Kevin Thomas Van Cott @KevinVanCott

Sometimes called Kevin Vandy. Software Engineer and OSS maintainer. Creator of material-react-table and mantine-react-table. TanStack maintainer. kevinvancott.dev Nebraska Joined November 2012

Tweets

4K
Followers

3K
Following

856
Likes

16K

Kevin Thomas Van Cott @KevinVanCott

a day ago

@Devinda_me it takes a lot of time (2-3 days of my free time), but it's also forcing me to create real benchmark testing.

0 0 1 10 0

View Details

Kevin Thomas Van Cott @KevinVanCott

2 days ago

Now that I've written a blog post 3 weeks in a row, I kinda just want to keep publishing one every week

2 0 13 604 0

View Details

Kevin Thomas Van Cott @KevinVanCott

3 days ago

You can definitely do that now if it makes sense. The row re-ordering example just doesn't really need that. All it needs is a `setData` call when the data is re-ordered, which shouldn't really be attached to the table. Or at least, I don't see a reason to. Data comes before the useTable definition. It's up to you to decide where it makes sense for your composable code to live. It can get added to the TanStack Table API if you want, or you can just write composable code along-side TanStack Table.

0 0 1 18 0

View Details

Kevin Thomas Van Cott @KevinVanCott

3 days ago

@culer_1997 @GregHoarau @sadmann17 @shadcn @tan_stack @partykit_io what part needs to be a custom feature on the table API?

1 0 0 40 0

View Details

Kevin Thomas Van Cott @KevinVanCott

4 days ago

@Revenode @ryanconnaughton I have an ongoing perf.md file where I'm trying to document most of the optimizations that I've been exploring during v9 development. That might be interesting for you to see if you can add on to. github.com/TanStack/table…

1 0 2 44 0

View Details

Kevin Thomas Van Cott @KevinVanCott

5 days ago

The blog post is out! Here's the surprisingly simple way that TanStack Table V9 can use a little as 10% memory as it used to. Hint: it takes advantage of Object prototypes tanstack.com/blog/tanstack-…

Kevin Thomas Van Cott @KevinVanCott

a week ago

3 12 124 29K 36

9 10 175 21K 86

View Details

Kevin Thomas Van Cott @KevinVanCott

4 days ago

Wow, I didn't realize how much faster Object.create could also be. I didn't measure that at all. Although, we're not replacing class constructors, we're just modifying how we create a new javascript object like `const foo = {...}`, so maybe the difference wouldn't have been as large for us.

0 0 0 33 0

View Details

Kevin Thomas Van Cott @KevinVanCott

4 days ago

There was a similar question that I answered here. I think there is a somewhat hacky way we could do class extending, but also, we'd pretty much need the javacript classes to have "multiple inheritance". Like we'd need to extend 12+ classes for each feature. Not to mention trying to support custom plugins like we currently do. x.com/KevinVanCott/s…

Kevin Thomas Van Cott @KevinVanCott

4 days ago

@MattiasBuelens In a normal library, using JavaScript classes makes sense. But for TanStack Table V9, all methods on these objects are created conditionally. Like we only want to have the `row.getVisibleCells()` method exist only if you are using the columnVisiblilityFeature. Or we only want to

0 0 4 596 0

1 0 0 21 0

View Details

Kevin Thomas Van Cott @KevinVanCott

4 days ago

@Revenode @ryanconnaughton I'm curious, when you say "object spread", are you talking about either of these 2 kinds of code snippets? Or something else?

1 0 0 26 0

View Details

Kevin Thomas Van Cott @KevinVanCott

4 days ago

I should probably update the article to discuss "Why not use a JavaScript class instead?" Here's my answer: x.com/KevinVanCott/s…

Kevin Thomas Van Cott @KevinVanCott

4 days ago

0 0 4 596 0

0 0 3 501 0

View Details

Kevin Thomas Van Cott @KevinVanCott

4 days ago

@Revenode @ryanconnaughton yeah, if you find further improvement on this, I'd be very interested.

1 0 1 21 0

View Details

Kevin Thomas Van Cott @KevinVanCott

4 days ago

Yeah, I've ripped out a lot of "readable" code out of the TanStack Table codebase recently, actually. All in the name of performance. Next week, my next article might be about the overall performance improvements from those refactors. Some refactors were a big deal, and some were more on the micro-optimization side. But micro-optimizations that scale to millions of potential operations can actually have a small noticeable when all added up. I've gotten rid of most `.map`, `.filter`, `.reduce`, and recursion patterns in our library code in favor of a lot more plain `for` loops with many more Sets and Map lookups. And actually, I suspect some of our memory usage savings can be attributed to the amount of closures that these refactors eliminated. As a preview, sorting performance is up to 2x faster than V8, but I haven't developed full benchmarks for that yet.

1 0 0 54 0

View Details

Kevin Thomas Van Cott @KevinVanCott

4 days ago

@ezra_yisma I've definitely seen people have many use cases where they were fetching hundreds of thousands of rows for their tables. I usually don't recommend doing that, but if it works and they don't have to implement server-side logic for all the table stuff, then it can be that way.

0 0 2 39 0

View Details

Kevin Thomas Van Cott @KevinVanCott

4 days ago

In a normal library, using JavaScript classes makes sense. But for TanStack Table V9, all methods on these objects are created conditionally. Like we only want to have the `row.getVisibleCells()` method exist only if you are using the columnVisiblilityFeature. Or we only want to have the `column.toggleSort()` method exist if you are using the rowSortingFeature. I don't know of a good dynamic way to create classes like that. When we make the objects and prototypes manually, we have full control to make these objects dynamically.

0 0 4 596 0

View Details

Kevin Thomas Van Cott @KevinVanCott

6 days ago

@culer_1997 @GregHoarau @sadmann17 @shadcn @tan_stack @partykit_io Yes, in the v9 beta docs

2 0 3 82 0

View Details

Kevin Thomas Van Cott @KevinVanCott

7 days ago

@GregHoarau @sadmann17 @shadcn @tan_stack @partykit_io V9 has an official plugin/custom feature system so that it's much easier for you to extend TanStack Table with custom features like that.

1 0 3 98 0

View Details

Kevin Thomas Van Cott @KevinVanCott

7 days ago

@theCTO 30 million downloads each not enough for you? Not necessarily resting in peace right now.

1 0 22 4K 0

View Details

Kevin Thomas Van Cott @KevinVanCott

7 days ago

@getpochi ok chatgpt

0 0 2 161 0

View Details

Kevin Thomas Van Cott @KevinVanCott

a week ago

Up to 90% less memory usage? I'm working on a new blogpost to examine the performance improvements of TanStack Table V9 and first up is memory usage. I just created some memory usage benchmarks for TanStack Table V9 vs V8, and it verifies that Table V9 may use up to 90% less memory than V8 on our example with 100k+ virtualized rows. This is kind of a best-case scenario, so I'll need to benchmark even more common table setups in order to be as transparent as possible. This improvement is due in large part to our refactor where we store object functions on a single prototype instead of creating them for every object. APIs like row.getVisibleCells, column.toggleSorting(), etc... now all defined just once on `table._rowPrototype`, `table._columnPrototype`, etc. When I first attempted this refactor, I was worried that it might be a waste of time for an improvement that could be negligible. For common smaller tables, it still kind of is, but for large, virtualized tables where the number of objects created scales linearly with memory usage, it's seeming to make all the difference.