refs #9178
- Migrated settings model regression to unit tests
- Removed redundant/unuseful post tests
- Extracted post model regression tests to API tests
- Renamed test suites for consistency
no issue
- Admin API v2 returned /404/, see comment in code base:
/**
* CASE: admin api should serve preview urls
*
* @NOTE
* The url service has no clue of the draft/scheduled concept. It only generates urls for published resources.
* Adding a hardcoded fallback into the url service feels wrong IMO.
*
* Imagine the site won't be part of core and core does not serve urls anymore.
* Core needs to offer a preview API, which returns draft posts.
* That means the url is no longer /p/:uuid, it's e.g. GET /api/v2/content/preview/:uuid/.
* /p/ is a concept of the site, not of core.
*
* The site is not aware of existing drafts. It won't be able to get the uuid.
*
* Needs further discussion.
*/
no-issue
The 2.17 migration included a bug which set the `is_private`, `amp` and `force_i18n` setting values to `'false'` when they should have been `'true'`
We've reverted these changes by reading the most recent backup file, and setting the value to `'true'` if the backup has it set to `'true'` AND the current db has it set to false.
We've also amended the broken migration, so that it does not cause this issue for future installs
no-issue
Currently if you run two migrations on the same day, the backup is overwritten. This change adds the `HH-mm-ss` to the file name, meaning that you get a unique backup for each migration.
refs #10599
- meta_description output wrong meta description
Only solves meta_description for this use case:
```
routes:
/:
data: page.{slug}
template: t
```
refs #10082
```
routes:
/news/:
data: post.news
```
The twitter_image was not available, because the context is [news, post] and the data is in `data.post`.
The context helper was incorrect. I think it is still not fully correct, but only focused on this use case.
The meta layer needs a full refactoring.
refs #10593
- Added `canonical_url` field to post&pages resources in Admin & Content APIs
- Support for canonical URL on metadata layer (used in {{ghost_head}} helper)
- Made sure the new field is not accessible from API v0.1
- Added handling same domain relative and absolute URLs
refs #10582
- I don't think this is a good idea
- If a user passses "null", we should treat it as a string
- I am not aware of a use case why people have "null" in their database
- If people send "null" via the API, we should respect this and accept a string
refs #10582
- otherwise we will forward string booleans to model layer
- causes trouble if we trigger events
- causes trouble if we want to add conditions to the model e.g. setting.get('value') ?
no issue
- e.g. /feed.xml/ was showing a 422
- any other none slug site requests showed a 422
- should be a 404
- context: site is talking to Content API
- it can happen that the API returns a 422
- the routing layer needs to handle this and always show a 404
- catched error in routing error handling
- need to see if there are more cases
closes#10595
* Added breaking test for img-url helper
Input from the content API was absolute, adding this test to verify my fix
* Updated existing test to breaking test for img-url
Had made a dumb assumption when building images sizes, this updates the
test to fail so we can verify the fix
* Refactored img-url to return as early as possible
This makes it a little easier to parse what's going on, and it also
allows us to remove the check for existence of the image url in the
getImageSizes function
* Refactored img-url config parsing to clean up core logic
Superficial refactor to make future changes able to focus on what rather
than how.
* Refactored internal image detection into helper
We're gonna need to know if the image is internal or not, when we force
the path to relative, if we pull this out now, we have access in the top
level logic
* Removed duplicate checks for internal image
Cleaning up and moving "higher-level" logic into the main function of
the module
* Renamed attr -> requestedImageUrl
Superficial refactor, trying to be more explicit about identifiers
* 🐛 Fixed img-url to output relative url by default
Includes a check to isInternalImage as we never want to make external
images relative.
* Returned early if img-url recieves external url
After realising we never want to deal with external urls, we can
continue to return as early as possible, letting us remove checks and
simplify the more complex logic for internal images.
* Cleaned up the internal image logic
Defining the three functions in order helps to see what operations are
going to happen and in which order, we can then return the result of
each operation applied to the next operation.
no issue
- Descreased ammount of posts inserted by 'posts:mu' fixture so that Travis doesn't timeout.
- The fix is just a patch, needs investigation into what change caused the timeout in the first place
refs #5162
- allow pagination and navigation partial helpers to have attributes passed through to them
- e.g. {{navigation header=true}} -> {{#if header}} will now work
- allows styling navigation to be done differently for different sections of the page
- properly create a data frame, and pass through "this" context
- means {{navigation header=true}} is the same as {{> navigation header=true navigation=@site.navigation}}
- our partial helpers, have the same behaviour exactly as if the partial was called directly
- this is additive, and improves behaviour
closes#10580
- The validation was failing because boolean values in settings can also be "0" and "1". 04c60b4ce1 explains the reason why these 2 new values are allowed
no issue
- Additional JSON.stringify call is redundant because it is already happening internally in Ignition (https://github.com/TryGhost/Ignition/blob/master/lib/logging/GhostLogger.js#L241)
- Left stringification in importer as is, because the use case there is also
putting errors into 'problems' array and seems like those values have to
be stringified
no issue
- Reported here: https://forum.ghost.org/t/in-version-2-16-3-found-bug/6065/3
---
Admin Client sends false or true booleans for `is_private` key.
The settings table has two columns "key" and "value". And "value" is always type TEXT.
If you pass value=false, the db will transform this value into "0".
`settingsCache.get('is_private')` is then always true, even though the value is meant to be false.
We should add a migration in v3 and normalize all setting values to ensure consistent database values. Furthermore, we should improve the handling around settings values in general.
For now, we protect parsing values from DB, which we anyway need to transform the values into the correct data type, because we always save strings. This will protect values being stored as "false" or "1" or whatever.
closes#10570
Added a conditional to only run makePathsAbsolute when database:client
is sqlite3, which keeps expected behaviour (make the
"database:connection:filename" path absolute when running SQLite) while
not breaking MySQL behaviour.
no issue
- trying to use the v0.1 Public API when it was disabled led to a confusing error message, see https://forum.ghost.org/t/403-forbidden-error-on-postman-api-call/6017
- adds an explicit check for the Public API being enabled in the client authentication step and throws a useful error message if client auth is attempted when it's disabled
closes#10512
- Removed field filtering in blog owner fetching because it didn't work before (fields weren't reduced) and now broke generated sql queries (ambiguous id field)
refs #10512
- Fixed ability to fetch specific fields when fetching tag resource by id
- Also only returning `url` field when specified in `fields` parameter
closes#10518
- we had a very generic logic to remove "unwanted" null values
- copied from v0.1
- originally added in 7d4107fec4
- this logic transformed: settings = [{key: 'key', value: null}] to [{key: 'key'}], which is wrong
- i've removed this generic logic completely, because i don't know which purpose it serves
- if there a specific case where we want to remove null values, we should either use the JSON schema or use a specific serializer for the target resource
- added tests to proof that settings API behaves as it should
- one test failed because we removed the isNull logic -> if you send published_at = null on a published post
- the model layer has a piece of logic to force a date if you set published_at to null if the status is published
- protected
closes https://github.com/TryGhost/Ghost/issues/10558
- added conditional to explicitly set `excerpt` to `null` in the API output serializer when a post has no `plaintext` or `custom_excerpt` value
no issue
- Migrated default scheduling adapter to use Got via the request proxy
- SchedulingDefault is the only module that was using superagent so removed it as a dependency
no issue
- was unable to revert 9dd7aff9c6, because it contains members changes
- functional calls did not work correctly, because the content and admin ctrl differentiation happend in the web layer
- `isContentAPI` returned true for `api.v2.settings.edit(data, {context: {internal:true{})`
- content & admin API are using different controllers
- we can just tell which ctrl is content API and which is not
- the direction fits for the content & admin API split
no issue
- throwing an object from a catch handler is not a good idea
- unexpected and broke functional call to API (always returned a 500, because API returned {err: err, method: ...}
* Required kid be a header claim as according to spec
https://tools.ietf.org/html/rfc7515#section-4.1.4 (JWT is an extension of JWS)
* Updated error message for missing kid
* Fixed admin-api key unit tests
* Fixed regression and acceptance tests
no-issue
- Added member auth middleware to siteApp
- Passed member as context in routing service
- set Cache-Control: private for member requests
- fucked up some tests
- Added member as global template variable
- Updated tokens to have expiry of subscription_period_end
no issue
We're creating tooling to convert HTML to Ghost flavoured mobiledoc, however we have cards that allow arbitrary content without a wrapper element which means that we're unable to do a 1:1 mapping of mobiledoc->html->mobiledoc. To work around this problem we now output HTML comments before/after the output of each card so that our converter can extract card content correctly when parsing HTML.
- added `createCard` method which wraps a card's `render()` method to add begin/end comments and updated all cards to use it
- only takes affect for newly added or re-saved posts/pages
refs #10438
- "null" means the resource does not exist (it was sett to "null"), which is not true
- we won't serve primary_tag and primary_author by default
- TODO: add the same change to the Content API v2 (raise issue)
no issue
- Content API v2 served primary_tag by default if members flag is enabled
- reference: b2201d4179
- it's safe to remove, because members is behind the dev flag
no issue
- if you html is NULL e.g. you create a draft post, we always set "html" to ""
- this get's marked as changed
- !this.get('html') was added for the 2.0 migration, because some posts had custom mobiledoc, but no html value
refs #9299
- `contextUser` returns a number and if the previous x_by is "1", then bookshelf marks it as changed ("1" !== 1)
- this is a left over from 0.x, because we still owner as id 1
- as soon as we fix 9299, we don't have to worry about this anymore, because we will fetch the owner id if we need it
no-issue
* Corrected function names for rpc methods
* Updated gateway to store tokens locally
* Fixed lint
* Added hardcoded 30 minute expiry for member tokens
* Added default contentApiAccess config;
* Updated validateAudience method
This is required for security, we need to restrict which domains can access
tokens meant for the content api
refs #10438
- To make response structure future proof and conform to the rest of API responses /images* now returns an object with url property instead of plain url string
no issue
- make use of filter instead of status=all or data.page
- nql was designed to filter data on database layer
- do not break v0.1
- we just got rid of the "status" query param, you should use the filter instead
- get rid of the ugly condition to remove page field if "fields" param was used
- allow filtering on model layer for "findOne"
- do not allow filtering for "findOne" on API layer for now
- the API controller defines what is allowed
- the model layer can allow more by default
- we can re-use the powerful filter logic without adding hacks
refs #10438, refs #10106
* Renamed existing pages ctrl
* Splitted posts & pages for Admin API v2
* Added pages JSON input schema for Admin API v2
* Removed single author for Content & Admin API v2
- single author is not documented
- single author usage is deprecated in v0.1
- single author usage is removed in API v2
* Splitted posts & postsPublic controller for v2
* Removed requirement to send `status=all` from Admin API v2
* Removed `status` option from pages Content API v2
* Removed `status` options from Users Admin API v2
no issue
- this is either documented, not does it work
- the Content API returns authors independing on the status
- filtering by status should not work, because otherwise you could guess the status of a user
- we do not expose the status (!)
refs #10438
- the `updated_at` functions as version control value
- it is required for collision detection
- we might redesign this feature at some point
no issue
- we have seen random test failures recently
- the cause: deadlocks
- @NOTE: Deadlocks can and will happen naturally in innodb when multiple transactions are running and they operate on the same table.
The challenge is just how to minimize, handle or avoid them.
---
Why did the deadlock occur?
The tests insert posts in parallel.
As soon you insert two posts, we will attach the relations.
The relations are basically: tags & authors.
Both tables use foreign keys:
post_id -> posts.id
author_id -> users.id
tag_id -> tags.id
Attaching relations runs through two stages:
- inserting or deleting the row (Bookshelf-Relations)
- updating the row because of sort order (Ghost)
2 or more transactions can create a deadlock on the target relation table because of X and S locks for the foreign key, which get automatically set.
Refs:
https://bugs.mysql.com/bug.php?id=48652https://www.chriscalender.com/advanced-innodb-deadlock-troubleshooting-what-show-innodb-status-doesnt-tell-you-and-what-diagnostics-you-should-be-looking-at/
Long-Term?
- investigate further
- retry deadlocks if we know it's fine?
- drop foreign key and handle in Bookshelf?
no issue
A new Zapier app will be released that uses the v2 Admin API which means it will require an ApiKey that is linked to an Integration.
- adds a `type` column to the `integrations` table with the following types allowed:
- `custom` (default) used by custom integrations added by users
- `builtin` used by built-in integrations that have their own UI and won't show up in the "Custom Integrations" list
- `internal` used by "internal" integrations such as the scheduler
- adds a `zapier` "builtin" integration to the fixtures
no-issue
* Refactored hideMembersOnlyContent to 3 "stages"
* Exported paymentConfigured flag from members service
* Updated Content-API to check members service for paymentConfigured
* Updated members content output serializer to remove content if plan required and no plan
* Updated isContentAPI method
* Moved api util test
refs #10438
- we now try to match by slug or id or email
- fallback to owner
- you cannot create a user via post endpoint
- Ghost uses the invite flow to add users
- get rid of `id` restriction on API level
refs #10438
refs #9100
- Added 'strip' attributes to properties that need to be ignored
- Relaxed 'uri' format to 'uri-reference'
- Made input array for posts more restrictive
Added JSON Schema validations for /tags endpoints
refs #10438
refs #9100
- Added JSON Schemas for POST/PUT /tags endpoints
- Added 'strip' keyword definition schema allowing to strip data and not throw errors on further validation stages
refs #10438
- these fields are not used
- no need to expose them in v2
- we will either remove them in the next major or use them for new features (will see)
no-issue
* Added getPublicConfig method to stripe payment processor
* Added getPublicConfig method to subscriptions service
* Added initial config endpoint for members api
* Added getConfig method to members gateway
refs #10438
- Additional check for present 'name' property before generating a
'slug'. Setting slug should not succeed and throw validation error in later
stages.
refs #10472
- Moved config related variable into function scope, so it can be reset by unit tests
- e47d1e275f broke the build and is being fixed by this commit
refs #9441
* Updated top-level ids to use const
* Removed one layer of indentation
* Added .eslintignore files for server and test tasks
* Added npm scripts for eslint
* Fixed lint command in w/ grunt
* Uninstalled grunt-eslint
* Added eslint config
refs #10438
- Added validation helper based on JSON schema
- Added schema validation for POST/PUT in /posts endpoints
- Refactored existing authors validation test suite
- Extended test coverage with a minimally required structure of post.add validator
refs #10461
- do not break the existing webhooks by keeping both payload formats for subscribers events
- refactored webhooks service to run models through target API version
- added new events described in the target issue reference
- this refactoring & enhancement is undocumented, further breaking changes will happen because we are actively working on: https://github.com/TryGhost/Ghost/issues/10438
refs #10461
- the model layer (only post & user) fetches the model after update
- i assume it was added to ensure a response with all fields
- quick fixing it for now to ensure API layer can access ".wasChanged" to be able to decide if a request modified a resource or not
@NOTE: Bookshelf does not physically update a resource if nothing has changed.
no issue
- the model & api layer suffered from missing fields when creating resources
- usually there is only a handful of fields which are required to insert a resource
- the other fields are nullable and/or get defaults assigned
- the API only returned the configured default fields and the fields you have sent to the API
- this resulted in a response with missing fields
- if you have listend on "created" event, the same happend
- you received a model with missing fields
- we now set the undefined fields to null on purpose to ensure a full model for both cases
@NOTE:
There is no endpoint to serve webhooks (not for v0.1, not for v2).
Exposing the secret is required if an integration fetches it's api keys and it's webhooks.
The secret is currently un-used and not implemented.
