no issue
- coverage reports were not really being used and were frequently ignored when merging PRs because we knew there were reductions in test coverage
- coveralls also causes some problems with Renovate because there can be random micro changes in reported coverage that meant a PR "failed" when really it was fine
no issue
- `@tryghost/kg-parser-plugins` contains the parser plugins used by the editor and was extracted so that they can be used in server-side html-to-mobiledoc conversion
no issue
- `cleanBasicHtml` has been extracted to an external package so that it can be used inside the extracted `@tryghost/kg-parser-plugins` package server-side
* ✨ Added caption support to code cards
no issue
* fix caption placeholder text
* add figure-to-code-card parser plugin, add language detection to code card parsers
* extract {{kg-action-bar}} component
* do not show card toolbar if caption has focus
no issue
- Added a language indicator when in rendered mode and a language input when in edit mode
- Allow code card language to be set with <code>```lang</code>+<kbd>Space/Enter</kbd> expansion
- previously <code>\`\`\`</code> would immediately create a code card, the <kbd>Space/Enter</kbd> is now necessary for the insertion to occur
- lang is optional <code>\`\`\`</code>+<kbd>Space/Enter</kbd> will insert a code card with no language selected
- requires <kbd>Enter</kbd> to be pressed to finalise the expansion and insert the card
- added hook for text expansions to skip newline creation for when they are triggered with <kbd>Enter</kbd>
- Set the code card editor's language mode based on selected language
- set the CodeMirror mode based on the code card payload language
- add a basic map of language short codes to their respective CodeMirror modes
- observe `mode` property in `{{gh-cm-editor}}` so that the mode is properly set when it's changed after initial render
closes https://github.com/TryGhost/Ghost/issues/9847
- the text expansion is triggered twice in this situation, the first time with `""` and the second with `"/"`. The second trigger was trying to match cards to `"/"` which returned an empty card set
- added a quick workaround that ensures we never include a leading `/` in the card selection query
closes https://github.com/TryGhost/Ghost/issues/10677
- adjusted the basic markdown text expansion behaviour to delete the MD chars and add a markup to the resulting range rather than deleting the whole range and re-adding it
- extracted basic markdown format behaviour into a helper function
- fixed the link markup expansion so that `[](https://example.com)` is recognised and creates a link with the URL as the text
Over in:
* https://github.com/whatwg/html/issues/3840
* https://bugzilla.mozilla.org/show_bug.cgi?id=1281135
I'm trying to come up with a model for `<link rel="stylesheet" disabled>` in
which Blink / WebKit and Firefox can agree on.
See that HTML spec issue for all the inconsistencies of WebKit / Blink, and the
following post for more context:
* https://groups.google.com/d/msg/mozilla.dev.platform/BdgNaChHnpY/mhXzCBwSCgAJ
---
Unfortunately, my change to Firefox breaks the Ghost Admin panel night-mode
switch (you can see it in Firefox Nightly).
This is because with my change, removing the `disabled` attribute from an
stylesheet behaves the same regardless of whether the `disabled` attribute is
added dynamically or not.
That means that adding the `disabled` attribute dynamically "unloads" the
stylesheet completely (just like when the attribute is there before inserting
the link in the document, or from the parser). Thus removing the attribute will
load the stylesheet again and fire a load event.
This is problematic for the code as-is, because it means that each time that the
load event fires when the disabled attribute is removed on an alternate, then
it's added again. :)
Prevent that from happening by removing the load event listener ASAP. What this
code wants is to only resolve the promise once after all.
Given this is so far the only regression from my change that has been reported
(over at https://bugzilla.mozilla.org/show_bug.cgi?id=1546707), I think fixing
the Ghost-Admin panel is worth it.
If this pattern is somehow common, then we'll probably revert that patch and go
back to the sad current state of affairs regarding interop :(
- currently, we only build renovate PRs
- that means we can't use automerge branch, as we wouldn't run any checks
- this swaps to running only renovate pushes, which means the branch always gets checked
