Skip to content

Sync publish with main for hvigor-plugin 0.1.5 - #38

Merged
Fiona2016 merged 10 commits into
publishfrom
main
Aug 28, 2026
Merged

Sync publish with main for hvigor-plugin 0.1.5#38
Fiona2016 merged 10 commits into
publishfrom
main

Conversation

@Fiona2016

Copy link
Copy Markdown
Collaborator

Brings main into the publish branch so the hvigor-plugin-v0.1.5 tag points at a commit the publish branch already contains, which the release workflow requires.

Includes the hvigor-plugin 0.1.5 changes (#37) and the npm publish DingTalk notification (#36).

Merge with a merge commit, not squash: squashing would rewrite the release commit and the tag validation would reject it.

The npm publish workflow had no notification of any kind, so 0.1.2, 0.1.3 and
0.1.4 all reached npm silently — the DingTalk steps only ever existed in the OHPM
SDK release. A build-time plugin ships less often than the SDK, which is exactly
why nobody notices a release without being told.

Mirrors the notification jobs in publish-ohpm.yml, including ignoreError so a
broken webhook cannot mark a release that already reached npm as failed. On a tag
the reported version is exact: the publish job already fails unless the tag
matches hvigor-plugin/package.json.

The DingTalk secrets are repository-level, so these jobs do not need the
Production environment the publish job runs in.
An unassigned FLASHCAT_BUILD_DIR= in a pipeline reaches the option as '', which
resolved to the module root. The sourcemap search walks recursively, so that
collects every product's sourceMaps.map and uploads an arbitrary one under the
current version — the same wrong-directory failure the product-aware default was
added to prevent, reached through a different door.

The warning for an empty key now names the apiKey option rather than only the
environment variable, and points at --no-daemon, which is the likeliest reason
the value arrived empty.
The demo wired symbol upload as fc-sdk-harmony-demo@0.1.0, but it reports RUM
under flashcat-harmony-demo (DemoConfig's default, and what demo_config.json
ships) at version 0.1.1 (AppScope versionName). ArkTS symbolication matches on
service and version, so both keys were wrong and the demo could never validate
symbolication end to end.

Also document that hvigor-plugin/dist must be built before opening the project:
entry/hvigorfile.ts imports it, hvigor evaluates that file on every invocation
including DevEco project sync, and dist is not checked in.
The gate ran the upload task and asserted nothing. With -p product=default the
probe's success and its fallback both resolve to build/default, and an upload
failure never fails the build, so the step stayed green even if reading the
product from the hvigor context was broken outright. It now requires the
"(product 'default')" suffix in the scan line, which only the probe path emits.
Removing the enabled option is a breaking change, and 0.1.3 documented the
dependency as ^0.1.3 — a range that picks up 0.1.4 automatically. Shipping it as
a patch would have changed behaviour under everyone already on 0.1.x without
them asking for it. 0.2.0 leaves that range where it is and makes the upgrade a
deliberate step.
Removing it assumed that naming the task on the command line is an equivalent
switch. It is not: in a pipeline, flipping a variable and editing the build
command are not the same cost, and the second may need review. Consumers who
want the upload off for one run reach for a variable, and hand-roll this gate
when the plugin does not offer it.

The defect was never that the option existed — it was that `enabled: false`
returned without a word, which in a build log is indistinguishable from a
successful upload. It now logs why it skipped.
With the enabled option kept, nothing here is breaking: the public type surface
is a superset of 0.1.3 and every behaviour change is a fix to something that
could not have worked. A patch release is the honest label, and consumers on
^0.1.3 pick it up as they should.
…tify

Notify DingTalk when hvigor-plugin is published to npm
hvigor-plugin 0.1.5: fix task registration, build-dir resolution, and three review findings
@Fiona2016
Fiona2016 merged commit 28ffcd7 into publish Aug 28, 2026
6 checks passed
@Fiona2016
Fiona2016 deployed to Production August 28, 2026 02:58 — with GitHub Actions Active
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