Reduce bundle size

[gnestor]

Follow up from #5722 (comment)

• xterm could be lazy loaded
• santitize-html too big
  • What's it doing with postcss?
• moment is too big
  • Import submodule
• vim keymap for codemirror
  • Lazy load keymaps
• do we need esprima? Can we lazy-load it?
• ajv could be lazy-loaded

cc @ian-r-rose

[gnestor]

How to lazy load large libraries in lab extensions: jupyterlab/jupyterlab-latex#107

[bollwyvl]

An example up for xterm on #5994. Not great, but does save ~350kb

Sanitizer is hard, as its API would have to change, as it is currently synchronous.

I didn't realize how brutal the blueprint stuff really is: because it's used by every toolbar, there's no point in moving it out of the vendors-main monolith.

Pretty late in the game at this point, but I feel like we either need to ramp up our usage to use every single bell and whistle in there... or forbid use of it in the critical path (any sidebar, menu, launcher, etc.), and figure out a way for them to get them conveniently lazy loaded.

[ian-r-rose]

Yeah, sanitizer it tough. A shame postcss is so large.

ajv may be doable. The settings API is mostly async, with just one synchronous function that would need to be a breaking change.

[bollwyvl]

The settings API is mostly async, with just one synchronous function that would need to be a breaking change.

Is it even worth it if it has to be loaded every time? Does settings loading block any UI drawing? If so, there's no real benefit, as it's an extra connection. The only place we can really win for "clean" first start is things that won't be used at all until you decide what you are going to do: codemirror actually would be a candidate, and we as sure as shooting would want to not be loading monaco in the hot path.

[bollwyvl]

I had a look at moment: we're only using two functions from moment (formatting and relative time). We already have date-fns in the lockfile, so i took a look at that. Because of how we vendor, it won't tree-shake out, but it's still a good deal smaller... will try to PR if I get a chance...

[ghost]

Can't personally vouch for it but have read about Day.Js as an alternative, formatting and relative time just cost 2kb+

⏰ Day.js 2KB immutable date library alternative to Moment.js with the same modern API

https://github.com/iamkun/dayjs

reference: moment/moment#2373 (comment)

[vidartf]

Somebody mentioned previously that moment.js could be optimized by using submodule imports. Is this the case?

[bollwyvl]

Somebody mentioned previously that moment.js could be optimized by using

submodule imports. Is this the case?.

Not really: everything is chained off a moment instance, so there's no way to get just the methods you want. It is possible to get not-stock builds. https://medium.com/@k2u4yt/momentjs-vs-date-fns-6bddc7bfa21e A huge whammy is the locale information, and we don't l10n, so it's not worth it... yet. I'm not down on moment (did some of the iso 8601 duration stuff), and have used it in extensions, but we're presently just not using it much. Day.js seems interesting, but I was just looking for something we (or one of our dependencies) already nominally "trust" and therefore would already have lying around, instead of adding new dependencies.