Skip to content

Commit 488f9f2

Browse files
Fix pnpm trust warning for per-member locks (#1268) (#1781)
* Start fix for #1268 Assisted-by: Claude Code:claude-opus-5-5 * Warn about trustLockfile after member-lock rollback In a pnpm workspace with sharedWorkspaceLockfile: false, hosted mode pins each member's own pnpm-lock.yaml and adds trustLockfile: true to the root pnpm-workspace.yaml. Rollback and remove only looked at the root lock when deciding whether that setting is still needed, so: - restoring a member lock never warned pnpm_trust_lockfile_left, silently leaving pnpm's lockfile tarball check turned off; - restoring the root lock while a member lock stayed hosted reported the setting as unneeded. The cleanup now treats the root lock and every member lock of the workspace as the locks the setting trusts. Fixes #1268 Assisted-by: Claude Code:claude-opus-5-5 * Keep trust cleanup safe for unlisted members Review follow-ups for the member-lock trust cleanup: - when the workspace's member locks cannot be listed, leave the trust setting alone instead of judging it by the root lock only, since a member lock may still be hosted; - only list member locks when the restore actually put a pnpm lock back, so rollbacks of other ecosystems don't walk the workspace; - reuse the new registry mock in the existing pnpm rollback helper. Assisted-by: Claude Code:claude-opus-5-5 --------- Co-authored-by: Claude <noreply@anthropic.com>
1 parent a8a10dc commit 488f9f2

3 files changed

Lines changed: 265 additions & 15 deletions

File tree

‎crates/socket-patch-cli/CLI_CONTRACT.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1185,7 +1185,7 @@ v5.0 replaces v4's per-purl reverts and whole-ledger reverse replay (`revert_rem
11851185

11861186
* **Scope.** The hosted pins are what lockfile discovery finds — `(purl, patch uuid, files wiring it)`, recognized only on `https://patch.socket.dev` or the `--patch-server-url` / `SOCKET_PATCH_SERVER_URL` origin. A scoped rollback (paths / identifiers / `--ecosystems`) restores exactly the pins in scope; each pin restores or refuses on its own (there is no whole-ledger replay, and a pre-v5 ledger's edits are never replayed). A pin discovery cannot see is out of reach: a lockless cargo `registry = "socket-patch-<uuid>"` pin, a nuget exact-id mapping with no `packages.lock.json`, a gem wired only in the `Gemfile` (pre-bundler-2.6 mixed state) — restore those files from version control.
11871187
* **What a restore does.** Every file wiring the pin is rewritten back to the DEFAULT UPSTREAM registry entry for `name@version`, re-resolving whatever the entry pins (tarball URL, integrity, checksum, hashes) from the public registry; only the hosted entries change and every other byte stays the file's own. A pin is **all-or-nothing**: refused in one of its files, it is restored in none of them, so no pin is left half hosted. Nothing reaches disk until every pin has resolved, and `--dry-run` resolves exactly like a wet run — registry lookups included — and skips only the write. Per format:
1188-
* **npm family** — `package-lock.json` / `npm-shrinkwrap.json`, `yarn.lock` (classic and berry), `pnpm-lock.yaml` / `shrinkwrap.yaml`, `bun.lock`: resolution + integrity (+ shasum where recorded) from the npm registry's version document (`SOCKET_NPM_REGISTRY`); a yarn berry lock whose settings name another registry reads that registry's document instead — the registry yarn itself resolves the package against (#1017): a scoped package's `npmScopes.<scope>.npmRegistryServer`, else `YARN_NPM_REGISTRY_SERVER`, else `npmRegistryServer`, each from the closest `.yarnrc.yml` (`YARN_RC_FILENAME`) that sets it among the lock's directory, every parent directory and then the home directory, with a `${VAR:-default}` / `${VAR-default}` reference to an unset variable read as its default, as yarn reads it (a reference to a variable that is set is not expanded — the rc file must not choose which of the process's variables lands in a requested URL — and, like a reference yarn cannot expand or an `npmScopes` written as a flow mapping, falls back to the default registry with `upstream_registry_fallback`), so a mirror's off-path `dist.tarball` keeps its `::__archiveUrl=` binding (a berry entry the hosted pin keyed by its tarball URL also takes the version document's `bin` back, in place of the served tarball's own spelling the pin wrote, as yarn writes the `npm:` entry, #1131), and a pnpm lock whose sibling settings name a registry reads that registry's document — `.npmrc` `registry` (or, for a scoped name, `@scope:registry`); on pnpm 10 a pnpm-workspace.yaml `registries` map instead when present; on pnpm 11+ (or an unknown major) pnpm-workspace.yaml `registries."@scope"` / `registry` / `registries.default` too, ahead of the matching `.npmrc` key — so a mirror's `tarball:` comes back as pnpm recorded it and a URL conventional under that registry stays derived (falling back to the default registry, with `upstream_registry_fallback`, when the mirror can't be read; a value holding an unexpanded `${VAR}` is read as unset). pnpm 12.0–12.12 (named by the same install record, package.json pin or rush.json `pnpmVersion` as tier 2 below) compare a scoped package's URL with the default registry's rather than its scope registry's, so they record `tarball:` for every package from a scope registry other than the default one, and the restore gives it back, both in the entry and when reading the lock's evidence below (#1619); pnpm 12.13+ derive it from the scope registry like pnpm <= 11. Whether a restored pnpm entry gets its `tarball:` back follows `lockfileIncludeTarballUrl` as the pnpm that wrote the lock read it (#902), from the strongest evidence available: (1) the lock's own unpinned registry resolutions (a bare one proves it off; a URL pnpm could have derived proves it on; pnpm 11+'s env lockfile document does not count); else (2) the settings file the installed pnpm major reads (`node_modules/.modules.yaml` `packageManager`, else package.json `packageManager`; a pre-9 lock or shrinkwrap means pnpm <= 8): `.npmrc` `lockfile-include-tarball-url` on pnpm <= 9, pnpm-workspace.yaml `lockfileIncludeTarballUrl` on pnpm >= 11 (also assumed for a lock carrying an env lockfile document), the workspace file then `.npmrc` on pnpm 10; else (3) pnpm 10's reading. A Rush lock (`common/config/rush/pnpm-lock.yaml` or a subspace lock, with `rush.json` at the Rush root) takes its pnpm major from rush.json `pnpmVersion` instead of tier 2's install record and package.json pin. A 9.0 lock may come from pnpm 9, 10 or 11+, so when tier 3's reading differs from pnpm 9's (`.npmrc` only) or pnpm >= 11's (pnpm-workspace.yaml only) — e.g. `.npmrc` on with the workspace file silent, or the workspace file setting it with `.npmrc` silent or disagreeing — the restore follows pnpm 10 but warns `upstream_pnpm_tarball_setting_guessed` (once per lock, naming the entries); a URL pnpm records anyway (not derivable from the registry) never warns. A `bun.lock` 4-tuple's registry slot is rebuilt the way Bun writes it (#992): `""` for a package from registry.npmjs.org, otherwise the full tarball URL — Bun 1.1.39–1.3.6 read `""` as npmjs whatever the project configures. The registry is the one Bun resolves the package against: a scope's `.npmrc` `@scope:registry` or `bunfig.toml` `[install.scopes]` entry, else `BUN_CONFIG_REGISTRY` / `NPM_CONFIG_REGISTRY`, then the `.npmrc` `registry` or `bunfig.toml` `[install] registry`. Bun reads these from the files beside the lock and from the user's own (#1276): `$XDG_CONFIG_HOME/.npmrc` when it exists, else `~/.npmrc` (never `NPM_CONFIG_USERCONFIG`), and the global bunfig `$XDG_CONFIG_HOME/.bunfig.toml` when `XDG_CONFIG_HOME` is set, else `~/.bunfig.toml`. A key set beside the lock wins over the user's own file of the same kind. Between the two kinds, Bun ≤ 1.3 takes any `.npmrc` over any bunfig and Bun ≥ 1.4 any bunfig over any `.npmrc`. A lockfileVersion-2 `bun.lock` is Bun ≥ 1.4's (Bun 1.3 ignores it); when the kinds disagree on a package of a lockfileVersion-0/1 `bun.lock` (which Bun 1.4 keeps as is) or a `bun.lockb`, the restore refuses that pin rather than guess. Only the conventional token variables (`NPM_TOKEN`, `NODE_AUTH_TOKEN`, `BUN_AUTH_TOKEN`) expand in the project's files; the user's own files expand any variable, as Bun does. Its version document's `dist.tarball` fills the slot, and when it can't be read (`upstream_registry_fallback`) the default registry's conventional URL is re-based on it. The `bun.lockb` takeover restore records the same URL. Side settings: a project `.npmrc` that is exactly `allow-remote=all\n` is deleted once no root npm lock entry is hosted, otherwise a remaining top-level `allow-remote=all` warns `npm_allow_remote_left`; a `pnpm-workspace.yaml` that is exactly the scaffold hosted mode creates is deleted once `pnpm-lock.yaml` is no longer hosted (a keyless file hosted mode spliced the scaffold into gets the spliced lines removed, #1096), otherwise a remaining `trustLockfile: true` warns `pnpm_trust_lockfile_left`. **`bun.lockb` (binary)**: `rollback` and `remove` refuse it (the checkout remedy). The hosted → vendored takeover and the eject DO restore it, since the vendor ledger then records the rebuilt record as its pre-vendor original: the native codec turns each hosted remote-tarball record back into Bun's npm registry record for `name@version` (the registry's `dist.tarball` + `dist.integrity`, the package metadata hash re-derived, the hosted URL string dropped from the string pool). The hosted rewrite keeps the registry record's inactive bytes (padding, semver) in the tarball record, so a lock it wrote comes back byte for byte — early writers' uninitialized padding included; a record without them (an older socket-patch or a Bun re-save) is rebuilt the way Bun writes one, and refused for a prerelease/build version. A lock the hosted rewrite had to normalize is marked in the root package's resolution value bytes (which no Bun reader reads): a binary format 1 lock it promoted to format 2 is demoted back to its exact format-1 bytes (verified by promoting it again, otherwise refused), and a lock whose workspace dependency behaviors it normalized is refused with the `git checkout -- bun.lockb` remedy.
1188+
* **npm family** — `package-lock.json` / `npm-shrinkwrap.json`, `yarn.lock` (classic and berry), `pnpm-lock.yaml` / `shrinkwrap.yaml`, `bun.lock`: resolution + integrity (+ shasum where recorded) from the npm registry's version document (`SOCKET_NPM_REGISTRY`); a yarn berry lock whose settings name another registry reads that registry's document instead — the registry yarn itself resolves the package against (#1017): a scoped package's `npmScopes.<scope>.npmRegistryServer`, else `YARN_NPM_REGISTRY_SERVER`, else `npmRegistryServer`, each from the closest `.yarnrc.yml` (`YARN_RC_FILENAME`) that sets it among the lock's directory, every parent directory and then the home directory, with a `${VAR:-default}` / `${VAR-default}` reference to an unset variable read as its default, as yarn reads it (a reference to a variable that is set is not expanded — the rc file must not choose which of the process's variables lands in a requested URL — and, like a reference yarn cannot expand or an `npmScopes` written as a flow mapping, falls back to the default registry with `upstream_registry_fallback`), so a mirror's off-path `dist.tarball` keeps its `::__archiveUrl=` binding (a berry entry the hosted pin keyed by its tarball URL also takes the version document's `bin` back, in place of the served tarball's own spelling the pin wrote, as yarn writes the `npm:` entry, #1131), and a pnpm lock whose sibling settings name a registry reads that registry's document — `.npmrc` `registry` (or, for a scoped name, `@scope:registry`); on pnpm 10 a pnpm-workspace.yaml `registries` map instead when present; on pnpm 11+ (or an unknown major) pnpm-workspace.yaml `registries."@scope"` / `registry` / `registries.default` too, ahead of the matching `.npmrc` key — so a mirror's `tarball:` comes back as pnpm recorded it and a URL conventional under that registry stays derived (falling back to the default registry, with `upstream_registry_fallback`, when the mirror can't be read; a value holding an unexpanded `${VAR}` is read as unset). pnpm 12.0–12.12 (named by the same install record, package.json pin or rush.json `pnpmVersion` as tier 2 below) compare a scoped package's URL with the default registry's rather than its scope registry's, so they record `tarball:` for every package from a scope registry other than the default one, and the restore gives it back, both in the entry and when reading the lock's evidence below (#1619); pnpm 12.13+ derive it from the scope registry like pnpm <= 11. Whether a restored pnpm entry gets its `tarball:` back follows `lockfileIncludeTarballUrl` as the pnpm that wrote the lock read it (#902), from the strongest evidence available: (1) the lock's own unpinned registry resolutions (a bare one proves it off; a URL pnpm could have derived proves it on; pnpm 11+'s env lockfile document does not count); else (2) the settings file the installed pnpm major reads (`node_modules/.modules.yaml` `packageManager`, else package.json `packageManager`; a pre-9 lock or shrinkwrap means pnpm <= 8): `.npmrc` `lockfile-include-tarball-url` on pnpm <= 9, pnpm-workspace.yaml `lockfileIncludeTarballUrl` on pnpm >= 11 (also assumed for a lock carrying an env lockfile document), the workspace file then `.npmrc` on pnpm 10; else (3) pnpm 10's reading. A Rush lock (`common/config/rush/pnpm-lock.yaml` or a subspace lock, with `rush.json` at the Rush root) takes its pnpm major from rush.json `pnpmVersion` instead of tier 2's install record and package.json pin. A 9.0 lock may come from pnpm 9, 10 or 11+, so when tier 3's reading differs from pnpm 9's (`.npmrc` only) or pnpm >= 11's (pnpm-workspace.yaml only) — e.g. `.npmrc` on with the workspace file silent, or the workspace file setting it with `.npmrc` silent or disagreeing — the restore follows pnpm 10 but warns `upstream_pnpm_tarball_setting_guessed` (once per lock, naming the entries); a URL pnpm records anyway (not derivable from the registry) never warns. A `bun.lock` 4-tuple's registry slot is rebuilt the way Bun writes it (#992): `""` for a package from registry.npmjs.org, otherwise the full tarball URL — Bun 1.1.39–1.3.6 read `""` as npmjs whatever the project configures. The registry is the one Bun resolves the package against: a scope's `.npmrc` `@scope:registry` or `bunfig.toml` `[install.scopes]` entry, else `BUN_CONFIG_REGISTRY` / `NPM_CONFIG_REGISTRY`, then the `.npmrc` `registry` or `bunfig.toml` `[install] registry`. Bun reads these from the files beside the lock and from the user's own (#1276): `$XDG_CONFIG_HOME/.npmrc` when it exists, else `~/.npmrc` (never `NPM_CONFIG_USERCONFIG`), and the global bunfig `$XDG_CONFIG_HOME/.bunfig.toml` when `XDG_CONFIG_HOME` is set, else `~/.bunfig.toml`. A key set beside the lock wins over the user's own file of the same kind. Between the two kinds, Bun ≤ 1.3 takes any `.npmrc` over any bunfig and Bun ≥ 1.4 any bunfig over any `.npmrc`. A lockfileVersion-2 `bun.lock` is Bun ≥ 1.4's (Bun 1.3 ignores it); when the kinds disagree on a package of a lockfileVersion-0/1 `bun.lock` (which Bun 1.4 keeps as is) or a `bun.lockb`, the restore refuses that pin rather than guess. Only the conventional token variables (`NPM_TOKEN`, `NODE_AUTH_TOKEN`, `BUN_AUTH_TOKEN`) expand in the project's files; the user's own files expand any variable, as Bun does. Its version document's `dist.tarball` fills the slot, and when it can't be read (`upstream_registry_fallback`) the default registry's conventional URL is re-based on it. The `bun.lockb` takeover restore records the same URL. Side settings: a project `.npmrc` that is exactly `allow-remote=all\n` is deleted once no root npm lock entry is hosted, otherwise a remaining top-level `allow-remote=all` warns `npm_allow_remote_left`; a `pnpm-workspace.yaml` that is exactly the scaffold hosted mode creates is deleted once neither `pnpm-lock.yaml` nor, under `sharedWorkspaceLockfile: false`, any member's own `pnpm-lock.yaml` is hosted any more (#1268; a keyless file hosted mode spliced the scaffold into gets the spliced lines removed, #1096), otherwise a remaining `trustLockfile: true` warns `pnpm_trust_lockfile_left`. **`bun.lockb` (binary)**: `rollback` and `remove` refuse it (the checkout remedy). The hosted → vendored takeover and the eject DO restore it, since the vendor ledger then records the rebuilt record as its pre-vendor original: the native codec turns each hosted remote-tarball record back into Bun's npm registry record for `name@version` (the registry's `dist.tarball` + `dist.integrity`, the package metadata hash re-derived, the hosted URL string dropped from the string pool). The hosted rewrite keeps the registry record's inactive bytes (padding, semver) in the tarball record, so a lock it wrote comes back byte for byte — early writers' uninitialized padding included; a record without them (an older socket-patch or a Bun re-save) is rebuilt the way Bun writes one, and refused for a prerelease/build version. A lock the hosted rewrite had to normalize is marked in the root package's resolution value bytes (which no Bun reader reads): a binary format 1 lock it promoted to format 2 is demoted back to its exact format-1 bytes (verified by promoting it again, otherwise refused), and a lock whose workspace dependency behaviors it normalized is refused with the `git checkout -- bun.lockb` remedy.
11891189
* **vlt** — `vlt-lock.json`: slot [2] from the registry's `dist.integrity`, slot [3] per the lock's own convention (see the vlt hosted-mode contract); every hosted instance of the pin together.
11901190
* **cargo** — `Cargo.lock` back on crates.io (source + the sparse index's checksum, `SOCKET_CRATES_INDEX`). When the lock already holds a crates.io block for the same `name@version` (a later dependent locked its own copy, #679), the Socket block merges into it instead: it is dropped (with a v1 lock's `[metadata]` line), and dependents' references take the spelling cargo writes, so no duplicate block is left that cargo cannot parse (#863), and no index lookup is needed; every `Cargo.toml` declaration loses its `registry = "socket-patch-<uuid>"` pin (the shorthand the rewriter produced collapses back); every `[registries.socket-patch-<uuid>]` block no manifest or lock still references leaves the project cargo config — including a superseded patch generation's block an earlier re-pin left behind (#864). A declaration it cannot unpin refuses.
11911191
* **golang** — the hosted `replace` and the socket module's go.sum lines go; the upstream module's two go.sum lines come back, hashed from the module proxy (`SOCKET_GOPROXY`, else `GOPROXY` / `GONOPROXY` / `GOPRIVATE` as go reads them) and cross-checked against the checksum database (`SOCKET_GOSUMDB_URL`, else `sum.golang.org` unless `GOSUMDB=off` / `GONOSUMDB` / `GOPRIVATE` say go would not ask it). A `replace` the user had before the hosted run is not recorded anywhere, so the restore lands on the plain upstream module.

0 commit comments

Comments
 (0)