We are now experiencing a much more serious problem. We have a page with 811 distinct usages of syntax highlighting, mostly as inline fragments. It will not render - instead we get a timeout.

"we" ?

To me this sounds like you are missing some of the caching layer optimizations, rather than a problem with this specific extension.

"We" is the organization I work for.

Indeed, our wiki is on a shared host and there are many things we cannot do. The workaround we chose is to only use syntax highlighting for block quotes, and to go back to <code><nowiki> tags for inline (XML) fragments.

But how do you explain the difference between 199 and 200 usages (slow rendering vs. hangup and eventual timeout)?

Just wondering - did you try what I suggested - making a page with 200 copies of <source lang="xml" enclose="none"><Test></source>

Did it work for you?

Tested a bit. It just takes that long to generate the output. The request times out before the server is able to finish it. On reading this isn't too much of a problem, since the server caches that, but a preview is the first render of something, so here your wait time is simply larger than you browser request time.

This also used to happen a lot with complex templates in the past, and here particularly it is caused by the fact that for each individual syntax highlight block, the renderer needs to start up a separate python process to generate the results. It's not a bug, but expected at this moment. But not really ideal either. I'll reword the description of the ticket.

This problem still exist for the latest LTS of MW/extension. I wrote a comment https://www.mediawiki.org/wiki/Topic:Vw4yfp0viknw4ree before i found this ticket. As already mentioned the reason for the slow rendering is that for each code block a separate python process is needed. I wonder if the code could be rewritten in a way that only a single conversion could be performed by merging the input of all blocks before calling the converting only once.

There is a workaround we described in https://github.com/Khan/pygments-server/issues/1. It uses a HTTP server to handle the requests.