Due to restrictions on how Electron package works, the `node_modules` should not be hoisted and not to use s/h-links at all. This is why we need to have two separate installs for electron and non-electron packages in the build.
Tested via the following script
```bash
#!/bin/bash
echo "step 1: clean up"
find . -name "node_modules" -prune -exec rm -rf '{}' +
# git clean -dfX
build_type=canary
echo "step 2: install web dependencies"
# firstly, build web static
PLAYWRIGHT_SKIP_BROWSER_DOWNLOAD=1 SENTRYCLI_SKIP_DOWNLOAD=1 PLAYWRIGHT_SKIP_BROWSER_DOWNLOAD=1 HUSKY=1 yarn
echo "step 3: generate assets"
BUILD_TYPE="$build_type" yarn workspace @affine/electron generate-assets
# cleanup node_modules
find . -name "node_modules" -prune -exec rm -rf '{}' +
echo "step 4: install electron dependencies"
# install electron deps
yarn config set nmHoistingLimits workspaces
yarn config set enableScripts false
yarn config set nmMode classic
PLAYWRIGHT_SKIP_BROWSER_DOWNLOAD=1 HUSKY=0 yarn workspaces focus @affine/electron @affine/monorepo
echo "step 5: build native"
# build native
yarn workspace @affine/native build
yarn workspace @affine/storage build
echo "step 6: build electron"
# build electron
yarn workspace @affine/electron build
echo "step 7: package electron"
# package
SKIP_GENERATE_ASSETS=1 BUILD_TYPE="$build_type" HOIST_NODE_MODULES=1 yarn workspace @affine/electron package
```
I suspect HMR does not working properly on dev because we have multiple entries.
One relative issue: https://github.com/webpack/webpack-dev-server/issues/2792/
I think we do not need multiple entries for polyfills & plugins after all. They could be in the same chunk, and could be later optimized through splitChunks option.
`ses.ts` is changed to `ses-lockdown.ts` because `ses.ts` does not pass circular dependency check by madge. I haven't looked through the real root cause though. See https://github.com/pahen/madge/issues/355
If network offline or API error happens, the `session` returned by the `useSession` hook will be null, so we can't assume it is not null.
There should be following changes:
1. create a page in ErrorBoundary to let the user refetch the session.
2. The `SessionProvider` stop to pull the new session once the session is null, we need to figure out a way to pull the new session when the network is back or the user click the refetch button.
Close#3287
<!--
copilot:all
-->
### <samp>🤖 Generated by Copilot at d3fdf86</samp>
### Summary
📄🚀🔗
<!--
1. 📄 - This emoji represents the page and edgeless modes of sharing a page, as well as the GraphQL operations and types related to public pages.
2. 🚀 - This emoji represents the functionality of publishing and revoking public pages, as well as the confirmation modal and the notifications for the user.
3. 🔗 - This emoji represents the sharing URL and the query parameter for the share mode, as well as the hooks and functions that generate and use the URL.
-->
This pull request adds a feature to the frontend component of AFFiNE that allows the user to share a page in either `page` or `edgeless` mode, which affects the appearance and functionality of the page. It also adds the necessary GraphQL operations, types, and schema to support this feature in the backend, and updates the tests and the storybook stories accordingly.
* Modify the `useIsSharedPage` hook to accept an optional `shareMode` argument and use the `getWorkspacePublicPagesQuery`, `publishPageMutation`, and `revokePublicPageMutation` from `@affine/graphql`
It seems there are some cases that [this upstream PR](https://github.com/toeverything/blocksuite/pull/4747) will cause data loss.
Because of some historical reasons, the page id could be different with its doc id.
It might be caused by subdoc migration in the following (not 100% sure if all white screen issue is caused by it) 0714c12703/packages/common/infra/src/blocksuite/index.ts (L538-L540)
In version 0.10, page id in spaces no longer has prefix "space:"
The data flow for fetching a doc's updates is:
- page id in `meta.pages` -> find `${page-id}` in `doc.spaces` -> `doc` -> `doc.guid`
if `doc` is not found in `doc.spaces`, a new doc will be created and its `doc.guid` is the same with its pageId
- because of guid logic change, the doc that previously prefixed with `space:` will not be found in `doc.spaces`
- when fetching the rows of this doc using the doc id === page id,
it will return EMPTY since there is no updates associated with the page id
The provided fix in the PR will patch the `spaces` field of the root doc so that after 0.10 the page doc can still be found in the `spaces` map. It shall apply to both of the idb & sqlite datasources.
Special thanks to @lawvs 's db file for investigation!
