diff options
Diffstat (limited to 'src/lib/strings/rich-text.ts')
-rw-r--r-- | src/lib/strings/rich-text.ts | 216 |
1 files changed, 0 insertions, 216 deletions
diff --git a/src/lib/strings/rich-text.ts b/src/lib/strings/rich-text.ts deleted file mode 100644 index 1df2144e0..000000000 --- a/src/lib/strings/rich-text.ts +++ /dev/null @@ -1,216 +0,0 @@ -/* -= Rich Text Manipulation - -When we sanitize rich text, we have to update the entity indices as the -text is modified. This can be modeled as inserts() and deletes() of the -rich text string. The possible scenarios are outlined below, along with -their expected behaviors. - -NOTE: Slices are start inclusive, end exclusive - -== richTextInsert() - -Target string: - - 0 1 2 3 4 5 6 7 8 910 // string indices - h e l l o w o r l d // string value - ^-------^ // target slice {start: 2, end: 7} - -Scenarios: - -A: ^ // insert "test" at 0 -B: ^ // insert "test" at 4 -C: ^ // insert "test" at 8 - -A = before -> move both by num added -B = inner -> move end by num added -C = after -> noop - -Results: - -A: 0 1 2 3 4 5 6 7 8 910 // string indices - t e s t h e l l o w // string value - ^-------^ // target slice {start: 6, end: 11} - -B: 0 1 2 3 4 5 6 7 8 910 // string indices - h e l l t e s t o w // string value - ^---------------^ // target slice {start: 2, end: 11} - -C: 0 1 2 3 4 5 6 7 8 910 // string indices - h e l l o w o t e s // string value - ^-------^ // target slice {start: 2, end: 7} - -== richTextDelete() - -Target string: - - 0 1 2 3 4 5 6 7 8 910 // string indices - h e l l o w o r l d // string value - ^-------^ // target slice {start: 2, end: 7} - -Scenarios: - -A: ^---------------^ // remove slice {start: 0, end: 9} -B: ^-----^ // remove slice {start: 7, end: 11} -C: ^-----------^ // remove slice {start: 4, end: 11} -D: ^-^ // remove slice {start: 3, end: 5} -E: ^-----^ // remove slice {start: 1, end: 5} -F: ^-^ // remove slice {start: 0, end: 2} - -A = entirely outer -> delete slice -B = entirely after -> noop -C = partially after -> move end to remove-start -D = entirely inner -> move end by num removed -E = partially before -> move start to remove-start index, move end by num removed -F = entirely before -> move both by num removed - -Results: - -A: 0 1 2 3 4 5 6 7 8 910 // string indices - l d // string value - // target slice (deleted) - -B: 0 1 2 3 4 5 6 7 8 910 // string indices - h e l l o w // string value - ^-------^ // target slice {start: 2, end: 7} - -C: 0 1 2 3 4 5 6 7 8 910 // string indices - h e l l // string value - ^-^ // target slice {start: 2, end: 4} - -D: 0 1 2 3 4 5 6 7 8 910 // string indices - h e l w o r l d // string value - ^---^ // target slice {start: 2, end: 5} - -E: 0 1 2 3 4 5 6 7 8 910 // string indices - h w o r l d // string value - ^-^ // target slice {start: 1, end: 3} - -F: 0 1 2 3 4 5 6 7 8 910 // string indices - l l o w o r l d // string value - ^-------^ // target slice {start: 0, end: 5} - */ - -import cloneDeep from 'lodash.clonedeep' -import {AppBskyFeedPost} from '@atproto/api' -import {removeExcessNewlines} from './rich-text-sanitize' - -export type Entity = AppBskyFeedPost.Entity -export interface RichTextOpts { - cleanNewlines?: boolean -} - -export class RichText { - constructor( - public text: string, - public entities?: Entity[], - opts?: RichTextOpts, - ) { - if (opts?.cleanNewlines) { - removeExcessNewlines(this).copyInto(this) - } - } - - clone() { - return new RichText(this.text, cloneDeep(this.entities)) - } - - copyInto(target: RichText) { - target.text = this.text - target.entities = cloneDeep(this.entities) - } - - insert(insertIndex: number, insertText: string) { - this.text = - this.text.slice(0, insertIndex) + - insertText + - this.text.slice(insertIndex) - - if (!this.entities?.length) { - return this - } - - const numCharsAdded = insertText.length - for (const ent of this.entities) { - // see comment at top of file for labels of each scenario - // scenario A (before) - if (insertIndex <= ent.index.start) { - // move both by num added - ent.index.start += numCharsAdded - ent.index.end += numCharsAdded - } - // scenario B (inner) - else if (insertIndex >= ent.index.start && insertIndex < ent.index.end) { - // move end by num added - ent.index.end += numCharsAdded - } - // scenario C (after) - // noop - } - return this - } - - delete(removeStartIndex: number, removeEndIndex: number) { - this.text = - this.text.slice(0, removeStartIndex) + this.text.slice(removeEndIndex) - - if (!this.entities?.length) { - return this - } - - const numCharsRemoved = removeEndIndex - removeStartIndex - for (const ent of this.entities) { - // see comment at top of file for labels of each scenario - // scenario A (entirely outer) - if ( - removeStartIndex <= ent.index.start && - removeEndIndex >= ent.index.end - ) { - // delete slice (will get removed in final pass) - ent.index.start = 0 - ent.index.end = 0 - } - // scenario B (entirely after) - else if (removeStartIndex > ent.index.end) { - // noop - } - // scenario C (partially after) - else if ( - removeStartIndex > ent.index.start && - removeStartIndex <= ent.index.end && - removeEndIndex > ent.index.end - ) { - // move end to remove start - ent.index.end = removeStartIndex - } - // scenario D (entirely inner) - else if ( - removeStartIndex >= ent.index.start && - removeEndIndex <= ent.index.end - ) { - // move end by num removed - ent.index.end -= numCharsRemoved - } - // scenario E (partially before) - else if ( - removeStartIndex < ent.index.start && - removeEndIndex >= ent.index.start && - removeEndIndex <= ent.index.end - ) { - // move start to remove-start index, move end by num removed - ent.index.start = removeStartIndex - ent.index.end -= numCharsRemoved - } - // scenario F (entirely before) - else if (removeEndIndex < ent.index.start) { - // move both by num removed - ent.index.start -= numCharsRemoved - ent.index.end -= numCharsRemoved - } - } - - // filter out any entities that were made irrelevant - this.entities = this.entities.filter(ent => ent.index.start < ent.index.end) - return this - } -} |