Work against dsh 0.1.2-rc.1 - #17
Merged
Merged
Conversation
npm semver never lets a prerelease satisfy a caret with a different version tuple, so a range naming only older rc lines matches no published dsh at all. dsh 0.1.2-rc.1 became npm `latest` on 2026-09-03 and installing this plugin beside it failed with ERESOLVE. Ranges now OR in each supported line, the way upstream bumps its own `^0.1.2-rc.1` pins. Proven: dsh 0.1.2-rc.1 plus all four dshworks plugins resolve to one hoisted 0.1.2-rc.1 tree, 528 packages, no ERESOLVE; the plugin imports and `dsh --profile web --patch cordis.patch.yml --dump-config` composes it.
dsh ships prereleases only, and npm semver never lets a prerelease satisfy a caret with a different version tuple. Every release moves the whole `@deepseek-ai/dsh-*` line outside any range already written, and no test here can see it: the break is in the range, not the code. `npm i` succeeding is the weaker question and it passes while broken. Installed alone, npm satisfies a stale `^0.1.0-rc.6` by HOISTING dsh-llm 0.1.0-rc.8 to the root and pushing dsh 0.1.2-rc.1's own copy into a nested node_modules: 691 packages instead of 528, zero warnings, and the plugin imports a four-release-old harness while the host imports the current one. Measured on the published @dshworks/dsh-watch 0.2.0: THIRTEEN harness packages split. So the daily check installs for real and asserts one version of every harness package — twice, because they fail separately: this tree (did we fix it?) and the published package (did the fix ship? a fix users cannot install is not a fix). Drift opens an issue; a red scheduled run notifies nobody. It reads red right now against every published plugin, and clears on publish.
The release check was red on all four PRs that fix the thing it checks. By construction: the published package is still the broken one until the PR merges and ships, so gating a PR on the published half makes the tripwire red on its own fix — which is how a tripwire gets switched off. A pull request now checks only what this branch would publish. The scheduled run keeps both halves, and it is the only run that can ever see the published one clear. docs/pricing.json carries the package version, so the 0.4.1 bump left it stale and the generated-artifact check caught it. Regenerated. The lesson is ordering: bump, then regenerate, then test — I tested before the bump.
lroolle
force-pushed
the
fix/dsh-0.1.2-rc.1
branch
from
September 4, 2026 05:53
01201da to
f77d6a3
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
dsh 0.1.2-rc.1 became npm
lateston 2026-09-03. This plugin does not workbeside it, in a way nothing here could have noticed.
1. The peer range excludes it
npm semver never lets a prerelease satisfy a caret with a different version
tuple, so
^0.1.0-rc.6matches 0.1.0-rc.8 and nothing after it — not0.1.1-rc.2, not 0.1.2-rc.1. That has been true since 2026-08-21.
The loud failure is
ERESOLVEwhen two of our plugins are installed together.The quiet one is what actually reaches users: installed alone, npm satisfies
the stale range by hoisting
@deepseek-ai/dsh-llm0.1.0-rc.8 to the rootand pushing dsh 0.1.2-rc.1's own copy into a nested
node_modules. 691packages instead of 528, zero warnings, thirteen harness packages resolved to
two different versions at once. The plugin imports a four-release-old harness;
the host imports the current one.
Ranges now OR in each supported line, the way upstream bumps its own
^0.1.2-rc.1pins.2. A daily check that fails loudly
scripts/check-dsh-release.mjsinstalls for real against dshlatestandasserts one version of every harness package — twice, because they fail
separately: this tree (did we fix it?) and the published package (did the fix
ship?). Drift opens an issue, because a red scheduled run notifies nobody.
It reads red against the currently published package and clears on publish.
Proof
packages, no ERESOLVE
dsh --profile web --patch cordis.patch.yml --dump-configcomposes each one