Fixes#395
Instead of leaving the toot body blank, it replaces it with a single “.” in
order for the fold/unfold CW behavior to not look *too* weird on upstream
Mastodon. Note that this does not fix upstream's CW-dropping behavior, as
that is decided at the time the toot is posted, not received.
Indeed, when the “Always enable the Content Warning field” setting is enabled,
sending a content-less toot with a CW would move the CW to the toot's content
and leave the toot not marked as sensitive.
* Downcase signed_headers string before building the signed string
The HTTP Signatures draft does not mandate the “headers” field to be downcased,
but mandates the header field names to be downcased in the signed string, which
means that prior to this patch, Mastodon could fail to process signatures from
some compliant clients. It also means that it would not actually check the
Digest of non-compliant clients that wouldn't use a lowercased Digest field
name.
Thankfully, I don't know of any such client.
* Revert "Remove dead code (#8919)"
This reverts commit a00ce8c92c.
* Restore time window checking, change it to 12 hours
By checking the Date header, we can prevent replaying old vulnerable
signatures. The focus is to prevent replaying old vulnerable requests
from software that has been fixed in the meantime, so a somewhat long
window should be fine and accounts for timezone misconfiguration.
* Escape users' URLs when formatting them
Fixes possible HTML injection
* Escape all string interpolations in Formatter class
Slightly improve performance by reducing class allocations
from repeated Formatter#encode calls
* Fix code style issues
* Add conversations API
* Add web UI for conversations
* Add test for conversations API
* Add tests for ConversationAccount
* Improve web UI
* Rename ConversationAccount to AccountConversation
* Remove conversations on block and mute
* Change last_status_id to be a denormalization of status_ids
* Add optimistic locking
Media modals push an history state so that pressing back on mobile closes them.
We made sure to not scroll when opening them, but not when *closing* them,
which caused some issues in rare cases.
