mirror of
https://github.com/discourse/discourse.git
synced 2026-08-05 19:38:04 +08:00
Previously, content that was already HTML-escaped — topic `fancy_title`s on the 404/oops pages, GitHub onebox labels, and topic/group/category `<meta>` descriptions — got escaped a second time, so readers (and crawlers) saw literal entities like `’` and `&` instead of `'` and `&`. This change escapes every value exactly once, by converting it to plain text before it reaches the renderer instead of flagging anything `html_safe`: - `Emoji.codes_to_img` escapes its text segments with `html_escape_once` so the already-escaped `fancy_title` and sanitized GitHub labels aren't escaped again. All of its callers render into element content, and the attribute escaping inside `emoji_img_tag` is untouched. - the new `ExcerptParser.to_plain_text` converts a stored excerpt (escaped text, the `…` truncation marker, and the hashtag placeholder markup the parser keeps) back to plain text, exposed as `Topic#plain_text_excerpt`. It's a tag-strip plus a memoized `htmlentities` decode — pure Ruby, a few microseconds per excerpt, no Nokogiri in the hot path. - `Category#plain_text_description` caches the pre-escape plain text that `description_text` previously discarded, and the category meta descriptions use it directly. This also fixes the double-escaped `og:description`/`twitter:description` on category pages, where `gsub_emoji_to_unicode` silently drops the `html_safe` flag of `description_text`. - the new `Group#bio_summary` mirrors `UserProfile#bio_summary`: a plain-text excerpt of the bio with links and images stripped and hashtags rendered as plain `#slug`. Marking the stored values `html_safe` would have worked for the column data — every writer of `topics.excerpt` escapes — but `topic.excerpt` can be swapped in-memory with a localized excerpt that isn't guaranteed to be escaped, so treating everything as unsafe plain text is both simpler and safer. |
||
|---|---|---|
| .. | ||
| plain.html.erb | ||
| show.html.erb | ||
| show.rss.erb | ||