This PR replaces #2057.
Implement a collective typecheck action that can be invoked in the root of the
monorepo with `yarn typecheck` or in watch-mode with `yarn watch:typecheck`.
Include the watch-mode typecheck action as part of `yarn start` (a.k.a
`yarn watch`).
To activate collective typecheck for a package in the monorepo, its
`package.json` file should specify:
```
{
"embark-collective": {
"typecheck": true
}
}
```
*-or-*
```
{
"embark-collective": {
"typecheck": {...}
}
}
```
Where `{...}` above is a `tsconfig.json` fragment that will be merged into the
config generated for the package according the same rules that `tsc` applies
when merging [configs][config].
When collective typecheck begins, it generates a `tsconfig.json` for the root
of the monorepo and for each package that is activated for the action. If the
generated JSON is different than what's on disk for the respective root/package
config, or if the config is not present on disk, then it will be
written. Changes to generated `tsconfig.json` files should be committed; such
changes will arise when there are structural changes to the monorepo, e.g. a
package is added, removed, moved and/or the directory containing it is
renamed. Since the configs are only generated at the beginning of collective
typecheck, when structural changes are made in the monorepo `yarn typecheck`
(or `yarn start` or `yarn watch:typecheck`) should be restarted.
Nearly all of the packages in the monorepo (i.e. all those for which it makes
sense) have been activated for collective typecheck. Even those packages that
don't contain `.ts` sources are activated because `tsc` can make better sense
of the code base as a whole owing to the project references included in the
generated `tsconfig.json` files. Also, owing to the fully cross-referenced
`tsconfig.json` files, it's possible for `tsc` to type check the whole code
base without babel (`yarn build` or `yarn watch:build`) having been run
beforehand.
**NOTE** that a *"cold typecheck"* of the whole monorepo is resource intensive:
on this author's 2019 MacBook Pro it takes around three minutes, the fans spin
up, and `tsc` uses nearly 0.5 GB of RAM. However, once a full typecheck has
completed, the next full typecheck will complete in a few seconds or less; and
when running in watch-mode there is likewise a *big* speedup once a full
typecheck has completed, whether that full check happened before it's running
in watch-mode or when watch-mode itself resulted in a full check before
switching automatically to incremental check, as well a corresponding *big*
reduction in resource consumption. A full check will be needed any time
`yarn typecheck` (or `yarn start` or `yarn watch:typecheck`) is run in a fresh
clone plus `yarn install`, or after doing `yarn reboot[:full]` or `yarn reset`.
The combination of options in each generated package-level `tsconfig.json` and
the root `tsconfig.base.json` result in `tsc` writing `.d.ts` files (TypeScript
declaration files) into the `dist/` directory of each package. That
output is intended to live side-by-side with babel's output, and therefore the
`"rootDir"` option in each generated config is set to `"./src"`.
In projects activated for collective typecheck, `.js` may be converted to `.ts`
and/or `.ts` sources may be added without any additional changes needed in
package-level `package.json`.
---
Reorganize types in `packages/core/typings` (a.k.a `@types/embark`) into
`packages/core/core` (`embark-core`), refactor other packages' imports
accordingly, and delete `packages/core/typings` from the monorepo. This results
in some similarly named but incompatible types exported from `embark-core`
(e.g. `Events` and `EmbarkEvents`, the latter being the one from
`packages/core/typings`); future refactoring should consolidate those types. To
avoid circular dependency relationships it's also necessary to split out
`Engine` from `embark-core` into its own package (`embark-engine`) and to
introduce a bit of duplication, e.g. the `Maybe` type that's now defined in
both `embark-i18n` and `embark-core`.
In the process of the types reorg, move many dependencies spec'd in various
`package.json` to the `package.json` of the package/s that actually depend on
them, e.g. many are moved from `packages/embark/package.json` to
`packages/core/engine/package.json`. Related to those moves, fix some Node.js
`require`-logic related to bug-prone dependency resolution.
Fix all type errors that appeared as a result of activating collective
typecheck across the whole monorepo.
Reactivate `tslint` in `packages/core/core` and fix the remaining linter errors.
Tidy up and add a few items in the root `package.json` scripts.
Bump lerna from `3.16.4` to `3.19.0`.
Bumpt typescript from `3.6.3` to `3.7.2`.
Bumpt tslint from `5.16.0` to `5.20.1`.
Make various changes related to packages' `import`/`require`ing packages that
weren't spec'd in their respective `package.json`. More refactoring is needed
in this regard, but changes were made as the problems were observed in the
process of authoring this PR.
[config]: https://www.typescriptlang.org/docs/handbook/tsconfig-json.html
Make changes by running these commands in the root of the monorepo:
**bugs**
```shell
npx lerna exec --concurrency 1 --stream -- \
'DIRPATH=$(realpath $PWD --relative-to=$LERNA_ROOT_PATH); \
npx json -I -f package.json -e "this.bugs=\
\"https://github.com/embark-framework/embark/issues\""'
```
**homepage**
```shell
npx lerna exec --concurrency 1 --stream -- \
'DIRPATH=$(realpath $PWD --relative-to=$LERNA_ROOT_PATH); \
npx json -I -f package.json -e "this.homepage=\
\"https://github.com/embark-framework/embark/tree/master/${DIRPATH}#readme\""'
```
Don't commit changes to private packages, with the exceptions of embark-typings
and embark-reset because those may switch from private to public, and also
because the latter will be included in `node_modules` of embark even if it is
private since embark-reset is presently a bundled dependency of embark.
Don't include the homepage and bugs fields in dapps generated from the template
packages, except for the demo. Set those dapps' description field to an empty
string.
Ensure every package (inc. private packages) has a description.
Ensure every package (inc. private packages) has a README that begins with:
```markdown
`[pkgJson.name]`
================
> [pkgJson.description]
Visit [embark.status.im](https://embark.status.im/) to get started with
[Embark](https://github.com/embark-framework/embark).
```
Don't include the README in dapps generated from the template packages, except
for the demo.
Previously, templates were in a subdirectory of `packages/embark`. Reorganize
them so that they are member packages of the monorepo. This allows them to
cleanly depend on other members of the monorepo,
e.g. `embarkjs-connector-web3`.
It is desirable for the templates, in the context of the monorepo, to specify
embark as a dependency, to take advantage of `npx embark test` (and it's a
"forward looking" setup re: how we plan to evolve embark). However, if embark
were to specify the template packages as dependencies a circular relationship
would be introduced, which is [unsupported by Lerna][circular]. Therefore,
revise the template generator so that all templates are resolved / fetched at
runtime, i.e. `boilerplate`, `demo`, and `simple` are no longer
"built-ins" *per se*. This change won't be apparent to embark's users, but it
does mean that the template generator won't work (in a production install of
embark) if it can't connect to the npm registry, i.e. when the user runs
`embark demo` or `embark new [--simple]`. When embark is inside the monorepo,
templates are resolved and copied from the yarn workspace rather than being
fetched from the registry, which is convenient for development. Also, any
template dependencies that are members of the monorepo are linked into the
copied template's `node_modules` rather than being installed from the registry,
again for convenience. During template generation, remove scripts and
dependencies that pertain only to membership in the monorepo; for now, that
involves removing embark as a dependency since we're not quite ready for that
arrangement to be the default, i.e. outside of the monorepo.
Refactor the root scripts so that more of them can consistently be used with
Lerna's filter options, e.g. `--scope` and `--ignore`. "Combo" scripts that
don't support filtering generally have a `:full` postfix.
Flip `clean` and `reset` scripts at the root and in the member packages for
consistency re: Lerna's notion of `clean` and embark's notion of `reset`. Have
each package run its `reset` script when its `clean` script is invoked (and
that's all for now), relying on `lerna clean` to delete packages'
`node_modules` in view of how Lerna's topological sorting works.
Lift the implementation of `embark reset` into a private package in
`packages/embark-reset` and make it a bundled dependency of embark. Packages in
`dapps/*` depend on `embark-reset` directly and make use of it with `npx
embark-reset` (but only in monorepo context). This removes a "wart" where
reboots could show errors when embark's sources aren't already built in
`packages/embark/dist`. Users will not notice any difference since `embark
reset` works as before, transparently making use of the `embark-reset`
package. The only downside to having it be a bundled dependency of embark is
that bundled deps have all of their `node_modules` included in the tarball
built with `npm pack` (that's why having the templates as bundled dependencies
of embark isn't a viable approach). However, `embark-reset` only has one
dependency, `rimraf`, which is a tiny module, so the cost seems acceptable.
As part of the reorganization, move `test_dapps` into `dapps/tests` and
`packages/embark/templates` into `dapps/templates`. Keep the directory names
short but revise the package names to facilitate simple filtering with
`embark-dapp-*`. Consolidate `.yarnrc` and `.gitignore` and clean up some
redundant ignore listings.
Scripts run with `--scope embark-dapp-*` use `--concurrency=1` to avoid
conflicts that could arise over network ports. The `ci:full` and `qa:full`
scripts use `--concurrency=1` in all scopes, for two reasons: resource
limitations on Travis and AppVeyor result in slower runs with concurrency >1,
and if something fails in those contexts it's easier to see what went wrong
when Lerna's output isn't interleaved from a bunch of scripts in `packages/*`.
Bump the Lerna version.
[circular]: https://github.com/lerna/lerna/issues/1198#issuecomment-442278902
Michael Bradley, Jr
Iuri Matias