can't post LaTeX

Stream: Zulip meta

[Kevin Buzzard]

So $$q$$ can just be $$a^{-1}+1$$, and the big question is how to make (a) work. Because $$1/(1-ax)$$ is getting bigger as $$x$$ increases to $$a$$ from below, if we can make sure that $$C\leq 1/(1-ap)$$ then this is going to work. Because $$p$$ is just a tiny bit less than $$a^{-1}$$, $$ap$$ will be a tiny bit less than 1, so $$(1-ap)$$ will be positive, so we can multiply both sides by $$(1-ap)$$ and reduce our problem to finding a $$p$$ such that $$C(1-ap)\leq 1$$. Doing the algebra we see $$C-Cap\leq 1$$ so $$C-1\leq Cap$$ and now we have to be a bit careful because we want to divide by $$C$$ to isolate $$p$$ but don't know the sign of $$C$$ so we can't control the inequality.

[Kevin Buzzard]

If I view source of that message, remove the triple quotes, and just try and post the LaTeX, then with 100% reliability the post fails for me (I get the red error icons like you do if you're on a train and lose connection)

[Kevin Buzzard]

here I split the paragraph into two, and I manage to send the message successfully.

[Kevin Buzzard]

On Ubuntu 22.04, using the desktop app.

[Alex Vandiver]

We have some performance problems rendering messages with a large number of separate $\LaTeX$ blocks in them; as you've found, splitting it into two messages is a workaround.

I expect we'll have a solution for this in the next month or so. https://github.com/zulip/zulip/issues/17425 is the bug to follow for updates.

[Kevin Buzzard]

yeah I opened that in 2021 :-) The situation is much better now -- when the post failed today I could still access the post in drafts so I could copy-paste to a file and not lose it; I just reposted it in chunks. But conversely my paragraph above was much smaller than the large multi-paragraph messages I was struggling with in 2021.

[Alex Vandiver]

The raw number of $$foo$$ blocks is the relevant limitation. The short summary is that it takes some time to render each one, and we time out trying to render the message if we take too long.

[Alex Vandiver]

I count 18 $\LaTeX$ blocks in there. My suspicion is that the paragraphs in 2021 may have been longer, but may not have had more $\LaTeX$ blocks in them.

[Alex Vandiver]

Hm. I guess your example in the 2021 bug report, reproduced below, has more than 18 blocks, and you did get it to render eventually:
109338371-18026080-785e-11eb-9f1b-97ca11c4fdba.png

[Alex Vandiver]

So there may be a performance regression in this codepath which is limiting how many "fit" before it times out.

[Alex Vandiver]

Tangentially, https://github.com/zulip/zulip/issues/28626 to make it easier to compose several messages, to be sent back-to-back, by splitting at paragraph breaks, is also making some progress.

[Alex Vandiver]

Alex Vandiver said:

I expect we'll have a solution for this in the next month or so. https://github.com/zulip/zulip/issues/17425 is the bug to follow for updates.

This is now live. Taking your above example:

So $q$ can just be $a^{-1}+1$, and the big question is how to make (a) work. Because $1/(1-ax)$ is getting bigger as $x$ increases to $a$ from below, if we can make sure that $C\leq 1/(1-ap)$ then this is going to work. Because $p$ is just a tiny bit less than $a^{-1}$, $ap$ will be a tiny bit less than 1, so $(1-ap)$ will be positive, so we can multiply both sides by $(1-ap)$ and reduce our problem to finding a $p$ such that $C(1-ap)\leq 1$. Doing the algebra we see $C-Cap\leq 1$ so $C-1\leq Cap$ and now we have to be a bit careful because we want to divide by $C$ to isolate $p$ but don't know the sign of $C$ so we can't control the inequality.

[Kevin Buzzard]

Thanks!

[Alex Vandiver]

Rendering $\LaTeX$ should now be fast enough that I do not expect messages with even copious amounts of it to fail to render. Please bring it to our attention if you're still seeing any failures of that form.