This is a big commit that involves drastic changes to how static assets
(.sql and template files, the whole frontend bundle) are handled.
listmonk distribution should be a self-contained single binary
distribution, hence all static assets should be bundled. After
evaluating several solutions, srtkkou/zgok seemed like the best bet but
it lacked several fundamental features, namely the ability to fall back
to the local filesystem in the absence of embedded assets (for instance,
in the dev mode). Moreover, there was a lot of room for cleanup.
After a PR went unanswered, github.com/knadh/stuffbin was created. Just
like zgok, this enables arbitrary files and assets to be embedded into a
compiled Go binary that can be read during runtime. These changes
followed:
- Compress and embed all static files into the binary during
the build (Makefile) to make it standalone and distributable
- Refactor static paths (/public/* for public facing assets,
/frontend/* for the frontend app's assets)
- Add 'logo_url' to config
- Remove 'assets_path' from config
- Tweak yarn build to not produce symbol maps and override
the default /static (%PUBLIC_URL%) path to /frontend
- Add a name / e-mail "quicksearch" input to the UI
- Implement row selection and aggregation at table level and a "select all" that selects all rows at the query level
- On selected subscribers, add bulk list management (add / remove / unsubscribe), blacklist, and delete
- Add notifications for campaign state change
- Add notifications for import state change
Related changes.
- Add a new 'templates' directory with HTML templates
- Move the static campaign template as a .tpl file into it
- Change Messenger.Push() to accept multiple recipients
- Change exhaustCampaign()'s behaviour to pass metadata to admin emails
When a campaign exceeds N number of message send errors, for instance
SMTP errors, it is now auto-paused until there is manual intervention.
For this, the master goroutine in runner.Run() that was synchronising
between the tick based DB scanner and subscriber fetching has been
split into two. A new queue aggregates send errors from workers
again a threshold after which the campaign is paused.
