urnetwork/build v2026.8.22-1026185650

on GitHub

Changes in v2026.8.22-1026185650

v2026.8.21-1025763520 ... v2026.8.22-1026185650 -- 31 commits across 10 components.

apple

urnetwork/apple d2567e195...9a8a7b22d

• tests: the selectable transport order is H1-first, matching the SDK (6a4080625)
  TransportStatsTests asserted [.h3, .h1, .dns, .dnsPump] in four places that read their value from the SDK. The SDK's order is H1-first, and it is explicit about it:
• ci: build and test the apple app on every push (e5e66859b)
  This repo had no .github directory at all, so nothing verified that the iOS and macOS apps still compile or that the unit suite still passes. The release pipeline (build/all/run.sh) archives and signs; it never runs networkTests. This closes that gap.

extension

urnetwork/extension 8f75f07d8...660e05e39

• fix(types): replace the nine any casts and drop an unused listener param (fbc4ec63f)
  Pure typing/style debt, no behaviour change -- as any erases at compile time and the one runtime edit (dropping a trailing listener parameter) is invisible to the caller.
• fix(connect): stop writing the SDK callback refs during render (4e1c494c7)
  useConnectionManager kept the latest authNetworkClient/removeNetworkClient in refs with bare assignments in the hook body:
• ci: add timeout and concurrency; correct the lint comment (c793894d5)
  Two things the workflow was missing and two the comment got wrong.
• ci: report lint without gating build and test (0f912dec0)
  eslint . fails on a fresh checkout of main with 12 pre-existing errors (10 no-explicit-any, one unused binding, two react-hooks ref-during-render in elements/src), which stopped the run before the tests or either build ran at all.
• ci: build and test the extension on push (9f4d6fdf6)
  Adds the repo's first CI. Runs the checks the repo already defines -- eslint, the tsconfig.test.json type-check, and the vitest suite -- then builds both store targets the way the Makefile does (npm run build, npm run build:firefox) and asserts each produced a real zip.

build

urnetwork/build 4de28e9dd...8c422231f

• changelog: one directory, one file per audience, per release (7ee456401)
  The previous commit gave apple, windows and linux a note of their own but scattered them into fastlane-shaped paths that each release overwrote, and left the full changelog inlined in the release body. This puts all six files in one place and keeps every release's set.
• changelog: one release note per storefront, app changes first (785a706fe)
  There is no such thing as "the store note". Four storefronts serve four different artifacts and not one of the three numbers that matter -- the limit, the format, the field -- is shared between them:
• extension: bump the submodule past the lint fixes (57ed7d8c2)
  The extension leg of the CI has been red since it was added, and it is not the workflow's fault: this repo pinned extension at 8f75f07d8b, which predates urnetwork/extension#36. npm run lint fails there with 12 errors -- 9 no-explicit-any, an unused parameter, and two react-hooks/refs violations, one of which was a real correctness bug (refs assigned during render, feeding a ConnectionManager built once with [] deps that invokes the auth thunk from timers).
• changelog: say so when running anonymously degrades the store note (25f5f0d8e)
  The store note holds CI/test/docs-only commits back by looking up which files each candidate commit touched. That is one extra API call per candidate, and anonymous callers get 60 requests/hour -- so on any real span the lookups start returning 403. commit_files() then fails OPEN by design, because a missed lookup must never delete somebody's work from the release notes, which means the commit is KEPT.
• changelog: generate release notes from the commits instead of copying a placeholder (1e816e157)
  Every changelog this repo has published says the same thing. all/run.sh copied metadata/en-US/changelogs/pending.txt verbatim into metadata/en-US/changelogs/<version code>.txt on every release, pending.txt is a static one-liner, and so 1025763520, 1025613560, 1025339670 and 1025223880 are byte-identical: "- Bug and performance fixes."

sdk

urnetwork/sdk e3c36a9ba...3739dcfac

• Raise iOS extension SDK size limit to 60 MiB (3739dcfac)
• sdk: keep runtime/pprof out of the iOS builds (497d82552)
  The iOS extension SDK is over its size budget and cannot simply be raised past it:

warp

urnetwork/warp 678b25a8b...e13673c63

• ci: bound the job, cross-compile the shipped platforms, race the suite (682388b50)
  Follow-up on the same workflow. Five additions, all verified against a clean ubuntu:24.04 before pushing:
• ci: quarantine TestNginxLogsOmitClientAddr, which fails on ubuntu-latest (a2b0a0503)
  The first run of this workflow went red, and the cause was an assumption in the previous commit that CI just disproved: that warp's nginx tests would skip on a runner. One of them does not.
• ci: build and test warp on every push (afb3f80ef)
  warp had no CI. This adds the same build-and-test job the other urnetwork Go repos run, so a push or PR compiles every package and runs the real test suite rather than nothing.

server

urnetwork/server 66af1304a...2c9e92b45

• feat(model,api,taskworker): bootstrap the egress prober's credential without a human (a2731f132)
  A deployment that had never had an operator hand-mint a prober credential had no egress probing at all, and said nothing about it. The steps were: create a network, POST /network/auth-client, paste the by_client_jwt into the prober's environment. Miss them and every provider goes unprobed, which is indistinguishable from a fleet that is simply healthy -- the failure is silent in exactly the way that matters.
• feat(model,api): detect providers that go dark, separately from egress health (ed51f6d60)
  Egress health is the wrong instrument for this question. It samples ~131 destinations across several classes and takes minutes per provider, so it can only sweep the fleet slowly -- measured on beta, a provider went days between measurements. For that whole window a provider that has stopped carrying traffic is invisible: it stays connected, keeps accepting clients, keeps being advertised, and answers nothing. Making health cheaper is not the fix; the sampling breadth is what makes it worth ...
• test(model): pin the egress-health age bound that stopped advertising blackholes (8debc157b)
  The bound landed with the fix; nothing pinned it. That is the gap worth closing, because the behaviour it enforces is a NEGATIVE one -- a provider disappearing from the map -- and a bound that silently stops applying looks exactly like a fleet that happens to be healthy.
• feat(model): make the active bandwidth probe budget a deployment setting (d3e5a95c2)
  The hourly probe budget is CAPACITY, and capacity is a property of the deployment rather than of this code. A 40-provider fleet on a 4-core box and a production network carrying ~100k providers cannot share one compiled-in number: each has to measure its own limits and set its own. That is the opposite of a behavioural knob like the sampling rate, which stays one value everywhere precisely so every deployment exercises the same path.
• feat(model): name the prober as the fault when the whole fleet fails identically (b6aee48a2)
  A prober whose platform jwt had been rejected reported EVERY provider as no_consensus for eight hours. Every individual record was well-formed, every endpoint returned 200, the due queue kept handing out work and the prober kept taking it. Per provider the data was indistinguishable from a genuinely unreachable fleet, and nothing anywhere said "credential".

proxy

urnetwork/proxy bb5794710...3e94ca0f2

• ci: add a concurrency group and a vet step (52910aebc)
  Two house conventions the first commit left out.
• ci: build and test proxy on every push (f76ff29f4)
  proxy had no CI. This adds the same build-and-test job the other urnetwork Go repos run, so a push or PR compiles every package and runs the real test suite rather than nothing.

localizations

urnetwork/localizations e4715b9f2...d64c666e2

• ci: build and test the localization store on every push (bd3fb6fd5)
  This repo had no CI. Nothing validated keys/*.yaml until the release build ran npm run gen and wrote into the app repos, so a malformed key -- the common failure when a translation lands -- surfaced far from where it was introduced.

glog

urnetwork/glog 45662c0f2...93fe11b3e

• ci: record the TestRollover flake this workflow surfaced (ed159a35e)
  The first CI run on this branch went red on TestRollover, then a re-run of the same commit went green. Root cause, reproduced locally 50k+ times and independent of CI:
• ci: add concurrency group and timeout to the glog workflow (36e3f1128)
  The previous commit renamed test.yml to build-and-test.yml but landed only the rename; this carries the content change that was meant to go with it.
• ci: build and test glog on every push (1ee639974)
  glog had no CI of its own. This adds a build-and-test workflow so a push to master, a PR against master, or a manual dispatch compiles the package and runs the real test suite.
• ci: build and test glog on every push (ec233ad58)
  glog had no CI. This adds the same build-and-test job the other urnetwork Go repos run, so a push or PR compiles the package and runs the real test suite rather than nothing.

goidenticons

urnetwork/goidenticons 408ab24fc...325750b38

• ci: bound the run and vet, and name the file for what it does (4a65af535)
  Follow-up on the same branch, aligning this workflow with the newest house convention (server/test.yml, linux/build.yml) rather than the oldest (connect/test.yml, sdk/test.yml, which predate all three of these habits):
• ci: build and test goidenticons on every push (9293c1409)
  goidenticons had no CI. This adds the same build-and-test job the other urnetwork Go repos run, so a push or PR compiles the package and runs the real test suite rather than nothing.

Filtered out of this changelog: 18 merge commits (both sides are already listed), 1 version-stamp commit written by all/run.sh on every release. Re-run all/changelog.py --from v2026.8.21-1025763520 --to v2026.8.22-1026185650 --filters= to include them.