[Vision alignment]: Comparative benchmarking - high-speed

[joshbruce]

What other markdown compilers are there for JS now?

Would be worthwhile to put together a list of compilers (markdown-it, for example) and a complex (but sane) markdown file to do comparative benchmarking.

If we aren't faster than them, then we aren't achieving the primary purpose of marked...high-speed.

See also #956

[worker8]

I found this site jsperf.com for comparing javascript library performance.

Someone wrote a test for comparing the performance for a list of markdown compilers:

https://jsperf.com/showdown-js-vs-marked/10

According to the tests, there are a few libraries ranked higher than marked. But without comparing the features of each markdown libs provide, this might not be a fair comparison... Like if they don't provide syntax highlight then for sure it will be faster.

This table can however be a starting point to learn about other markdown libraries.

[Feder1co5oave]

newer version:
https://jsperf.com/showdown-js-vs-marked/19
also:
https://github.com/commonmark/commonmark.js#performance

[calculuschild]

Two things. First, is there a good online benchmark anymore? The jsperf links above seem to be broken.

Second, running the Marked bench script on my own machine is showing Commonmark and Marked-It as both faster and more spec compliant. Can you point to any techniques that used to give you the edge that aren't working anymore, or recent updates that introduced a lot of slowdown? I'd be interested in taking a look.

[UziTech]

I don't know of any online benchmarks.

I think #1686 made a bit of a slow down but also markdown-it has been getting faster with recent releases. I don't know of any simple ways to increase the speed.

[calculuschild]

I think #1686 made a bit of a slow down

I'm sure it did, there was a lot going on in there. I wonder if... there's a way to show kind of a benchmark score per release? Run some kind of automated benchmark on all releases up to this point, and collect that info somewhere useful so it could be plotted and compared, maybe as a page on the documentation?

[UziTech]

We tried to integrate the benchmarks into our testing but test servers are extremely inconsistent

[asbjornu]

This MeasureThat benchmark works (tested in Firefox 131.0.3 on macOS Sonoma 14.7 on a MacBook Pro M2):

The results seem to be rather consistent with Remarkable being the fastest and Remark the slowest, which is a bit surprising given pngwn/MDsveX#20 (comment), but the benchmark may just be using Remark wrong. Remarked ends up in 5th place.