It's not about being picky. It's about communicating needs, and setting boundaries that are designed to satisfy those needs without overwhelming anybody's system to the point of saturation and degraded performance.
In the context of this blog post, what if the SLA was <100ms for an initial response, with some mandatory fields, but then any additional information which happens to be loaded within that 100ms automatically is included. With anything outside the 100ms is automatically sent in a followup message?
From wikipedia, "ndjson" used to include single-line comments with "//" and needed custom parsers for it, but the spec no longer includes it. So now they are the same.
https://github.com/ndjson/ndjson.github.io/issues/1#issuecom...
None of the above is actually good enough to build on, so a thousand little slightly-different ad hoc protocols bloom. For example, is empty line a keepalive or an error? (This might be perfectly fine. They're trivial to program, not like you need a library.)
I like to represent tree data with parent, type, and data vectors along with a string table, so everything else is just small integers.
Sending the string table and type info as upfront headers, we can follow with a stream of parent and data vector chunks, batched N nodes at a time. Tye depth- or breadth-first streaming becomes a choice of ordering on the vectors.
I'm gonna have to play around with this! Might be a general way to get snappier load time UX on network bound applications.
But, indeed, depth vectors are nice and compact. I find them harder to work with most of the time, though, especially since insertions and deletions become O(n), compared to parent vector O(1).
That said, I do often normalize my parent vectors into dfpo order at API boundaries, since a well-defined order makes certain operations, like finding leaf siblings, much nicer.
I think you can still have the functionality described in the article: you would send “hole” markers tagged with their level. Then, you could make additional requests when you encounter these markers during the recovery phase, possibly with buffering of holes. It becomes a sort of hybrid DFS/BFS approach where you send as much tree structure at a time as you want.
Better to rethink it from scratch instead of trying to put a square peg in a round hog.
Seems like it is never about merit of technological design. As some CS professor put it, tech is more about fashion than tech now days. IMHO that is true, and often also comes down to the technological context surrounding the industry at the time, and now days if the code is open sourced/FOSS.
* I made it up - and by extension, the status quo is 'correct'.
Having progressive or partial reads would dramatically speed up applications, especially as we move into an era of WASM on the frontend.
A proper binary encoded format like protobuf with support for partial reads and well defined streaming behavior for sub message payloads would be incredible.
It puts more work on the engineer, but the improvement to UX could be massive.
Following the example, why is all the data in one giant request? Is the DB query efficient? Is the DB sized correctly? How about some caching? All boring, but if rather support and train someone on boring stuff.
There is something wrong with adding a "fancy" feature to an off-the-shelf option, if said "fancy" feature is realistically "a complicated engineering question, for which we can offer a leaky abstraction that will ultimately trip up anybody who doesn't have the actual mechanics in mind when using it".
Your comment focuses on desired outcomes (i.e., "nice" things), but fails to acknowledge the reality of tradeoffs. Over engineering a solution always creates problems. Systems become harder to reason with, harder to maintain, harder to troubleshoot. For example, in JSON arrays are ordered lists. If you onboard an overengineered tool that arbitrarily reorders elements in a JSON array, things can break in non-trivial ways. And they often do.
I think the issue with the example json is that it's sent in OOP+ORM style (ie nested objects), whereas you could just send it as rows of objects, something like this;
{
header: "Welcome to my blog",
post_content: "This is my article",
post_comments: [21,29,88], # the numbers are the comment ID's
footer: "Hope you like it",
comments: {21: "first", 29: "second", 88: "third" }
}
syntax = "proto3";
service DirectiveAffectsService {
rpc Get(GetPageWithPostParams) returns (PageWithPost);
}
message GetPageWithPostParams {
string post_id = 1;
}
message PageWithPost {
string page_header = 1;
string page_footer = 2;
string post_content = 3;
repeated string post_comments = 4;
repeated CommentInPost comments_for_post = 5;
}
message CommentInPost {
string comment_id = 1;
string comment_text = 2;
}
GraphQL could use Progressive JSON to serialize subscriptions.
For @defer, the transport is a stream[2] where subsequent payloads are tracked ("pending") and as those payloads arrive they are "patched" into the existing data at the appropriate path.
Both of these are effectively implementations of this Progressive JSON concept.
I can picture a future where both the JSX and object payloads are delivered progressively over the wire in the same stream, and hydrate React and Relay stores alike. That could potentially simplify and eliminate the need for Relay to synchronise it's own updates with React. It would also potentially be a suitable compile target for decoupling, React taking ownership of a normalised reactive object store (with state garbage collection which is a hard problem for external stores to solve) with external tools like Relay and TanStack Query providing the type-generation & fetching implementation details.
Using `new Promise()` in the client-side store would also mean you could potentially shrink the component API to this. `useFragment(fragment IssueAssignee on Issue { assignee { name }}, issue)` could instead be `const assignee = use(issue.assignee)` and would suspend (or not) appropriately.
[1]: https://github.com/facebook/relay/blob/main/packages/relay-r...
[2]: https://github.com/graphql/graphql-wg/blob/main/rfcs/DeferSt...
I got really, really sick of XML, but one thing that XML parsers have always been good at, is realtime decoding of XML streams.
It is infuriating, waiting for a big-ass JSON file to completely download, before proceeding.
Also JSON parsers can be memory hogs (but not all of them).
I’ve written a lot of APIs. I generally start with CSV, convert that to XML, then convert that to JSON.
CSV is extremely limited, and there’s a lot of stuff that can only be expressed in XML or JSON, but starting with CSV usually enforces a “stream-friendly” structure.
* I do acknowledge you qualified the question with "better".
The type and value are encoded the same as DER, but the length is different:
- If it is constructed, the length is omitted, and a single byte with value 0x00 terminates the construction.
- If it is primitive, the value is split into segments of lengths not exceeding 255, and each segment is preceded by a single byte 1 to 255 indicating the length of that segment (in bytes); it is then terminated by a single byte with value 0x00. When it is in canonical form, the length of segments other than the last segment must be 255.
Protobuf seems to not do this unless you use the deprecated "Groups" feature, and this is only as an alternative of submessages, not for strings. In my opinion, Protobuf also seems to have many other limits and other problems, that DER (and DSER) seems to do better anyways.
[ ["aaa", "bbb"], { "name", "foo" } ]
Start array
Start array
String aaa
String bbb
End array
Start object
Key name
String foo
End object
End arrayYou can do that, in a specialized manner, with PHP, and Streaming JSON Parser[0]. I use that, in one of my server projects[1]. It claims to be JSON SAX, but I haven’t really done an objective comparison, and it specializes for file types. It works for my purposes.
[0] https://github.com/salsify/jsonstreamingparser
[1] https://github.com/LittleGreenViper/LGV_TZ_Lookup/blob/main/...
In response to "Json is just a packing format that does have that [streaming] limitation".
?page=3&size=100
Yes, breadth-first is always an option, but JSON is a heterogenous structured data source, so assuming that breadth-first will help the app start rendering faster is often a poor assumption. The app will need a subset of the JSON, but it's not simply the depth-first or breadth-first first chunk of the data set.
So for this reason what we do is include URLs in JSON or other API continuation identifiers, to let the caller choose where in the data tree/graph they want to dig in further, and then the "progressiveness" comes from simply spreading your fetch operation over multiple requests.
Also often times JSON is deserialized to objects so depth-frst or breadth-first doesn't matter, as the object needs to be "whole" before you can use it. Hence again: multiple requests, smaller objects.
In general when you fetch JSON from a server, you don't want it to be so big that you need to EVEN CONSIDER progressive loading. HTML needs progressive loading because a web page can be, historically especially, rather monolithic and large.
But that's because a page is (...was) static. Thus you load it as a big lump and you can even cache it as such, and reuse it. It can't intelligently adapt to the user and their needs. But JSON, and by extension the JavaScript loading it, can adapt. So use THAT, and do not over-fetch data. Read only what you need. Also, JSON is often not cacheable as the data source state is always in flux. One more reason not to load a whole lot in big lumps.
Now, I have a similar encoding with references, which results in a breadth-first encoding. Almost by accident. I do it for another reason and that is structural sharing, as my data is shaped like a DAG not like a tree, so I need references to encode that.
But even though I have breadth-first encoding, I never needed to progressively decode the DAG as this problem should be solved in the API layer, where you can request exactly what you need (or close to it) when you need it.
Right. Closer to the end of the article I slightly pivot to talk about RSC. In RSC, the data is the UI, so the outermost data literally corresponds to the outermost UI. That's what makes it work.
It's encoded like progressive JSON but conceptually it's more like HTML. Except you can also have your own "tags" on the client that can receive object attributes.
Not again, please.
This is more of a post on explaining the idea of React Server Components where they represent component trees as javascript objects, and then stream them on the wire with a format similar to the blog post (with similar features, though AFAIK it’s bundler/framework specific).
This allows React to have holes (that represent loading states) on the tree to display fallback states on first load, and then only display the loaded component tree afterwards when the server actually can provide the data (which means you can display the fallback spinner and the skeleton much faster, with more fine grained loading).
(This comment is probably wrong in various ways if you get pedantic, but I think I got the main idea right.)
do you think a new data serialization format built around easier generation/parseability and that also happened to be streamable because its line based like jsonld could be useful for some?
What immediately comes to mind is using a uniform recursive tree instead, where each node has the same fields. In a funny way that would mimic the DOM if you squint. Each node would encode it's type, id, name, value, parent_id and order for example. The engine in front can now generically put stuff into the right place.
I don't know whether that is feasible here. Just a thought. I've used similar structures in data driven react (and other) applications.
It's also efficient to encode in memory, because you can put this into a flat, compact array. And it fits nicely into SQL dbs as well.
As of right now, I could only replace the JSON tool calling on LLM's on something I fully control like vLLM, and the big labs probably are happy to over-charge a 20-30% tokens for each tool call, so they wouldn't really be interested on replacing json any time soon)
also it feels like battling against a giant which is already an standard, maybe there's a place for it on really specialized workflows where those savings make the difference (not only money, but you also gain a 20-30% extra token window, if you don't waste it on quotes and braces and what not
Thanks for replying!
It’s become a thing, even beyond RSCs, and has many practical uses if you stare at the client and server long enough.
I've been trying to create go/rust ones but its way harder than just json due to all the context/state they carry over
From an outsider's perspective, if you're sending around JSON documents so big that it takes so long to parse them to the point reordering the content has any measurable impact on performance, this sounds an awful lot like you are batching too much data when you should be progressively fetching child resources in separate requests, or even implementing some sort of pagination.
And the most annoying antipattern is showing empty state UI during loading phase.
Quoting the article:
> You don’t actually want the page to jump arbitrarily as the data streams in. For example, maybe you never want to show the page without the post’s content. This is why React doesn’t display “holes” for pending Promises. Instead, it displays the closest declarative loading state, indicated by <Suspense>.
> In the above example, there are no <Suspense> boundaries in the tree. This means that, although React will receive the data as a stream, it will not actually display a “jumping” page to the user. It will wait for the entire page to be ready. However, you can opt into a progressively revealed loading state by wrapping a part of the UI tree into <Suspense>. This doesn’t change how the data is sent (it’s still as “streaming” as possible), but it changes when React reveals it to the user.
[…]
> In other words, the stages in which the UI gets revealed are decoupled from how the data arrives. The data is streamed as it becomes available, but we only want to reveal things to the user according to intentionally designed loading states.
https://www.haskellpreneur.com/articles/slaying-a-ui-antipat...
Since React is functional programming it works well with parallelization so there is room for experiments.
> Especially if it involves content jumping around.
I remember this from the beginning of Android, you'll search for something and click on it and the time it takes you to click the list of results changed and you clicked on something else. Happens with adds on some websites, maybe intentionally?
> And the most annoying antipattern is showing empty state UI during loading phase.
Some low quality software even show "There are no results for your search" when the search didn't even start or complete.
It’s been so long since I used ember that I’ve forgotten the terms, but essentially the rearranged the tree structure so that some of the children were at the end of the file. I believe it was meant to handle DAGs more efficiently but I may have hallucinated that recollection.
But if you’re using a SAX style streaming parser you can start making progress on painting and perhaps follow-up questions while the initial data is still loading.
Of course in a single threaded VM, you can snatch Defeat from the jaws of Victory if you bollocks up the order of operations through direct mistakes or code evolution over time.
In .NET land, Utf8JsonReader is essentially this idea. You can parse up until you have everything you need and then bail on the stream.
https://learn.microsoft.com/en-us/dotnet/standard/serializat...
I suppose it's because doing it breadth-first means you need to come up with a way to reference items that will arrive many lines later, whereas you don't have that need with depth-first serialisation.
I think that's a very contemporary problem and worth pursuing, but I also somehow won't see that happening in real-time (with the priority to reduce latency) without necessary metadata.
Progressively loaded JPEGs just apply some type of "selective refinement" to chunks of data, and for Progressive selective refinement to work it's necessary to "specify the location and size of the region of one or more components prior to the scan"[0][1]. If you don't know what size to allocate, then it's quite difficult(?) to optimize the execution. This doesn't seem like the kind of discussion you'd like to have.
Performance aware web developers are working with semantic awareness of their content in order to make tweaks to the sites loading time. YouTube might prefer videos (or ads) to be loaded before any comments, news sites might prioritize text over any other media, and a good dashboard might prioritize data visualizations before header and sidebar etc.
The position of the nodes in any structured tree tells you very little about the preferred loading priority, wouldn't you agree?
[0] https://jpeg.org/jpeg/workplan.html
[1] https://www.itu.int/ITU-T/recommendations/rec.aspx?id=3381 (see D.2 in the PDF)
EDIT: Btw thanks for your invaluable contributions to react (and redux back then)!
RSC streams outside-in because that's the general shape of the UI — yes, you might want to prioritize the video, but you have to display the shell around that video first. So "outside-in" is just that common sense — the shell goes first. Other than that, the server will prioritize whatever's ready to be written to the stream — if we're not blocked on IO, we're writing.
The client does some selective prioritization on its own as it receives stuff (e.g. as it loads JS, it will prioritize hydrating the part of the page that you're trying to interact with).
Personally I prefer that sort of approach - parsing a line of JSON at a time and incrementally updating state feels easier to reason and work with (at least in my mind)
- Sending a request for posts, then a request for comments, resulting in multiple round trips (a.k.a. a "waterfall"), or,
- Sending a request for posts and comments, but having to wait until the commends have loaded to get the posts,
...you can instead get posts and comments available as soon as they're ready, by progressively loading information. The message, though, is that this is something a full-stack web framework should handle for you, hence the revelation at the end of the article about it being a lesson in the motivation behind React's Server Components.
What is innovative trying to build a framework that does it for you.
Progressive loading is easy, but figuring out which items to progressively load and in which order without asking the developer/user to do much extra config is hard.
Do developers even control the order in which stuff is loaded? Tha depends on factors beyond a developer's control, such as the user's network speed, the origin server's response speed, which resources are already cached, how much data each request fetches for user A or user B, etc.
It's a solved problem. Use HTTP/2 and keep the connection open. You now have effectively a stream. Get the top-level response:
{
header: "/posts/1/header",
post: "/posts/1/body",
footer: "/posts/1/footer"
}
This still involves multiple round-trips though. The approach laid out in the article lets you request exactly the data you need up-front and the server streams it in as it becomes available, e.g. cached data first, then data from the DB, then data from other services, etc.
However, sequentially-dependent requests are about as slow with HTTP/2 as HTTP/1.1. For example, if your client side, after loading the page, requests data to fill a form component, and then that data indicates a map location, so your client side requests a map image with pins, and then the pin data has a link to site-of-interest bubble content, and you will be automatically expanding the nearest one, so your client side requests requests the bubble content, and the bubble data has a link to an image, so the client requests the image...
Then over HTTP/2 you can either have 1 x round trip time (server knows the request hierarchy all the way up to the page it sends with SSR) or 5 x round trip time (client side only).
When round trip times are on the order of 1 second or more (as they often are for me on mobile), >1s versus >5s is a very noticable difference in user experience.
With lower latency links of 100ms per RTT, the UX difference between 100ms and 500ms is not a problem but it does feel different. If you're on <10ms RTT, then 5 sequential round trips are hardly noticable, thought it depends more on client-side processing time affecting back-to-back delays.
https://github.com/rgraphql/rgraphql
But it was too graphql-coupled and didn't really take off, even for my own projects.
But it might be worth revisiting this kind of protocol again someday, it can tag locations within a JSON response and send updates to specific fields (streaming changes).
Compared with waiting on a blank page for ages, sometimes it's nice to see text content if it's useful, and to be able to click navigation links early. It's much better than pages which look like they have finished loading but important buttons and drop-downs are broken without any visible indication because there's JS still loading in the background. I'm also not fond of pages where you can select options and enter data, and then a few seconds after you've entered data, all the fields reset as background loading completes.
All the above are things I've experienced in the last week.
We tried to do that with GraphQL, http2,... And arguably failed. Until we can properly evolve web standards we won't be able to fix the main issue. Novel frameworks won't do it either
The idea is that every frontend has a dedicated backend with exactly the api that that frontend needs.
If you're optimizing for time to first render, or time to visually complete, then you need to render the page using as little logic as possible - sending an empty skeleton that then gets hydrated with user data over APIs is fastest for a user's perception of loading speed.
If you want to speed up time to first input or time to interactive you need to actually build a working page using user data, and that's often fastest on the backend because you reduce network calls which are the slowest bit. I'd argue most users actually prefer that, but it depends on the app. Something like a CRUD SAAS app is probably best rendered server side, but something like Figma is best off sending a much more static page and then fetching the user's design data from the frontend.
The idea that there's one solution that will work for everything is wrong, mainly because what you optimise for is a subjective choice.
And that's before you even get to Dev experience, team topology, Conway's law, etc that all have huge impacts on tech choices.
I think that OP's point is that these optimization strategies are completely missing the elephant in the room. Meaning, sending multi-MB payloads creates the problem, and shaving a few ms here and there with more complexity while not looking at the performance impact of having to handle multi-MB payloads doesn't seem to be an effective way to tackle the problem.
This is often repeated, but my own experience is the opposite: when I see a bunch of skeleton loaders on a page, I generally expect to be in for a bad experience, because the site is probably going to be slow and janky and cause problems. And the more the of the site is being skeleton-loaded, the more my spirits worsen.
My guess is that FCP has become the victim of Goodhart's Law — more sites are trying to optimise FCP (which means that _something_ needs to be on the screens ASAP, even if it's useless) without optimising for the UX experience. Which means delaying rendering more and adding more round-trips so that content can be loaded later on rather than up-front. That produces sites that have worse experiences (more loading, more complexity), even though the metric says the experience should be improving.
I think it's more the bounce rate is improving. People may recall a worse experience later, but more will stick around for that experience if they see something happen sooner.
The number of websites needlessly forced into being SPAs without working navigation like back and forth buttons is appalling.
It’s only fastest to get the loading skeleton onto the page.
My personal experience with basically any site that has to go through this 2-stage loading exercise is that:
- content may or may not load properly.
- I will probably be waiting well over 30 seconds for the actually-useful-content.
- when it does all load, it _will_ be laggy and glitchy. Navigation won’t work properly. The site may self-initiate a reload, button clicks are…50/50 success rate for “did it register, or is it just heinously slow”.
I’d honestly give up a lot of fanciness just to have “sites that work _reasonably_” back.
- FE is short for the Front End (UI)
- BFF is short for Backend For Frontend
One example a programmer might understand - rather than needing to send the grammar and code of a syntax highlighter to the frontend to render formatted code samples, you can keep that on the backend, and just send the resulting HTML/CSS to the frontend, by making sure that you use your syntax highlighter in a server component instead of a client component. All in the same language and idioms that you would be using in the frontend, with almost 0 boilerplate.
And if for some reason you decide you want to ship that to the frontend, maybe because you want a user to be able to syntax highlight code they type into the browser, just make that component be a client component instead of a server component, et voila, you've achieved it with almost no code changes.
Imagine what work that would take if your syntax highlighter was written in Go instead of JS.
Dan is one of the best explainers in React ecosystem but IMO if one has to work this hard to sell/explain a tech there's 2 possibilities 1/ there is no real need of tech 2/ it's a flawed abstraction
#2 seems somewhat true because most frontend devs I know still don't "get" RSC.
Vercel has been aggressively pushing this on users and most of the adoption of RSC is due to Nextjs emerging as the default React framework. Even among Nextjs users most devs don't really seem to understand the boundaries of server components and are cargo culting
That coupled with fact that React wouldn't even merge the PR that mentions Vite as a way to create React apps makes me wonder if the whole push for RSC is for really meant for users/devs or just as a way for vendors to push their hosting platforms. If you could just ship an SPA from S3 fronted with a CDN clearly that's not great for Vercels and Netflifys of the world.
In hindsight Vercel just hiring a lot of OG React team members was a way to control the future of React and not just a talent play
Basically: If you replace the "$1" placeholders from the article with URIs you wouldn't need a server.
(In most cases you don't need fully dynamic SSR)
The big downside is that you'd need a good pipeline to also have fast builds/updates in case of content changes: Partial streaming of the compiled static site to S3.
(Let's say you have a newspaper with thousands of prerendered articles: You'd want to only recompile a single article in case one of your authors edits the content in the CMS. But this means the pipeline would need to smartly handle some form of content diff)
I’ll just correct the allegation about the Vite — it’s being worked on but the ball is largely in the Vite team’s court because it can’t work well without bundling in DEV (and the Vite team knows it and will be fixing that). The latest work in progress is here: https://github.com/facebook/react/pull/33152.
Re: people not “getting” it — you’re kind of making a circular argument. To refute it I would have to shut up. But I like writing and I want to write about the topics I find interesting! I think even if you dislike RSC, there’s enough interesting stuff there to be picked into other technologies. That’s really all I want at this point. I don’t care to convince you about anything but I want people to also think about these problems and to steal the parts of the solution that they like. Seems like the crowd here doesn’t mind that.
As someone who’s been building web UI for nearly 30 years (scary…), I’ve generally been fortunate enough that when some framework I use introduces a new feature or pattern, I know what they’re trying to do. But the only reason I know what they’re trying to do is because I’ve spent some amount of time running into the problems they’re solving. The first time I saw GraphQL back in 2015, I “got” it; 10 years later most people using GraphQL don’t really get it because they’ve had it forced upon them or chose it because it was the new shiny thing. Same was true of Suspense, server functions, etc.
I don't want to have a fleet of Node/Bun backend servers that have to render complex components. I'd rather have static pages and/or React SPA with Go API server.
You get similar result with much smaller resources.
It allows for dynamism (user only sees the menus that they have permissions for), you can already show those parts that are already loaded while other parts are still loading.
(And while I prefer the elegance and clean separation of concerns that come with a good REST API, it's definitely more work to maintain both the frontend and the backend for it. Especially in caes where the backend-for-frontend integrates with more backends.)
So it's the new PHP (with ob_flush), good for dashboards and big complex high-traffic webshop-like sites, where you want to spare no effort to be able to present the best options to the dear customer as soon as possible. (And also it should be crawlable, and it should work on even the lowest powered devices.)
There’s of course a third option: the solution justifies the complexity. Some problems are hard to solve, and the solutions require new intuition.
It’s easy to say that, but it’s also easy to say it should be easier to understand.
I’m waiting to see how this plays out.
Where it gets a little more controversial is if you want to run Next.js in full fat mode, with serverless functions for render paths that can operate on a stale-while-revalidate basis. Currently it is very hard for anyone other than Vercel to properly implement that (see the opennextjs project for examples), due to undocumented "magic". But thankfully Next.js / Vercel have proposed to implement (and dogfood) adapters that allow this functionality to be implemented on different platforms with a consistent API:
https://github.com/vercel/next.js/discussions/77740
I don't think the push for RSC is at all motivated by the shady reasons you're suggesting. I think it is more about the realisation that there were many good things about the way we used to build websites before SPA frameworks began to dominate. Mostly rendering things on the server, with a little progressive enhancement on the client, is a pattern with a lot of benefits. But even with SSR, you still end up pushing a lot of logic to the client that doesn't necessarily belong there.
Seeing efforts like this (started by the main dev of Next.js working at Vercel) convinces me that the Vercel team is honestly trying to be a good steward with their influence on the React ecosystem, and in general being a beneficial community player. Of course as a VC-funded company its purpose is self-serving, but I think they're playing it pretty respectably.
That said, there's no way I'm going to run Next.js as part of a server in production. It's way too fat and complicated. I'll stick with using it as a static site generator, until I replace it with something simpler like Vite and friends.
Do I wish that it were something like some kind of Haskell-style monad (probably doable in TypeScript!) or a taint or something, rather than a magic string comment at the top of the file? Sure, but it still doesn't seem to be a big deal, at least on my team.
This was before the hook era.
Read a few of his many comments in any React issue and see what I mean. We are truly gifted. Dan you are my idol!
Think about how this could be done recursively, and how scoping could work to avoid spaghetti markup.
Aftertext: https://breckyunits.com/aftertext.html
There are many formats out there. If payload size is a concern, everyone is far better off enabling HTTP response compression instead of onboarding a flavor-of-the-month language.
Last thing one want in a wire format is white space sensitivity and ambiguous syntax. Besides, if you are really transferring that much json data, there are ways to achieve it that solves the issues
Check our mark miller's E stuff and thesis - this stuff goes all the way back to the 80s.
Check our mark miller's E stuff and thesis - this stuff goes all the way back to the 80s.
It's called "being part of the curriculum" and apparently the general insights involved aren't, so far.
---
I broadly see this as the fallout of using a document system as an application platform. Everything wants to treat a page like a doc, but applications don't usually work that way, so lots of code and infra gets built to massage the one into the other.
- https://overreacted.io/one-roundtrip-per-navigation/
- https://overreacted.io/jsx-over-the-wire/
The tldr is that endpoints are not very fluid — they kind of become a "public" API contract between two sides. As they proliferate and your code gets more modular, it's easy to hurt performance because it's easy to introduce server/client waterfalls at each endpoint. Coalescing the decisions on the server as a single pass solves that problem and also makes the boundaries much more fluid.
Re your "JSX Over The Wire" post, I think we've gone totally around the bend. A piece of code that takes 0 or more responses from a data backend and returns some kind of HTML is a web service. Like, that's CGI, that's PHP, that's Rails, Node, Django, whatever. If the argument here is "the browser should have some kind of state tracking/reactivity built in, and until that day we have a shim like jQuery or the old school thin React or the new school htmx" then OK, but this is so, so much engineering to elide `onclick` et al.
---
I kind of worry that we've spent way, way too much time in these weeds. There's millions and millions of lines of React out there, and certainly the majority of it is "stitch the responses of these N API calls together into a view/document, maybe poll them for updates from time to time", to the degree that AI just does it now. If it's so predictable that a couple of video cards can do it in a few seconds, why have we spent gazillions of engineering years polishing this?
Most of the time nobody needs this, make sure your database indexes are correct and don’t use some under powered serverless runtime to execute your code and you’ll handle more load than most people realize.
If you’re Facebook scale you have unique problems, most of us doesn’t.
> React has become a behemoth requiring vendor specific hosting
React is sub-3kB minified + gzip'ed [0], and the grand majority of React apps I've deployed are served as static assets from a fileserver.
My blog runs off of Github Pages, for instance.
People will always find a way to invent problems for themselves, but this is a silly example.
You know that the author of this post is the creator of React and that he's been pushing for RSC/Vercel relentlessly, right?
btw reactdom is ~30kb gzipped so React minimal bundle is around 35kb
Sure, I could’ve been clearer, but you did forget react-dom. And good luck getting RSC going on GH pages.
I typically compare Vanilla, Angular, SolidJS, Svelte, Vue Vapor, Vue, and React Hooks, to get a good spread of the major JS frameworks right now. Performance-wise, there are definitely differences, but tbh they're all much of a muchness. React famously does poorly on "swap rows", but also there's plenty of debate about how useful "swap rows" actually is as a benchmark.
But if you scroll further down, you get to the memory allocation and size/FCP sections, and those demonstrate what a behemoth React is in practice. 5-10× larger than SolidJS or Svelte (compressed), and approximately 5× longer FCP scores, alongside a significantly larger runtime memory than any other option.
React is consistently more similar to a full Angular application in most of the benchmarks there than to one of the more lightweight (but equally capable) frameworks in that list. And I'm not even doing a comparison with microframeworks like Mithril or just writing the whole thing in plain JS. And given the point of this article is about shaving off moments from your FCP by delaying rendering, surely it makes sense to look at one of the most significant causes to FCP, namely bundle size?
[0]: https://krausest.github.io/js-framework-benchmark/2025/table...
A good delta format is Mendoza [1] (full disclosure: I work at Sanity where we developed this), which has Go and JS/TypeScript [2] implementations. It expresses diffs and patches as very compact operations.
Another way is to use binary digging. For example, zstd has some nifty built-in support for diffing where you can use the previous version as a dictionary and then produce a diff that can be applied to that version, although we found Mendoza to often be as small as zstd. This approach also requires treating the JSON as bytes and keeping the previous binary snapshot in memory for the next delta, whereas a Mendoza patch can be applied to a JavaScript value, so you only need the deserialized data.
This scheme would force you to compare the new version for what's changed rather than plug in exactly what's changed, but I believe React already needs to do that? Also, I suppose the Mendoza applier could be extended to return a list of keys that were affected by a patch application.
If we imagine a streaming protocol of key/value pairs that are either snapshots or deltas:
event: snapshot
data: {"topPost":[], "user": {"comments": []}}
pending: topPosts,user.comments
event: delta
data: [17,{"comments":[{"body":"hello world"}]},"user"]
pending: topPostsIf pending arrays are just returned as empty arrays, how do I know if it’s empty because it’s actually empty, or empty because it’s pending?
GraphQL’s streaming payloads try to get the best of both worlds, at any point in time you have a valid payload according the GraphQL schema - so it’s possible to render some valid UI, but it also communicates what paths contain pending data, and then subsequent payloads act as patches (though not as sophisticated as Mendoza’s).
Of course, you could do it in-band, too:
{"comments": {"state": "pending", "values": []}}
["foo", "bar", "$1"]
["baz", "gar", "$2"]
And then a higher level is just iterating the array, and doesn't have to think about the promise. Like a Python generator or Ruby enumerator. I see that Javascript does have async generators, so I guess you'd be using that.
The "sort of" is that you can stream the array contents without literally splatting. The caller doesn't have to reify the whole array, but they could.
EDIT: To this not-really-a-proposal I propose adding a new spread syntax, ["foo", "bar", "...$1"]. Then your progressive JSON layer can just deal with it. That would be awesome.
>The format is a leading row that indicates which type of stream it is. Then a new row with the same ID is emitted for every chunk. Followed by either an error or close row.
and have
{ progressive: "false", a:"value", b:"value b", .. }
on top of that add some flavor of message_id, message_no (some other on your taste) and you will have a protocol to consistently update multiple objects at a time.
Probably the simplest one is to refactor the JSON to not be one large object. A lot of "one large objects" have the form {"something": "some small data", "something_else": "some other small data", results: [vast quantities of identically-structured objects]}. In this case you can refactor this to use JSON lines. You send the "small data" header bits as a single object. Ideally this incorporates a count of how many other objects are coming, if you can know that. Then you send each of the vast quantity of identically-structed objects as one-line each. Each of them may have to be parsed in one shot but many times each individual one is below the size of a single packet, at which point streamed parsing is of dubious helpfulness anyhow.
This can also be applied recursively if the objects are then themselves large, though that starts to break the simplicity of the scheme down.
The other thing you can consider is guaranteeing order of attributes going out. JSON attributes are unordered, and it's important to understand that when no guarantees are made you don't have them, but nothing stops you from specifying an API in which you, the server, guarantee that the keys will be in some order useful for progressive parsing. (I would always shy away from specifying incoming parameter order from clients, though.) In the case of the above, you can guarantee that the big array of results comes at the end, so a progressive parser can be used and you will guarantee that all the "header"-type values come out before the "body".
Of course, in the case of a truly large pile of structured data, this won't work. I'm not pitching this as The Solution To All Problems. It's just a couple of tools you can use to solve what is probably the most common case of very large JSON documents. And both of these are a lot simpler than any promise-based approach.
What if instead of streaming JSON, we streamed CSV line by line? That'd theoretically make it way easier to figure out what byte to stream from and then parse the CSV data into something usable... like a Javascript object.
I think you'd also need to have some priority mechanism for which order to send your triple store entries (so you get the same "breadth first" effect) .. and correctly handle missing entries.. but that's the data structure that comes to mind to build off of
But anyway this is a different custom framework which follows the principle of resource atomicity and a totally different direction than GraphQL approach which follows the principle of aggregating all the data into a big nested JSON. The big JSON approach is convenient but it's not optimized for this kind of lazy loading flexibility.
IMO, resource atomicity is a superior philosophy. Field-level atomicity is a great way to avoid conflicts when supporting real-time updates. Unfortunately nobody has shown any interest or is even aware of its existence as an alternative.
We are yet to figure out that maybe the real issue with REST is that it's not granular enough (should be field granularity, not whole resource)... Everyone knows HTTP has heavy header overheads, hence you can't load fields individually (there would be too many heavy HTTP requests)... This is not a limitation for WebSockets however... But still, people are clutching onto HTTP; a transfer protocol originally designed for hypertext content, as their data transport.
Very simple API, takes a stream of string chunks and returns a stream of increasingly complete values. Helpful for parsing large JSON, and JSON being emitted by LLMs.
Extensively tested and performance optimized. Guaranteed that the final value emitted is identical to passing the entire string through JSON.parse.
[0]https://github.com/beenotung/best-effort-json-parser
[1]https://github.com/unjs/destr
[2]https://www.npmjs.com/package/json-parse-even-better-errors
I wonder if Gemini Diffusion (and that class of models) really popularize this concept as the tokens streamed in won't be from top to bottom.
Then we can have a skeleton response that checks these chunks, updates those value and sends them to the UI.
I'm one of the developers of BAML.
Please, don't be the next clueless fool with a "what about X" or "this is completely useless" response that is irrelevant to the point of the article and doesn't bother to cover the use case being proposed here.
I have a filter I wrote that just reformats JSON into line-delimited text that can be processed immediately by line-oriented UNIX utilities. No waiting.
"The client can't do anything with JSON until the server sends the last byte."
"Would you call [JSON] good engineering?"
I would not call it "engineering". I would call it design.
IMO, djb's netstrings^1 is better design. It inspired similar designs such as bencode.^2
1. https://cr.yp.to/proto/netstrings.txt (1997)
2. https://wiki.theory.org/BitTorrentSpecification (2001)
"And yet [JSON's] the status quo-that's how 99.9999%^* of apps send and process JSON."
Perhaps "good" does not necessarily correlate with status quo and popularity.
Also, it is worth considering that JSON was created for certain popular www browsers. It could piggyback on the popularity of that software.
Also, not a single note about error handling?
There is already a common practice around streaming JSON content. One JSON document per line. This also breaks JSON (removal of newline whitespace), but the resulting documents are backwards compatible (a JSON parser can read them).
Here's a simpler protocol:
Upon connecting, the first line sent by the server is a JavaScript function that accepts 2 nullable parameters (a, b) followed by two new lines. All the remaining lines are complete JSON documents, one per line.
The consuming end should read the JavaScript function followed by two new lines and execute it once passing a=null, b=null.
If that succeeds, it stores the return value and moves to the next line. Upon reading a complete JSON, it executes the function passing a=previousReturn, b=newDocument. Do this for every line consumed.
The server can indicate the end of a stream by sending an extra new line after a document. It can reuse the socket (send another function, indicating new streamed content).
Any line that is not a JavaScript function, JSON document or empty is considered an error. When one is found by the consuming end, it should read at most 1024 bytes from the server socket and close the connection.
--
TL;DR just send one JSON per line and agree on a reduce function between the producer and consumer of objects.
We acknowledge that streaming data is not a problem that JSON was intended, or designed, to solve, and ... not do that.
If an application has a usecase that necessitates sending truly gigantic JSON objects across the wire, to the point where such a scheme seems like a good idea, the much better question to ask is "why is my application sending ginormeous JSON objects again?"
And the answer is usually this:
Fat clients using bloated libraries and ignoring REST, trying to shoehorn JSON into a "one size fits all" solution, sending first data, then data + metadata, then data + metadata + metadata describing the interface, because finally we came full circle and re-invented a really really bad version of REST that requires several MB of minified JS for the browser to use.
Again, the solution is not to change JSON, the solution is to not do the thing that causes the problem. Most pages don't need a giant SPA framework.
Any benefits using this over jsonl + json patch?
behnamoh•1d ago