Update to Lerna v3 and switch to a 2-stage release process #1325
Conversation
docs/contributing/development.md
Outdated
| ## Releasing a new version | ||
|
|
||
| 1. Decide whether the new version should be a major/minor/patch/prerelease version bump. | ||
| 2. From the root of the Neutrino repository, run: `git checkout -b version-bump master` |
There was a problem hiding this comment.
Can we make this git checkout -b version-bump upstream/master to ensure we are using this from upstream or does that not matter? Also should we pair this with:
git fetch upstream --quiet && git checkout -b version-bump upstream/masterto ensure master is up to date?
docs/contributing/development.md
Outdated
|
|
||
| If bumping to a new major version, or incrementing the pre-release version, | ||
| you will want to manually increase the presets' version of `neutrino` in | ||
| `peerDependencies`. |
There was a problem hiding this comment.
Maybe we could also add a note here on how to make this easier via sed? This is the command I use on macOS:
sed -i '' -- 's/"neutrino": "^9.0.0-0"/"neutrino": "^9.0.0-rc.0"/g' packages/*/package.json(D.)There was a problem hiding this comment.
Yeah good idea. The example sed command in your comment didn't work for me, however this did:
sed -i 's/"neutrino": "^9.0.0-0"/"neutrino": "^9.0.0-rc.0"/g' packages/*/package.json
It might be worth testing that on your machine to check the OS X sed (which iirc is a little quirky) handles that form ok (or whether it needs an additional -e before the script command).
package.json
Outdated
| "release:prepare": "lerna version", | ||
| "release:publish": "lerna publish from-package", | ||
| "release:publish-next": "yarn release:publish --dist-tag next", | ||
| "release:ci": "lerna publish prerelease --registry http://localhost:4873 --no-git-reset --yes", |
There was a problem hiding this comment.
I don't see prerelease as a valid positional for the lerna publish command, am I missing something?
https://github.com/lerna/lerna/tree/master/commands/publish#positionals
There was a problem hiding this comment.
It also appears that SKIP_CHANGELOG isn't being respected:
https://travis-ci.org/neutrinojs/neutrino/jobs/506989820#L606
There was a problem hiding this comment.
I think updating the version script to the following should work:
[ -z $SKIP_CHANGELOG ] || yarn changelog --package && git add CHANGELOG.mdThere was a problem hiding this comment.
I don't see
prereleaseas a valid positional for thelerna publishcommand, am I missing something?
The lerna publish command is a superset of lerna version, so supports the options lerna version supports:
https://github.com/lerna/lerna/tree/master/commands/version#semver-bump
I'd originally wanted to have the yarn release:ci command follow the actual release process as closely as possible - ie: make it do something like:
yarn release:prepare prerelease --yes && yarn release:publish --registry http://localhost:4873 --yes
However it turns out when lerna publish is run using either of the from-git or from-package options, lerna insists that the working directory is clean, which it won't be in CI, since we deliberately miss out the Commit the changes using: `git commit -a -m 'vN.N.N' step that is performed during a real release.
As such, I avoided this by making yarn release:ci perform the publish in one step, by running lerna publish in "do all steps" mode (ie: by specifying a version bump, rather than from-git or from-package).
It also appears that
SKIP_CHANGELOGisn't being respected:
Yeah I saw that locally on Windows too. Yarn must not propagate the modified env for sub-commands, strange.
package.json
Outdated
| "release:prepare": "lerna version", | ||
| "release:publish": "lerna publish from-package", | ||
| "release:publish-next": "yarn release:publish --dist-tag next", | ||
| "release:ci": "lerna publish prerelease --registry http://localhost:4873 --no-git-reset --yes", |
There was a problem hiding this comment.
| "release:ci": "lerna publish prerelease --registry http://localhost:4873 --no-git-reset --yes", | |
| "release:ci": "SKIP_CHANGELOG=true lerna publish from-package --registry http://localhost:4873 --no-git-reset --yes --canary prerelease", |
There was a problem hiding this comment.
Re this suggested change - moving the SKIP_CHANGELOG here is a good idea, however the from-package option conflicts with the other options passed, so doesn't make sense in this context. See:
https://github.com/lerna/lerna/blob/a7ad9b60d27b390fde21fd2837f2d97320c4603e/commands/publish/index.js#L135-L143
Re --canary, we used to use that, however it used to append the Git SHA to the version in a way that created a semver-spec-violating version string if the Git SHA happened to have a leading zero (which caused errors in Verdaccio), so stopped doing so. However it looks like that was fixed in lerna v3 (lerna/lerna/issues/1118), so we can switch back to using it :-)
There was a problem hiding this comment.
However it looks like that was fixed in lerna v3 (lerna/lerna/issues/1118), so we can switch back to using it :-)
Ah --canary ignores the Neutrino version in lerna.json and instead tries to base it on the Git tags, which (a) doesn't work well on Travis (see https://travis-ci.org/neutrinojs/neutrino/jobs/507436675#L564), (b) means it uses a v8.x.x version rather than v9.x.x, since we don't yet have any Neutrino v9 Git tags.
Given we're not actually the packages in canary form, we don't benefit much from --canary, so I've just switched this back to how it was before.
|
I tested out this PR in Travis with: #1327 and it appears to be working with my suggestions. |
Since it relies on git tags, which doesn't work well on Travis, plus means it tries to using a v8.x.x version number rather than v9.x.x, since there are no git tags for v9 yet. When `--canary` isn't used, the new version will instead be based on that in `lerna.json` which works fine.
Releases will now occur in two phases, with the first being code-only changes that can be reviewed in a PR, followed by the tagging/publishing phase once it's merged.
Fixes #639.
Closes #1311.