer.pezkuwichain.io is a placeholder and treasury./
governance. redirect back to this app. externalLinks is empty until a
service indexes a Pezkuwi chain; their logos are removed too.
- Identity icons: the centrifuge, joystream, litentry, dicle and paseo
entries are removed.
typesBundle carried overrides for 140 chains inherited from upstream;
none of them is a Pezkuwi chain. The live Pezkuwi and Zagros runtimes
(pezkuwichain, asset-hub-pezkuwichain, people-pezkuwichain, zagros and
its asset hub, bridge hub, collectives, coretime and people) match no
entry and need none: they carry metadata v14+.
The spec files, the Cere chain bundle, the custom derives and the 32
npm packages they came from are removed, among them GPL-3.0-only ones
(pontem, moonbeam, subsocial, jamton) that were bundled into this
Apache-2.0 app. The spec/chain/derives mechanism stays, empty, for a
chain that ever needs one. typesBundle.ts is regenerated (15 lines).
yarn.lock loses 3250 lines, elliptic and the old @polkadot/api trees
with them.
The connection list showed Pezkuwi, Zagros and Dicle; Dicle was never
launched (none of its six RPC hosts resolves). The foreign endpoint
lists inherited from upstream (prodChains, testChains, Paseo, zkVerify)
were exported but used nowhere. Both are removed; the resulting list is
the previous one without the Dicle header and its six entries.
- yarn.lock refreshed within existing ranges for the open advisories.
- copy-webpack-plugin 11 -> 14 (serialize-javascript 7) and
terser-webpack-plugin 5.6: build:www output is identical except four
chunks where terser now emits `a ?? b` for `null == a ? b : a`.
- testcontainers (page-addresses, test-support) and devtron were never
imported: removed (undici, @fastify/busboy, highlight.js went with
them).
- The crypto-browserify webpack fallback was dead: build:www is
byte-identical without it, so the fallback and the dependency are gone
and a future `crypto` import fails the build instead of shipping a
polyfill.
- braces resolves to @pezkuwi/braces 3.0.4, which limits nesting depth
(CVE-2026-93687; no upstream release).
@logion/node-api (Apache-2.0) and @digitalnative/type-definitions (MIT)
were imported only for their type definitions, and each brought an old
@polkadot/api tree with it (util-crypto 6/7, elliptic, uuid). The
definitions are now in the spec files, with their source and license.
Checked by serialising each resulting bundle with key order and
undefined preserved: byte-identical to what the packages produced
(logion 27715 bytes, standard and opportunity 1778 each).
RegistryTypes does not model a struct's `_alias` map, which the registry
accepts; the merged latest logion types are cast at that one point.
recoverAddress used secp256k1/elliptic.js, which put elliptic (open
advisories, no fixed version) into the shipped bundle. secp256k1Recover
from @pezkuwi/util-crypto (@noble/curves, or wasm) with the 'keccak'
hash type returns the same expanded key without its 0x04 prefix.
The spec's signature vector had been broken by the rebrand: the signed
message was rewritten from "Kusama" to "Dicle", so recovery gave another
address and the test failed on the old code too (it only runs in
test:all). Restored; old and new code pass it, and changing one word of
the message makes it fail.
react-hooks 7 rules on as errors; the 458 existing findings (309 files,
React Compiler rules only) are recorded in eslint-suppressions.json, so
any new one fails lint and fixed ones must be pruned.
When an amount had more decimals than the selected unit allows, the
function set result to -1 and then overwrote it, so the conversion went
on with a negative exponent. BN computes 10 ** -n as 10 ** n: with 12
chain decimals, "1.1234567890" in milli became 13345678900 instead of
1123456789. Such input is now invalid.
The decimals were also cut to the chain's decimals, which silently
dropped digits of an asset with more decimals than the chain; the cut is
removed, and with it the api argument it was the only use of.
- dev ^0.86.4 at the root and in packages/apps; directives renamed by
pezkuwi-dev-migrate-eslint10.
- eslint --fix with the 0.86.4 config; one fix broke syntax
(typescript-eslint prefer-optional-chain turned
`(isStarted && onStart) && onStart()` into `(isStarted && onStart?.();`
in TxButton) and is corrected by hand.
- The remaining findings fixed rather than disabled: optional chains
(each suggestion reviewed), `cause` on rethrown errors, `if` instead of
`a && promise` in effects, parallel instead of sequential awaits in
Promise.all, Error rejections, no `delete` on arrays (the storage query
cache is a Map; an upload's file list no longer keeps a hole), no
`[object Object]` stringification (TFunction's second argument is text
or options), dead code removed (referenda's never-shown expand button,
react-query's unused CoreDescriptor).
The generator put `/* eslint-disable simple-import-sort/imports */` in
every generated image module, but only the two index files import
anything; eslint 10 reports the other 1533 as unused directives. It also
wrote import paths without the `.js` extension the committed index files
carry, so regenerating no longer reproduced them. Both fixed; the
generator now reproduces all 1535 committed files byte for byte.
- continue-on-error is removed; lint, test, build:code and build:i18n
each have to pass, with fail-fast off so each reports on its own
- after each step the tree must be unchanged: eslint --fix, the i18n
catalogue and typesBundle.ts are all rewritten by these steps
- the chmod step is dropped; the scripts are run through node
build:code regenerates this file from the chain specs; 7a4bbea edited
it by hand and removed the zeitgeist market and court types, so every
build rewrote it and the committed copy was never what the build used.
It is the generator's output again; lint, build and tests pass on it.
The resolutions held the whole tree on wasm-crypto 7.5.17, util-crypto
14.0.25, api 16.5.36, ui 3.17.2 and extension-dapp 0.62.20, the
releases where ed25519, secp256k1, derivation and bip39 were not
implemented in wasm-crypto. Every @pezkuwi pin moves to the release
that has them: common 14.0.26, wasm 7.5.20, api 16.5.37, ui 3.17.3 and
extension 0.62.27; @pezkuwi/dev ^0.85.12. After `yarn dedupe` each
@pezkuwi package resolves to a single version.
Every 6.x release is inside both open advisories, so no 6.x update closes
them: GHSA-wrjc-x8rr-h8h6 (open redirect via a backslash in <Link> and
useNavigate) and GHSA-337j-9hxr-rhxg (SSR hydration; not reachable here,
this is a client-only HashRouter app). 7.18 in library mode keeps every API
the app imports (Routes, Route, Link, NavLink, Navigate, useLocation,
useParams, HashRouter, MemoryRouter) and bundles @remix-run/router.
The one behaviour v7 changes, relative links under splat routes, does not
occur: no route path ends in '*', tab links are absolute (basePath) and the
top level picks the page from location.pathname.
Measured: 17 routes opened in Chromium against the live 6.x build and this
build give the same URL, active tab, tab count and tab-click target, with no
page errors. Tests 89/89, lint and type check green.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Both came with the upstream template: a pull request labelled '-auto' was
approved for an upstream maintainer's account and squash-merged once its
checks reported, through the upstream author's own actions. The label does
not exist here and neither workflow ever acted (2 runs each, both on other
labels), but the path was open: it merges without review, and the PR checks
it waited on run with continue-on-error.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- react-router 6.30.6 / @remix-run/router 1.23.4 (open redirect, XSS advisories)
- styled-components 6.5.3, which brings postcss 8.5.28
- query-string 9.5.1 for decode-uri-component 0.5.0; parse() is the only API
used and is unchanged
- in-range lockfile refresh
Bundle compared against the live build of 570586b: vendor chunks and the
loader change; the one page chunk that differs only renumbers the
query-string module. Tests 89/89, lint and type check green.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
push-main.yml now builds build:www on every push to main and, once the
production environment is approved, copies it to the web host and switches
to it. The host keeps releases side by side and serves a 'current' symlink,
so going live is one rename and rollback is a dispatch with an earlier sha.
.github/deploy/release.sh checks the switch through nginx for the public
name and restores the previous release if the new one is not served.
Ships env-config.js, which index.html loads: on a static host the request
fell through to index.html and the browser ran HTML as a script.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
None of it could run here: the repository has no Pages site, no Pinata or
Crust credentials and no npm token, and docker/build.sh pushed to the
upstream author's Docker Hub account. Remove the release scripts, the IPFS
uploader with @pinata/sdk (axios 0.21) and @crustio/crust-pin, and
public/ipfs, whose pin.json and redirect pointed at another project's IPFS
build from this site.
README: the hosted URL is apps.pezkuwichain.io; drop the IPFS and desktop
chapters and the pull of a Docker image that does not exist. .gitignore
listed '.log' where '*.log' was meant, which let start.log in.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Only the Electron renderer ever set isElectron to true; the web entry passed
false. With that package gone the flag is constant, so remove it from
ApiCtxRoot, the context types and its consumers, together with the
pezkuwi.js.org link base it selected. In-browser account storage and the
WebUSB ledger option now follow the settings alone, as they already did on
the web.
Runtime environment detection (getEnvironment, ExternalWarning) reads the
user agent and is left as is.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
sort-keys has rejected packages/react-api/src/Api.tsx since signedExtensions
was added after signer, so every lint run on main stopped there.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
packages/apps-electron was never released, and electron 28, electron-builder
and their tooling (got, tar, app-builder-lib, builder-util-runtime) were most
of the dependency advisories left in this repo. Drop the package, its scripts,
the build:electron CI step, its i18n namespace and the deps only it used
(concurrently included).
The shipped web bundle changes by one string: "apps-electron" leaves the
i18n namespace list.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
PezkuwiChain requires AuthorizeCall as the first signed extension in
all transactions. Without this extension properly configured, transactions
fail with "bad signature" errors.
Added signedExtensions configuration to:
- ApiPromise.create() in Api.tsx
- getApi() in StakingAsync.tsx
- Test API creation functions
This fix enables proper transaction signing for PezkuwiChain mainnet
and teyrchains (Asset Hub, People Chain).
- Fix Vec type casting in nominate, kick, and multisig operations
- Update all @pezkuwi packages to latest stable versions (API 16.5.36, Common 14.0.25)
- Regenerate logo SVG files
- Update PEZKUWI_GENESIS to correct hash (0x1fc56b6a5fcd50358707f69b2b0bb0c8ab1fea2bcfc5eba7279efbcafef642af)
- Change highlight color from pink/orange to Kurdistan green (#86e62a)
- Add critical CSS in index.html to fix initial theme load issue
- Fix metadata badge showing incorrectly on every page load
- Fix block time text rendering vertically in explorer
- Update SVG logos with correct branding colors
