Skip to content

fix(deps): development cannot npm install - #596

Merged
rubenvdlinde merged 2 commits into
developmentfrom
fix/development-cannot-npm-install
Aug 27, 2026
Merged

fix(deps): development cannot npm install#596
rubenvdlinde merged 2 commits into
developmentfrom
fix/development-cannot-npm-install

Conversation

@rubenvdlinde

Copy link
Copy Markdown
Contributor

npm install fails on development today

A bare install — no --legacy-peer-deps — is an ERESOLVE. The quality suites pass because they do pass that flag, which accepts a tree only npm-with-an-existing-lock can produce, never one a fresh clone can.

npm reports one conflict at a time, so this had to be peeled layer by layer. Every layer is the same shape: a bump landed past what a toolchain package's peer range allows.

package was now
@babel/core ^8.0.1 ^7.22.9
@babel/preset-env ^8.0.2 ^7.22.9
gridstack ^13.2.0 ^12.6.0
postcss-html ^2.0.0 ^1.0.0
stylelint-config-html ^2.0.0 ^1.1.0
stylelint-config-recommended-vue ^2.0.0 ^1.6.1
webpack-cli ^7.2.2 ^6.0.1

The babel pair is irreconcilable, not merely stale

@babel/preset-env@8 requires @babel/core ^8.0.0; @nextcloud/webpack-vue-config@7.0.4 requires ^7.22.9. No version satisfies both. A loop that keeps applying whichever demand npm reported last oscillates between the two forever — it is a deadlock, not a chain.

So both babel pins revert together. Moving one alone is what produced the deadlock, and it is why the open babel-8 dependabot PRs cannot land individually.

gridstack

Pinned to ^12.6.0 so development installs now. ConductionNL/nextcloud-vue#801 widens that peer to accept 13 — measured, not assumed: identical build and 6752 passing tests on both majors — and dependabot can raise it again once 2.20.0 publishes.

Verification

Every command run without --legacy-peer-deps:

before after
npm install ERESOLVE rc=0
npm run build rc=0
npm run lint rc=0
npm run stylelint rc=0

A bare `npm install` on development fails with ERESOLVE. The quality
suites pass because they install with --legacy-peer-deps, which accepts a
tree only npm-with-an-existing-lock can produce, never one a fresh clone
can.

npm reports one conflict at a time, so this had to be peeled layer by
layer. Every layer is the same shape: a bump landed past what a toolchain
package's peer range allows.

  @babel/core                        ^8.0.1 -> ^7.22.9
  @babel/preset-env                  ^8.0.2 -> ^7.22.9
  gridstack                          ^13.2.0 -> ^12.6.0
  postcss-html                       ^2.0.0 -> ^1.0.0
  stylelint-config-html              ^2.0.0 -> ^1.1.0
  stylelint-config-recommended-vue   ^2.0.0 -> ^1.6.1
  webpack-cli                        ^7.2.2 -> ^6.0.1

@babel/preset-env 8 requires @babel/core ^8.0.0 while
@nextcloud/webpack-vue-config@7.0.4 requires ^7.22.9 -- no version
satisfies both, so the babel pins revert together. Moving one alone is
what produced the deadlock, and is why the babel-8 dependabot PRs cannot
land individually.

gridstack is pinned to ^12.6.0 so development installs now.
ConductionNL/nextcloud-vue#801 widens that peer to accept 13 (measured:
identical build and 6752 passing tests on both majors); dependabot can
raise it again once 2.20.0 publishes.

Verified with no --legacy-peer-deps: npm install rc=0 (was ERESOLVE),
build rc=0, lint rc=0, stylelint rc=0.
The previous commit deleted package-lock.json before installing. That
turns a five-package pin into a full re-resolution: on pipelinq it moved
172 package versions, added 64 and removed 132, when five were intended.

One of those unintended moves broke boot. dexie went 4.4.4 -> 4.4.5, and
@conduction/nextcloud-vue's published dist BUNDLES its own dexie copy, so
the app loaded two and Dexie throws at module load:

  pageerror: Two different versions of Dexie loaded in the same app:
             4.4.5 and 4.4.4

The E2E boot gate caught it -- "the bundle loaded but rendered nothing" --
while build, lint, stylelint and unit tests were all green. A passing
build says nothing about whether the app mounts.

Starting from development's lock and letting npm move only what the
manifest forces cuts the change to 85/17/46 and leaves dexie alone.

Control: development's own E2E run is 309 passed / 1 failed with no
dexie pageerror and no boot-gate failure, so the breakage was mine.
@github-actions

Copy link
Copy Markdown
Contributor

Quality Report — ConductionNL/hermiq @ d522897

Check PHP Vue Security License Tests
lint
phpcs
phpmd
psalm
phpstan
phpmetrics
eslint
stylelint
build
check-specs
check-manifest
test-l10n
format
check-l10n-js
check-schema-l10n
composer ✅ 123/123
npm ✅ 739/739
app:check-code ⏭️
info.xml
REUSE
PHPUnit
Newman
Playwright
Hydra gates

Quality workflow — 2026-08-27 11:58 UTC

Download the full PDF report from the workflow artifacts.

@github-actions

Copy link
Copy Markdown
Contributor

Quality Report — ConductionNL/hermiq @ d82edb1

Check PHP Vue Security License Tests
lint
phpcs
phpmd
psalm
phpstan
phpmetrics
eslint
stylelint
build
check-specs
check-manifest
test-l10n
format
check-l10n-js
check-schema-l10n
composer ✅ 123/123
npm ✅ 742/742
app:check-code ⏭️
info.xml
REUSE
PHPUnit
Newman
Playwright
Hydra gates

Quality workflow — 2026-08-27 13:28 UTC

Download the full PDF report from the workflow artifacts.

@rubenvdlinde
rubenvdlinde merged commit ea9491a into development Aug 27, 2026
86 checks passed
rubenvdlinde added a commit that referenced this pull request Aug 27, 2026
Rebuilt on current development rather than merged, because #596
("development cannot npm install") landed while the first attempt was in
CI and rewrote both package.json and package-lock.json. Resolving a
lockfile conflict by hand produces a tree neither side generated; a
fresh install from the fixed baseline produces one npm did.

#596 also explains the E2E failure the earlier attempt hit. I had been
weighing flake against a 2.17->2.19 regression, and it was neither: the
install itself was broken on development, which is what #596 fixed.
Worth stating because the first attempt's evidence pointed away from a
regression already - no commit touched CnAiCompanion/CnAiInput.vue
between the two tags, and `cn-ai-input-dictation-error` was
byte-identical in both - but "not that" is not the same as knowing what.

#596 left the pin at ^2.17.0, so the bump is still needed.

Now ^2.20.0 rather than ^2.19.0: 2.20.0 published while this was in
flight and is the current latest. It carries the named index sources
(`entitySource`) and the `flow` page type, which is what lets hermiq
drop its custom flow pages in a follow-up.

Control build against the regenerated lock: webpack compiles, 26
artifacts.

Co-authored-by: Conduction Release Bot <release-bot@conduction.nl>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant