archetech / archon / 34260439067

87%

push

github

fix(ci): Seed the datachannel prebuild the repo actually installs (#1081)

* fix(ci): Seed the datachannel prebuild the repo actually installs

The prefetch list named @ipshipyard/node-datachannel, which reached the
tree through helia -> @libp2p/webrtc and left with them. The apparatus
looked dead, but the package it defends against is still here: filecoin-
pin brings node-datachannel 0.32.3 to the Filecoin wallet, and its
install script is `prebuild-install -r napi || build from source` --
exactly the unretried download that fails an image build when GitHub
blinks.

Upstream publishes the same asset shape at murat-dogan/node-datachannel,
so the target is one line and the env var loses its scope: prebuild-
install derives npm_config_node_datachannel_local_prebuilds from the
package name, so all 24 images that seed follow.

@bitgo/secp256k1 is not a candidate: it is pure JS, and the native
secp256k1 resolves through node-gyp-build, which reads prebuilds shipped
inside the package rather than downloading any.

Closes #1080

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GYGMfX3foBcVqP7AQNXjWj

* fix(ci): Seed prebuilds only where a seeded package is installed

All 24 images copied ./prebuilds and set the env var. Each runs npm ci
at the root and again in the one service it copies, and node-datachannel
is in one service lockfile, so 23 of them copied binaries nothing would
read and advertised a native dependency their service does not have.

The guard required that uniformity, so it moves to the invariant worth
holding: an image seeds if and only if the service it installs pulls a
seeded package. That is stricter than what it replaced -- it now fails
both an image that seeds without needing to and one that needs to
without seeding.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GYGMfX3foBcVqP7AQNXjWj

* fix(ci): Count the root lockfile when de... (continued)

3948 of 4536 branches covered (87.04%)

Branch coverage included in aggregate %.

8574 of 9000 relevant lines covered (95.27%)

725.42 hits per line