refs #9601
- implementation of resource listener updated
- if you define two collections: `featured:true` (1) and `page:false` (2) you can run into the following bug:
- you create a published post (owned by (2))
- you change the status to featured
- still owned by (2), because the filter still matches (it's still not a static page)
- this adaption fixes the behaviour
- less smart logic, but less error prone
no issue
- if you define no collections, but a static route, it can happen that the target template to render
makes use of the {{ghost_head}} helper
- the {{ghost_head}} helper tries to create the primary rss feed url
- at the moment: no collections, no primary rss feed url
- if we offer the option to define custom rss rules, this function might need an extension
no issue
- this mock eat already too much of my/our time
- the idea of adding a knex mock was definitely a failed approach/try
- it's too much to maintaince and have not found a module which does this already
- we have to support any query format
- this is too crazy
- the idea was to use the knex mock for model unit tests, because if we want to unit test models we have to
run through bookshelf, because the whole model layer depends on bookshelf e.g. events
- for now we simply use the real database
- we could use the sqlite3 memory mode, but that would mean every unit test runs on sqlite3
- something to consider for later e.g. run unit tests on one matrix
- run the rest on another matrix for sqlite + mysql
closes#9675
- with dynamic routing we have introduced a breaking change, which we have overseen
- Ghost does not return absolute urls, that's why the clients need to concat the blog url and the resource url
- with 1.24.0 Ghost returned resource urls including the subdirectory
- this caused trouble for e.g. zapier or the preview feature in the admin client
- revert breaking change and ensure we only expose resource urls without subdirectory
no issue
- from now on: you have to manually reconfigure your slack hook after importing your data
- we were running into trouble that Ghost had import slack hooks, because it can happen very fast
that you are importing someone's slack hook
refs #9601
### Dynamic Routing
This is the beta version of dynamic routing.
- we had a initial implementation of "channels" available in the codebase
- we have removed and moved this implementation
- there is now a centralised place for dynamic routing - server/services/routing
- each routing component is represented by a router type e.g. collections, routes, static pages, taxonomies, rss, preview of posts
- keep as much as possible logic of routing helpers, middlewares and controllers
- ensure test coverage
- connect all the things together
- yaml file + validation
- routing + routers
- url service
- sitemaps
- url access
- deeper implementation of yaml validations
- e.g. hard require slashes
- ensure routing hierarchy/order
- e.g. you enable the subscriber app
- you have a custom static page, which lives under the same slug /subscribe
- static pages are stronger than apps
- e.g. the first collection owns the post it has filtered
- a post cannot live in two collections
- ensure apps are still working and hook into the routers layer (or better said: and register in the routing service)
- put as much as possible comments to the code base for better understanding
- ensure a clean debug log
- ensure we can unmount routes
- e.g. you have a collection permalink of /:slug/ represented by {globals.permalink}
- and you change the permalink in the admin to dated permalink
- the express route get's refreshed from /:slug/ to /:year/:month/:day/:slug/
- unmount without server restart, yey
- ensure we are backwards compatible
- e.g. render home.hbs for collection index if collection route is /
- ensure you can access your configured permalink from the settings table with {globals.permalink}
### Render 503 if url service did not finish
- return 503 if the url service has not finished generating the resource urls
### Rewrite sitemaps
- we have rewritten the sitemaps "service", because the url generator does no longer happen on runtime
- we generate all urls on bootstrap
- the sitemaps service will consume created resource and router urls
- these urls will be shown on the xml pages
- we listen on url events
- we listen on router events
- we no longer have to fetch the resources, which is nice
- the urlservice pre-fetches resources and emits their urls
- the urlservice is the only component who knows which urls are valid
- i made some ES6 adaptions
- we keep the caching logic -> only regenerate xml if there is a change
- updated tests
- checked test coverage (100%)
### Re-work usage of Url utility
- replace all usages of `urlService.utils.urlFor` by `urlService.getByResourceId`
- only for resources e.g. post, author, tag
- this is important, because with dynamic routing we no longer create static urls based on the settings permalink on runtime
- adapt url utility
- adapt tests
no issue
- discovered while coding
- the value was always false, because we've tried to read the value from the config object
- the value lives in the database and is accessible via the labs service
no issue
- we sanitise any incoming slug on the model layer e.g uppercase -> lowercase
- and when importing e.g. an uppercase slug, the importer was trying to compare the uppercase slug with the sanitised slug
refs #9601
- replace raw knex queries by Bookshelf queries
- optimise lot's of test setups, so we don't experience a massive slow down in the test run
- this has troubled in the past e.g. with normalisation, any custom model logic - the test env always had to simulate things
- there are for sure thousands things which can be optimised now, but because of time, we do them step by step
- this is especially important for the url service (https://github.com/TryGhost/Ghost/issues/9601), because we have to ensure that inserting/updating/removing resources will trigger model events
`grunt test-all` with SQLite finishes in 2,5-3min. (on master: 1-2min)
`grunt test-all` with MySQL finishes in 4min. (on master: 3min)
**NOTE: We want to move as much as possible routing and integration tests to unit tests. This will speed up the test run again.** See #9342. But we need to find time for that. Any help is welcome!
refs https://github.com/TryGhost/Ghost/pull/9592
- we add bookshelf-relations step by step if we need it
- with https://github.com/TryGhost/Ghost/pull/9592 we have rewritten the test env to use Bookshelf
- this is important for our new url service
- because the service is listening on model updates and updates the urls based on the model events
- so with moving to Bookshelf, we need any easy way to add relations
- the test env inserts test fixtures
- it adds permissions and each permission get's roles attached
- `models.Permission.add({roles: [...]})
no issue
- removed the `routeKeywords` property from the config and used hard coded keywords.
- removed `routeKeywords` from public configuration API endpoint, as it's no longer used in the Admin.
closes#9547
- you setup a blog with the following owner:
- email: test@ghost.org
- name: test
- slug: test
- now you import a JSON db file, which holds the exact same owner
- this owner won't be imported, because it's a duplicate
- but the slug is different (!)
- the importer tries to find a matching existing user, but won't find anything
- the importer then send an empty authors array `post.authors=[]` into the model layer
- this is not allowed -> this would mean, you are actively trying to unset all authors
no issue
- if multiple queries run in a transaction, the model events are triggered before the txn finished
- if the txn rolls back, the events are anyway emitted
- the events are triggered too early
- solution:
- `emitChange` needs to detect that a transaction is happening
- it listens on a txn event to determine if events should be triggered
no issue
- if you delete all content, we expect two events
- `post.deleted` and `post.unpublished`
- `post.unpublished` was never triggered, because the api implementation made use of `collection.invoke(`destroy`)`
- what happened?
- you fetch all posts (columns:id)
- you destroy the post (only id column is available)
- the model events are triggered
- but you have no access to a default set of data
- the result is that the event handler can't even tell if this is a post or a page
- added a proper test to ensure which events are triggered
refs #9548
- we always receive date strings from the client in ISO format
- we ensure that we transform these strings into JS dates for comparison
- when the client sends relations, we need to ensure that relations are checked as well
- will only work for the post model for now, because this is the only model which uses `bookshelf-relations`
- added unit tests
- removed some model tests, which do the same
refs #9548
- do not forward `tag.parent` to the model layer
- the model layer should only know `tag.parent_id`
- and the API should only expose `tag.parent` (this is an API feature)
- currently Ghost has a mixture of using `toJSON` and the API validation layer for this
- we just continue with this for now (no time to fix this)
- disallow sending nested-nested relations
- unsupported
- see comment for more information
- this can cause problems with calling `hasChanged` on relations
- add unit tests
no issue
This PR adds the server side logic for multiple authors. This adds the ability to add multiple authors per post. We keep and support single authors (maybe till the next major - this is still in discussion)
### key notes
- `authors` are not fetched by default, only if we need them
- the migration script iterates over all posts and figures out if an author_id is valid and exists (in master we can add invalid author_id's) and then adds the relation (falls back to owner if invalid)
- ~~i had to push a fork of bookshelf to npm because we currently can't bump bookshelf + the two bugs i discovered are anyway not yet merged (https://github.com/kirrg001/bookshelf/commits/master)~~ replaced by new bookshelf release
- the implementation of single & multiple authors lives in a single place (introduction of a new concept: model relation)
- if you destroy an author, we keep the behaviour for now -> remove all posts where the primary author id matches. furthermore, remove all relations in posts_authors (e.g. secondary author)
- we make re-use of the `excludeAttrs` concept which was invented in the contributors PR (to protect editing authors as author/contributor role) -> i've added a clear todo that we need a logic to make a diff of the target relation -> both for tags and authors
- `authors` helper available (same as `tags` helper)
- `primary_author` computed field available
- `primary_author` functionality available (same as `primary_tag` e.g. permalinks, prev/next helper etc)
closes#9520
- it contains a dependency bump of the latest Bookshelf release
- Bookshelf introduced a bug in the last release
- see https://github.com/bookshelf/bookshelf/pull/1583
- see https://github.com/bookshelf/bookshelf/pull/1798
- this has caused trouble in Ghost
- the `updated_at` attribute was not automatically set anymore
---
The bookshelf added one breaking change: it's allow to pass custom `updated_at` and `created_at`.
We already have a protection for not being able to override the `created_at` date on update.
We had to add another protection to now allow to only change the `updated_at` property.
You can only change `updated_at` if you actually change something else e.g. the title of a post.
To be able to implement this check i discovered that Bookshelfs `model.changed` object has a tricky behaviour.
It remembers **all** attributes, which where changed, doesn't matter if they are valid or invalid model properties.
We had to add a line of code to avoid remembering none valid model attributes in this object.
e.g. you change `tag.parent` (no valid model attribute). The valid property is `tag.parent_id`.
If you pass `tag.parent` but the value has **not** changed (`tag.parent` === `tag.parent_id`), it will output you `tag.changed.parent`. But this is wrong.
Bookshelf detects `changed` attributes too early. Or if you think the other way around, Ghost detects valid attributes too late.
But the current earliest possible stage is the `onSaving` event, there is no earlier way to pick valid attributes (except of `.forge`, but we don't use this fn ATM).
Later: the API should transform `tag.parent` into `tag.parent_id`, but we are not using it ATM, so no need to pre-optimise.
The API already transforms `post.author` into `post.author_id`.
no issue
- just discovered that we had confusing function names in our test utility
- e.g. `posts` -> default posts from the data generator
- e.g. `users` -> extra users not from our data generator
- now:
- e.g. `posts` -> default posts from the data generator
- e.g. `users` -> default users from the data generator
- e.g. `users:extra` -> extra users not from our data generator
no issue
- replace logic for preparing nested tags
- if you have nested tags in your file, we won't update or update the target tag
- we simply would like to add the relationship to the database
- use same approach as base class
- add `posts_tags` to target post model
- update identifiers
- insert relation by foreign key `tag_id`
- bump bookshelf-relations to 0.1.10
no issue
- change behaviour from updating user references after the actual import to update the user reference before the actual import
- updating user references after the import is way less case intense
- that was the initial decision for updating the references afterwards
- but that does not play well with adding nested relations by identifier
- the refactoring is required for multiple authors
- if we e.g. store invalid author id's, we won't be able to add a belongs-to-many relation for multiple authors
- bookshelf-relations is generic and always tries to find a matching target before attching a model
- invalid user references won't work anymore
- this change has a very good side affect
- 17mb takes on master ~1,5seconds
- on this branch it takes ~45seconds
- also the memory usage is way lower and stabler
- 40mb takes 1,6s (times out on master)
no issue
- Ghost does not support adding an author by relation (`post.author = {id: '..'}`)
- Ghost does not support editing an author by relation (`post.author = {id: '..'}`)
- only `author_id` is allowed
refs https://github.com/TryGhost/Ghost/issues/3658
- the `validateSchema` helper was a bit broken
- if you add a user without email, you will receive a database error
- but the validation error should catch that email is passed with null
- it was broken, because:
- A: it called `toJSON` -> this can remove properties from the output (e.g. password)
- B: we only validated fields, which were part of the JSON data (model.hasOwnProperty)
- we now differentiate between schema validation for update and insert
- fixed one broken import test
- if you import a post without a status, it should not error
- it falls back to the default value
- removed user model `onValidate`
- the user model added a custom implementation of `onValidate`, because of a bug which we experienced (see https://github.com/TryGhost/Ghost/issues/3638)
- with the refactoring this is no longer required - we only validate fields which have changed when updating resources
- also, removed extra safe catch when logging in (no longer needed - unit tested)
- add lot's of unit tests to proof the code change
- always call the base class, except you have a good reason
no issue
- `isNew` does not work in Ghost, because Ghost does not use auto increment id's
- see https://github.com/bookshelf/bookshelf/issues/1265
- see https://github.com/bookshelf/bookshelf/blob/0.10.3/src/base/model.js#L211
- we only had one occurance, which was anyway redundant
- if you add a user, `hasChanged('password') is true
- if you edit a user and the password has changed, `hasChanged('password')` is true as well
NOTE #1:
1. We can't override `isNew` and throw an error, because bookshelf makes use of `isNew` as well, but it's a fallback if `options.method` is not set.
2. It's hard to re-implement `isNew` based on `options.method`, because then we need to ensure that this value is always set (requires a couple of changes)
NOTE #2:
If we need to differentiate if a model is new or edited, we should manually check for `options.method === insert`.
NOTE #3:
The unit tests are much faster compared to the model integration tests.
I did a comparision with the same test assertion:
- unit test takes 70ms
- integration test takes 190ms
no issue
- this commit cleans up the usages of `include` and `withRelated`.
### API layer (`include`)
- as request parameter e.g. `?include=roles,tags`
- as theme API parameter e.g. `{{get .... include="author"}}`
- as internal API access e.g. `api.posts.browse({include: 'author,tags'})`
- the `include` notation is more readable than `withRelated`
- and it allows us to use a different easier format (comma separated list)
- the API utility transforms these more readable properties into model style (or into Ghost style)
### Model access (`withRelated`)
- e.g. `models.Post.findPage({withRelated: ['tags']})`
- driven by bookshelf
---
Commits explained.
* Reorder the usage of `convertOptions`
- 1. validation
- 2. options convertion
- 3. permissions
- the reason is simple, the permission layer access the model layer
- we have to prepare the options before talking to the model layer
- added `convertOptions` where it was missed (not required, but for consistency reasons)
* Use `withRelated` when accessing the model layer and use `include` when accessing the API layer
* Change `convertOptions` API utiliy
- API Usage
- ghost.api(..., {include: 'tags,authors'})
- `include` should only be used when calling the API (either via request or via manual usage)
- `include` is only for readability and easier format
- Ghost (Model Layer Usage)
- models.Post.findOne(..., {withRelated: ['tags', 'authors']})
- should only use `withRelated`
- model layer cannot read 'tags,authors`
- model layer has no idea what `include` means, speaks a different language
- `withRelated` is bookshelf
- internal usage
* include-count plugin: use `withRelated` instead of `include`
- imagine you outsource this plugin to git and publish it to npm
- `include` is an unknown option in bookshelf
* Updated `permittedOptions` in base model
- `include` is no longer a known option
* Remove all occurances of `include` in the model layer
* Extend `filterOptions` base function
- this function should be called as first action
- we clone the unfiltered options
- check if you are using `include` (this is a protection which could help us in the beginning)
- check for permitted and (later on default `withRelated`) options
- the usage is coming in next commit
* Ensure we call `filterOptions` as first action
- use `ghostBookshelf.Model.filterOptions` as first action
- consistent naming pattern for incoming options: `unfilteredOptions`
- re-added allowed options for `toJSON`
- one unsolved architecture problem:
- if you override a function e.g. `edit`
- then you should call `filterOptions` as first action
- the base implementation of e.g. `edit` will call it again
- future improvement
* Removed `findOne` from Invite model
- no longer needed, the base implementation is the same
closes#9445
- redirects all asset requests if https is configured (theme, core, images)
- re-use and extend our url-redirect middleware
- add proper integration tests for our express site app (no db interaction, component testing required for such important use cases)
- i added some more general tests
- should avoid mixed content warnings in the browser
closes#9314
* added fixtures for contributor role
* update post api tests to prevent contributor publishing post
* update permissible function in role/user model
* fix additional author code in invites
* update contributor role migration for knex-migrator v3
* fix paths in contrib migration
* ensure contributors can't edit or delete published posts, fix routing tests [ci skip]
* update db fixtures hash
* strip tags from post if contributor
* cleanup post permissible function
* excludedAttrs to ignore tag updates for now (might be removed later)
* ensure contributors can't edit another's post
* migration script for 1.21
no issue
- returning and remembering the data, which was imported, is...
- not required when using the API
- not required when importing via script
- required for tests
- added an option to have control over it
- make more usage of local variables
- the GC cannot tidy up variables, which are defined outside of a loop, but used in the loop
- try to keep less memory in process
- reduce the number of properties we have to remember
no issue
- if you import a JSON file with a post, which has an unknown author,
the target user was removed from the blog
- Ghost can handle this case and still succeeds with import
- but we have stored an `author_id` in the database, which does not map to any user and won't map in the future
- this can trouble if we add support for multiple authors
- currently, we only return the `author_id` to the client and the client can map with `author_id` with users fetched by the API
- if it does not find a user, it just falls back to a different user
- but multiple authors have to be included explicit (`include=authors`) and we will return a mapped (author_id => user) result
- it won't be able to find the user, because we lookup the database
- this would result in an error
- there is in general no reason to import (or store) an unknown/invalid `author_id` into the database
- on import, we show you a warning and you can choose a different author if you want
- solution: fallback to owner user and extend warning
- it's not a behaviour change, you still can import unknown author id's and the import won't fail
- but we ensure valid author id's
- updated test
- further more: returning `author={}` when requesting `include=author` could trouble with ember currently
- it expects the author to be returned
no issue
- reported in slack (https://ghost.slack.com/files/U8QV8DXQB/F8TSBQ532/image.png)
- do not expose old release notification
- e.g. you are on 1.20.0
- you receive a notification for 1.20.1 to update
- you update to 1.20.1
- ensure we protect exposing the release notification (compare against blog version)
- protect against wrong formats
- @TODO: the notifications could store a `version` property
- by that we could use `notification.version` and don't have to match the version in the message
closes#5071
- Remove hardcoded notification in admin controller
- NOTE: update check notifications are no longer blocking the admin rendering
- this is one of the most import changes
- we remove the hardcoded release message
- we also remove adding a notification manually in here, because this will work differently from now on
-> you receive a notification (release or custom) in the update check module and this module adds the notification as is to our database
- Change default core settings keys
- remove displayUpdateNotification
-> this was used to store the release version number send from the UCS
-> based on this value, Ghost creates a notification container with self defined values
-> not needed anymore
- rename seenNotifications to notifications
-> the new notifications key will hold both
1. the notification from the USC
2. the information about if a notification was seen or not
- this key hold only one release notification
- and n custom notifications
- Update Check Module: Request to the USC depends on the privacy configuration
- useUpdateCheck: true -> does a checkin in the USC (exposes data)
- useUpdateCheck: false -> does only a GET query to the USC (does not expose any data)
- make the request handling dynamic, so it depends on the flag
- add an extra logic to be able to define a custom USC endpoint (helpful for testing)
- add an extra logic to be able to force the request to the service (helpful for testing)
- Update check module: re-work condition when a check should happen
- only if the env is not correct
- remove deprecated config.updateCheck
- remove isPrivacyDisabled check (handled differently now, explained in last commit)
- Update check module: remove `showUpdateNotification` and readability
- showUpdateNotification was used in the admin controller to fetch the latest release version number from the db
- no need to check against semver in general, the USC takes care of that (no need to double check)
- improve readability of `nextUpdateCheck` condition
- Update check module: refactor `updateCheckResponse`
- remove db call to displayUpdateNotification, not used anymore
- support receiving multiple custom notifications
- support custom notification groups
- the default group is `all` - this will always be consumed
- groups can be extended via config e.g. `notificationGroups: ['migration']`
- Update check module: refactor createCustomNotification helper
- get rid of taking over notification duplication handling (this is not the task of the update check module)
- ensure we have good fallback values for non present attributes in a notification
- get rid of semver check (happens in the USC) - could be reconsidered later if LTS is gone
- Refactor notification API
- reason: get rid of in process notification store
-> this was an object hold in process
-> everything get's lost after restart
-> not helpful anymore, because imagine the following case
-> you get a notification
-> you store it in process
-> you mark this notification as seen
-> you restart Ghost, you will receive the same notification on the next check again
-> because we are no longer have a separate seen notifications object
- use database settings key `notification` instead
- refactor all api endpoints to support reading and storing into the `notifications` object
- most important: notification deletion happens via a `seen` property (the notification get's physically deleted 3 month automatically)
-> we have to remember a seen property, because otherwise you don't know which notification was already received/seen
- Add listener to remove seen notifications automatically after 3 month
- i just decided for 3 month (we can decrease?)
- at the end it doesn't really matter, as long as the windows is not tooooo short
- listen on updates for the notifications settings
- check if notification was seen and is older than 3 month
- ignore release notification
- Updated our privacy document
- Updated docs.ghost.org for privacy config behaviour
- contains a migration script to remove old settings keys
no issue
- with 29e143fa9a import queries no longer run in parallel
- this commit simply adds a small code snippet to reflect the importer behaviour
1) duplicate slugs *within* a file are getting ignored
2) existing posts in the database and posts to import with the same slug, result in duplicates
Further improvements regarding duplication detection will happen via #8717.
closes#8717
- this is now required, because we run import queries sequentiell
- this code protects two cases:
- you have duplicate slugs in the JSON file (the first get's inserted, the second get's ignored)
- you have an existing slug in the database and you try to import the same slug, get's ignored
refs #9178
- i am not super happy about `const imageLib = require('../lib/image')`
- i don't really like the name `imageLib`
- but i had no better idea 😃
- if we use the same name in the whole project, it's very easy to rename the folder or the variable
no issue
> Deprecation warning: value provided is not in a recognized ISO format. moment construction falls back to js Date(), which is not reliable across all browsers and versions.
