And remedy weird API in the process. Autocomplete now exposes `onSelectionChange` to indicate that the user has selected another completion, rather than returning the chosen completion via onUpArrow etc.
Fixesvector-im/riot-web#4835
because using `body` gives inconsistent results - sometimes it will contain markdown and sometimes not, and this may not correspond with the `formatted_body`.
TODO: Do quoting proper - using `in_response_to`.
(not formatted_body)
This is because draft-js has regressed with a bug that causes some entities to not exist within a given ContentState - see vector-im/riot-web#4756
This makes sure that the length of the range for a completed Entity = the length of the text in the decoration, which apparently draftjs assumes when calculating selection state offsets.
Fixes https://github.com/vector-im/riot-web/issues/4666
Tab-completing a room alias inserts a markdown link, which is now stripped before a command is parsed. We also strip the MD links when doing the autocompletion itself (getAutocompleteQuery).
There is still an issue where `/invite` will not work at all with tab-completion - the text of a user Pill is not the userID but rather the display name, which cannot be used as an argument to the command.
This makes sure the selection state is correct when moving the cursor through an entity in the composer but only in rich text mode. https://github.com/vector-im/riot-web/issues/4666 still persists despite this, probably because the text content inserted during a completion in markdown mode has a different length to the text node within the entity.
fixesvector-im/riot-web#4654
Instead of blindly stripping all MD mentions, only strip those that were tab-completed. We do this by adding the `isCompleted` flag to the Entity data.
This isn't an entirely side-effect-free refactoring:
- the text of the timeline pills is now either the room ID/alias or user ID/ display name of the linked resource (which means that until we do a roundtrip to get user displaynames, mentions for users not in the current room will have their user IDs shown instead of what was in the link body).
- timeline links to rooms without avatars are now rendered as links
- fixed issue that would throw an error whilst rendering (i.e. unusable client) a room link to a room that the client doesn't know about
These will appear decorated because they are inserted as entities. It was necessary to modify pills to have an explicit linkText that is derived from the `href` being pillified (and is thus no longer the inserted completion but rather the display name (or user ID) or room alias.
by inspecting whether multiple lines are selected or the selection is empty. If either of these are true, insert a code block surrounding the selection, otherwise insert single backticks around the selection for inline code formatting.
This modifies the composer completion such that completing a room or user will insert an IMMUTABLE matrix.to LINK Entity for the range that was replaced. Display names will not have a colon after their name anymore as it seemed strange that we would insert one after a pill.
Any links in the composer that are recognised as matrix.to links will be decorated as `<span>`s with CSS classes (one of mx_UserPill or mx_RoomPill). This implementation has the nice bonus that switching to markdown (and back) will Just Work.
This will have some CSS changes coming to better match the design.
Hitting return in a code-block will now split the block into two code blocks. (Holding shift will still insert a soft newline into the current block).
We still need to make it a bit more obvious that consecutive code-blocks
are not contiguous - https://github.com/vector-im/riot-web/issues/4535
The easiest way to stop the user from inserting whitespace onto the end of an entity is to toggle the entity state of the whitespace that was just entered. This allows the user to continue drafting a message without editing the link content.
This is for pasted `<a>` tags that have been copied from a website. We probably also want to be storing entities for substrings of content that are determined to be URLs.
which is now https://github.com/vector-im/riot-web/issues/4537 <- there.
This does two things:
- Track which query was the most recent one requesting completion and only process completions for that one. (In this case the empty string "" doesn't have any completions but we still track it so that previous calls with non-empty queries would not race and cause completions to be shown when we actuall don't want any.)
- Make the "do we want to show the AutoComplete box?" logic a bit more sane
This behaviour was present in the old composer but implemented using local storage. This is unecessary as we don't really care about our drafts across clients, the important thing is that our draft is kept when switching rooms.
As a bonus, ifnore vertical arrow key presses when a modifier key is pressed so that the room switching keys (alt + up/down arrow) don't also cause history browsing (or autocomplete browsing).
As suggested by @dbkr (because this is what we do for the old composer), only move through history when the caret is that the beginning of the first line (block) or end of the last.
This has the nice property of being able to move the caret up to a really long message: fixes https://github.com/vector-im/riot-web/issues/4471
Because converting them to HTML when they got sent as HTML seems a bit pointless when they're just going to get transformed back again when retrieved from history.
Fixes regression https://github.com/vector-im/riot-web/issues/4465
When there are no styled blocks or inline styles applied within blocks, just send text instead of HTML.
Also, don't add <br /> for the last <p> (the last block).
Fixes https://github.com/vector-im/riot-web/issues/3147
