Long messages which respond to another message get split incorrectly

[tgr]

When writing a message that's longer than the IRC limit, while responding to another message, the bridge will prefix some words from the replied-to message (ie. it turns the response relationship into a quote) and splits the message into shorter chunks that do not get truncated on IRC. It seems the chunk calculation ignores that the quotation made the message longer, though, so you end up with some truncation happening in the middle:


(link to original comment)

[Half-Shot]

Hey @tgr, thanks for reporting. I'm finding it difficult to reproduce this bug, could you send me an example message and reply text so I can reproduce it locally?

[tgr]

Sorry for the slow response! Reproduction steps:

• Visit #wikipedia-hu:matrix.org (which is manually bridged to #wikipedia-hu@Freenode) in Riot-web and an IRC client (I've been using irssi).
• In Riot-web, click "Reply" next to this message.
• Type in grin: alapvetően itt az a nehézség, hogy az IRC szar, a Matrix meg nem az. Ha egy másik olyan rendszerrel kéne illeszteni, ami tud rich text tartalmat kezelni, szoba történetét valamilyen strukturált módon megjeleníteni, lehet hozzászólásokat szerkeszteni, vannak benne normális identitások, rendes topic API stb. stb. akkor normálisabban működne. De vízözön előtti technológia (régebbi, mint a web), úgyhogy ez van. Ezért is lenne fontos, hogy a Wikimédia mozgalom átálljon valamire, amitől egy mai felhasználó nem szalad el sikítva. Én reménykedem, hogy a magyar wiki és chapter ebben úttörő szerepet tud játszani. as the reply, and send. (The name at the beginning is via tab-completion for grin@grin.hu.)
• View the message in the IRC client.

This time, it failed in a somewhat different way:

(Also all the non-ASCII characters are broken, which I assume is an artifact of the message being truncated in the middle of an UTF-8 byte sequence.)

I tried to reproduce with a fresh message too, but that worked fine. So my original assumption that this was caused just by the combination of quoting + long message seems to be incorrect. Which also means I don't have an idea what the local reproduction steps would be :(

[Half-Shot]

Thanks @tgr, and yes it might be that we aren't splitting sensible for non-ASCII, I remember there are quirks in JS around multi-byte characters. I can at least take a look at that one, but I think it may be difficult to reproduce your original bug, so will see what I can achieve.

[Half-Shot]

This may actually be #706