Please make Open Source Github repo, so we can add our versions with another stack, or improve your versions. I really wanna show how Rust + Htmx version would be fast
i agree. would be fun to make one FE e.g. in Svelte/htmx and swap backends (Go, Rust, JS, PHP, C#, Gleam, Zig, ...) with same specs and different solutions.
@@jonklaric Yes, but that isn't the point of this video. A few of us pointed it would be fun to handle all backends with the same restrictions. (API, Caching, Db) The stacks ain't comparable in the video. imho
The NextJS implementation contains so much implicit knowledge you need to keep in your head to understand how all this "easy" "baked in" stuff works behind the scenes. Yes on something like elixir you have more lines of code but once you understand the mvc pattern it stays the same and does not change all the time. You can move around and understand the code base without knowing how all the cache & await connection etc. works. Nevermind the fact that in Next all of this implicit knowledge changes every two months
registered 2 days ago kekw. Also 1. Why should a person not know how cache and await works ? Await is nowadays essential and caching pretty much as well 2. Also Next.js 15 is now much more explicit with caching
@@martinlesko1521 I'm not talking about await. I'm talking about await 'connection' and whatever that means. The fact that you have to state that the newly released version 15000 of Next makes caching now much better kind of proves my point. They change things so much and give you these random new abstractions every couple of months that make it less sustainable to keep your projects up to date.
If I have a look at some of the comments, this video kind of drives home Prime's point about DX just being familiarity with the tools/language/... I'm not saying the niceties in Next aren't real, but the other stacks most definitely include features that would improve performance or "DX", but weren't used in this video because of lack of knowledge. The "number of files"-benchmark also just indicates a style difference. Splitting a project with a given number of lines over 10 or 20 files just moves the balance from searching within a file to searching for a file. With proper naming conventions and editor support, there's something to be said about both approaches. Very nice vdeo though, on a more global level this clearly indicated the pros and cons of not only the stacks themselves, but of the technology choices. You could sub rails for laravel, go for rust, next for nuxt and come to very similar conclusions because the "approach" is the same.
I’ve recently started using Hexagonal Architecture. It’s a lot of folders and files and more boilerplate than I’m used to, but I know where everything is, everything goes. And anyone who knows hex arch will be able to make sense of the code and add to it ez pz
You must be kidding, comparing nextjs just fetching and caching the whole db. And using redis to store clicks. It’s not a fare comparison with postgres+not having any cache and not quering the whole list from db at once. And even in this totally nextjs promoted comparison elixir shows itself better ;)
Also exaggerating how much file swapping there is in rails. A rails controller is tiny, the model is tiny. It looks like he generated his initial thing with scaffold and then just created an additional view and used that. You can see the jbuilder files etc. What is this project, one controller, one view and maybeee two models? That plus about 5 minutes spent in routes. The rest of it will be migrations, something he actually praised. The comments around the quality of documentation are very valid though. It’s like a brick wall and they’re incredibly comprehensive but hard to use as reference manuals
“I used a lot of Elixir back in the day…”. Doesn’t know what the priv folder is. I’m sorry mate, but that is nonsense. Either you didn’t do a lot of Elixir, or you never read any documentation of Elixir or Erlang. That is not Phoenix Specific.
Theo is not a real elixir engineer its clear that phoenix blows all of these modern frameworks out of the water but since nobody writes elixir most people dont know.
not knowing priv folder and saying "i used elixir back in the day" is valid, dunno what you find strange for his lack of knowledge, he didn't say he has a lot of experience now, but back then he had
I find that weird that the Go stack was more of a graphql stack. Graphql isn't specific to go at all, and there is tons other ways to have a simpler webstack running with go. Would have been fairer, though more work, to do a stack without graphql and try graphql on both go and node for example. Though appreciate the video and effort put into this.
Same. I think a more fair stack would have been Go + Templ + HTMX. But ultimately he would have used exactly the same frameworks for all Projects like (Go/Elixir/Ruby/...) + React or something like that. Edit: typo
Don't even need htmx IMHO, just the html template from standard lib would be enough for what he was doing. Then no setup required, works like a breeze.
For the Elixir/Phoenix app, you shouldn't have needed a controller. The fetching/sorting functions should be defined within your context (so roundest_phoenix folder, not rounded_phoenix_web), and your Liveview page should simply call these functions. You shouldn't write business logic in the controller (or Liveview, or anywhere in the roundest_phoenix_web folder) to begin with. I haven't needed to use a controller in Phoenix for well over a year at this point
And start_async instead of the task for the vote, and assign_async and corresponding component for that initial loading render for before the socket connects. I haven't needed to use the connected check in a while
The video makes no sense since you are obviously junior on several of these stacks. The same thing should be done but with senior developers, and they should have exactly the same specs
Fun fact about Rails is that u complain about things because u don’t know the framework. I don’t understand why u didn’t use turbo frame for your Rails demo when it is the default front framework for Rails… it’s like if u use Next without component .
You could have done something very similar to your Next cache in the Elixir version, by using a GenServer or an ETS table. I've found that Elixir/Phoenix projects can go crazy far w/o a database thanks to the OTP primitives.
@@abdelazizlaissaoui9079 But then that was a strange stack choice previously. And when you go forward in time you should also revisit your choices. Idioms change and you should always pick the best and/or widely used technology.
He could’ve used binary serialization via gRPC. Keep tanstack-query and just use raw sql or sqlc. That probably would’ve been the fastest of these stacks in terms of network
I really appreciate videos like this. Great to see your takes on these frameworks with a bit deeper understanding of some of the gripes, along with the things they get right
I really dislike this trend of using pokemon as sample data, clearly the best solution will always be to simply store the dataset in memory and everything else is overengineered. Doing something like returning a random wikipedia article would be way better.
React server component version seems a bit unfair in the comparison because you didn't use any database, where other projects did use it. Yes, it might not change the performance or anything in that regard, but since you ephesise on the line count - that's not a fair comparison. You could do the same thing on the GO and other versions to fetch data from the api directly in resolvers and that would decrease the number of lines in the project. Moreover this comparison doesn't address the performance and cost, only DX. Would be interested to actually know how much it would cost to run these versions and how performant would they be. Maybe an idea for the next video )
For the Rails app it should be a single request. It's SSR, not JS slop. 12:45 there is nothing stopping you from writing explicit changes in the add_mons_field. Using the naming convention does it for you, though. You could copy what's in the conventionally named migration into the add_mon_field migration and it would do that exact same thing when ran.
A Rails developer would be able to implement a solution faster than the the others, including api caching etc. No contest. So much explaining was required for all the JS versions. Even revalidatePage required a ton of explaining
Actually, fly doesn’t run docker. They released a really interesting video about this on their YT but TLDR they spin up a firecracker VM and then unpack all of the docker image layers onto that VMs filesystem. This gives them better isolation than just docker would while also providing better performance than VM + docker.
Develop that using a more appropriate stack and connect the two using APIs or MQs. Rails doesn’t do everything, but what it does do has a convention for use.
here here, I can't figure out where all of the hate for Rails comes from. Yeah that's true it's perfect for every single use case, but your car isn't a rocket. Does that mean you should get rid of it?
@@martinlesko1521 You can avoid convention whenever you want. There’s this whole section in the doctrine about it, “sharp knives” (I don’t love doctrines/manifestos btw) about it. The convention is there to make things quick. Ruby is a pretty powerful language and you can do whatever you want with rails. It’ll just be on you. Even then it’s actually not too bad and Laravel actually falls apart much quicker when you colour outside the lines. It’s a bit of a misconception.
I never faced the issues you mention with GraphQL. It’s a config issue in the way you defined the schema. And you can generate hooks for each query and mutation with codegen, all of which will be type-safe.
Elixir/Phoenix also has async assigns - this could possibly speed things up as you wouldn't need to block on fetching the next pair. Or it would be quite trivial to have a small 'next' cache genserver or something that gives you back a pair and reloads the next one in the background. So Elixir was fast, but there is still room to make it even faster I think.
Not sure what was going on with Rails, but it is definitely not normal for a new Rails project to take 6+ minutes to set up, especially with Ruby and all the gems already installed. That should take seconds, not minutes. It's not really fair to fault Rails for old Stack Overflow answers being wrong, though. That's going to happen for out-of-date answers on any topic in web development. Rails has just been around for a long time. The official setup guide is up to date, so I'd recommend using that. That being said, a lot of your other criticisms were definitely valid.
I think it's fair to point out that many answers about rails are going to be out of date. It's not specific to rails, but it's something that should be mentioned since it's something developers will experience.
@@smitjel3739 Devtools are definitely a good reason to dock points from a language. It's not like vscode is some niche editor with limited tooling, it's the most used and supported code editor. It's a massive hassle to have to change your entire dev env and/or use paid software to have a decent DX.
Turns out revalidatePath made your nextjs project refetch the whole database? Would generate a nice bill when deployed on Vercel... Kinda proves the point that there is a lot of magic going on in that framework. Having to rely on a personal connection to get it figured out is not the greatest look. What are we supposed to learn from a comparison about 2 projects with postgres, 2 projects with graphql (just generate protobufs and it would save 85% of the frustration from the GO part) and projects with a cached database and a kv store?
yeah, pattern matching in function signature is AMAZING. You just ditch if/switches from your code and have specialized methods to handle just that path that would be inside a conditional. It is something wonderful that I miss dearly.
@@ThugLifeModafocah simple function overloading, lost in the JS world. Remember learning about them and finding it so cool when learning Java in college lol
@@MKTV-1483 yeah, kind of overloading maybe linked with multimethod... I don't know. Anyways, it's too powerful because often, the overload happens on types, but in elixir you can do it types (using guards) and with map (struct) values (which I like more than guards). It is awesome.
So I'm also building in Elixir and Liveview and really have to agree with your painpoints. Seeding and env files are weird. I wrote the seedcode in a way so I can just connect to the server via IEX and copy paste it. For the env files i wrote a .env.exs file and imported it in the config. That worked pretty well. Regardless of that programming in Elixir is just wonderful and the responsiveness of Phoenix Liveview is insane.
Forgive me, I'm not a web dev, as an infra guy, couldn't you just dump your seed data from a dev db and import into prod? Is that not seen as elegant or am I missing something obvious?
if I was a private equity I would create a framework to gaslight and lure naive developers/managers into think 'this is easy' and play their ego into paying extreme markup for something you can host on a $5 VPS and then continue milking you.
Besides the total number of files and lines of code, it would also be nice to get an idea of how many lines of codes and files you have to create/edit yourself using something like a git diff. Although Elixir Phoenix has a large number of files and lines of code, the amount you have to write yourself is probably comparable or maybe even less than the other stacks. I am happy to see that Phoenix is number 1 in terms of performance though.
Phoenix shares a lot of DNA with Rails. It’s going to be low and Phoenix encourages reusability. Rails should also be very low. Part of its productivity is due to how little you actually need to write in order to build an application. That’s why there’s so much convention in place and a lot of the Rails “magic” is just good organisation and sensible defaults.
@@EightNineOne This is a feature of full-stack frameworks that are very opinionated imo. The same can be said about Laravel. JS frameworks lack in this regard because they are not opinionated enough.
@@Voidstroyeroh f’sure. Laravel is basically just rails in PHP. (Phoenix improves upon rails is a load more ways though IMO) I think what I’m driving at is even with how opinionated rails, django, Laravel etc are, rails still requires little boilerplate by comparison to get you up and running.
@Theo: The nullable-field issues with `| null` can be fixed by appending `!` to the end of each field on the GraphQL schema (look at the `field-nullability` page in the Apollo client docs)
11:15 You can have that in Rust if you use Diesel, the Diesel CLI is used to run the migrations, and it syncs the macros with your schema. And thanks to the strong typing, if you forget about it, your code won’t compile until you update the structs that should have changed with the schema.
@@Cuptial-ev9tb thats still weird bro there a like less that 10% of go projects that use bloated stuff like these just use normal go lang templates wth is this
laravel when? :-) but find it a super unfair comparison IMO if you are caching all in nextjs and then comparing this to the other frameworks where you are saving in db? 😅
Thanks. Great video. You might like SolidStart. I built the same project in Next and SolidStart, and I prefer SolidStart. Either way, this video helped me realize I picked great stacks. The others seem so unnecessarily complicated and verbose.
Hey Theo, I have always wandered if web dev complexity should be ranked based on the following 1. How easy is it to setup auth 2. How easy is it to setup Db 3. How easy is it to deploy What would you rate each of these rank in the above.
Will it be the new 1 billion row challenge ? Criteria: DevExp (How fast a new developer with some kind of experience can run pop into your approach?) + Less entropy possible + Max Performance possible
Returning data on POST at 1:08:48 comes at a cost: refreshing the page unintentionally upvotes. Typically POST requests mutate server state and return a redirect code to tell the browser where to go next / fetch the updated state, and leave the browser in state such that hitting refresh will not cause unintended mutation.
The thing for me with elixir/phoenix is it adds far less churn in the codebase... So fed up with npm dependencies adding breaking changes or React throwing out new ways of doing stuff in every new version... React apps always end up adding massive refactoring churn as pen testing always forces you to upgrade everything to latest.
The phoenix config files are very well made. I wish more franeworks adopted their separation (nestjs did!) however, yes it is intimidating the first time you see it in the same way that tailwind is intimidating. As for seeds, lol you are never ever supposed to do what you did. Seeds are script meant to ease local development, nothing more. If you wanted to push data to production, you’d import it via psql or turso cli or whatever else it is you are using. 😢
Appreciate all the effort put into this. And ben as well. Can’t say I know too much about caching, but it seems that cache the whole db vs having to query it in the other stacks is pretty unfair?
I had the SAME issue trying to connect to my local postgres with RoR. I was a brand new dev at the time, but it took me about an entire day or three to figure it out 😄. I was too dumb to even know what to ask stack overflow. Ya know and asking 'Help I can't connect to my database in RoR' on stack overflow is a quick way to get roasted, and your account banned haha. Looking back, I get it , that can mean a million different things, but yeah I didn't know any better at the time.
Please make Open Source Github repo, so we can add our versions with another stack, or improve your versions. I really wanna show how Rust + Htmx version would be fast
Great idea and free content for theo!
i agree. would be fun to make one FE e.g. in Svelte/htmx and swap backends (Go, Rust, JS, PHP, C#, Gleam, Zig, ...) with same specs and different solutions.
Even Go + HTMX would be way faster and smaller than Go + GraphQL + React frontend....
@@jonklaric Yes, but that isn't the point of this video. A few of us pointed it would be fun to handle all backends with the same restrictions. (API, Caching, Db) The stacks ain't comparable in the video. imho
Yeah... i think its more about DX than raw speed...
The NextJS implementation contains so much implicit knowledge you need to keep in your head to understand how all this "easy" "baked in" stuff works behind the scenes. Yes on something like elixir you have more lines of code but once you understand the mvc pattern it stays the same and does not change all the time. You can move around and understand the code base without knowing how all the cache & await connection etc. works. Nevermind the fact that in Next all of this implicit knowledge changes every two months
registered 2 days ago kekw. Also
1. Why should a person not know how cache and await works ? Await is nowadays essential and caching pretty much as well
2. Also Next.js 15 is now much more explicit with caching
@@martinlesko1521 I'm not talking about await. I'm talking about await 'connection' and whatever that means. The fact that you have to state that the newly released version 15000 of Next makes caching now much better kind of proves my point. They change things so much and give you these random new abstractions every couple of months that make it less sustainable to keep your projects up to date.
@@Felix-g8t1p "random new abstractions"
bruh
@@martinlesko1521 which one experimental cacheLife or unstable_cache?
"implicit knowledge" lmao in a video with RoR in it, this is your criticism.
Remember guys, there is a huge difference between a real developer/engineer and a content creator
If I have a look at some of the comments, this video kind of drives home Prime's point about DX just being familiarity with the tools/language/... I'm not saying the niceties in Next aren't real, but the other stacks most definitely include features that would improve performance or "DX", but weren't used in this video because of lack of knowledge.
The "number of files"-benchmark also just indicates a style difference. Splitting a project with a given number of lines over 10 or 20 files just moves the balance from searching within a file to searching for a file. With proper naming conventions and editor support, there's something to be said about both approaches.
Very nice vdeo though, on a more global level this clearly indicated the pros and cons of not only the stacks themselves, but of the technology choices. You could sub rails for laravel, go for rust, next for nuxt and come to very similar conclusions because the "approach" is the same.
@@wouterhenderickx6293 these are great points.
I’ve recently started using Hexagonal Architecture. It’s a lot of folders and files and more boilerplate than I’m used to, but I know where everything is, everything goes. And anyone who knows hex arch will be able to make sense of the code and add to it ez pz
You must be kidding, comparing nextjs just fetching and caching the whole db. And using redis to store clicks. It’s not a fare comparison with postgres+not having any cache and not quering the whole list from db at once. And even in this totally nextjs promoted comparison elixir shows itself better ;)
he glazes next to a fault i feel like..
Yeah... DB roundtrips with transaction management live compared to in-memory caching and KV stores....
RSC is smart isn't it ? ;))
Also, try moving that codebase away from Vercel… 🥶
Rails was faster too 😂😂😂
I think its pretty clear that you don't just introduce graphql into your project to make an SPA with 2 images and buttons.
Lol a JavaScript developer complaining about the number of folders in a project is crazy sounding 😅
Page.tsx intensifies
Also exaggerating how much file swapping there is in rails. A rails controller is tiny, the model is tiny. It looks like he generated his initial thing with scaffold and then just created an additional view and used that. You can see the jbuilder files etc.
What is this project, one controller, one view and maybeee two models? That plus about 5 minutes spent in routes. The rest of it will be migrations, something he actually praised.
The comments around the quality of documentation are very valid though. It’s like a brick wall and they’re incredibly comprehensive but hard to use as reference manuals
Node_modules enter the chat
@@naughtiousmaximus7853you know you can use non-file based routing in js frameworks right?
@@programmerjowo huh? That could be said for any language
Calling Go an old stack.
You're under arrest.
Why are you so insistent on not learning Rails properly?
Actually, you can use Turbo Frames to do the experience in a single request in the Rails version
With what he’s doing he couple probably just get away with turning morphing on
This
“I used a lot of Elixir back in the day…”. Doesn’t know what the priv folder is. I’m sorry mate, but that is nonsense. Either you didn’t do a lot of Elixir, or you never read any documentation of Elixir or Erlang. That is not Phoenix Specific.
i'm getting a feeling Theo doesn't do anything more than read the first page of popular frameworks and claims he is a developer
Theo is not a real elixir engineer its clear that phoenix blows all of these modern frameworks out of the water but since nobody writes elixir most people dont know.
He dropped the ball terribly also with Rails. I guess he unmasked himself.
not knowing priv folder and saying "i used elixir back in the day" is valid, dunno what you find strange for his lack of knowledge, he didn't say he has a lot of experience now, but back then he had
wow that elixir app demo was such nice to look at. its so smooth and fast when compared to nextjs demo for some reason. Need to try it manually.
I find that weird that the Go stack was more of a graphql stack. Graphql isn't specific to go at all, and there is tons other ways to have a simpler webstack running with go.
Would have been fairer, though more work, to do a stack without graphql and try graphql on both go and node for example.
Though appreciate the video and effort put into this.
These were indeed all 4 fully different implementations. It's not a good comparison at all.
Same. I think a more fair stack would have been Go + Templ + HTMX.
But ultimately he would have used exactly the same frameworks for all Projects like
(Go/Elixir/Ruby/...) + React or something like that.
Edit: typo
Don't even need htmx IMHO, just the html template from standard lib would be enough for what he was doing. Then no setup required, works like a breeze.
and let’s not forget graphql is a piece of shit
Its because Theo used GQL when he used Go, the video is about the tech stacks he used in his career
For the Elixir/Phoenix app, you shouldn't have needed a controller. The fetching/sorting functions should be defined within your context (so roundest_phoenix folder, not rounded_phoenix_web), and your Liveview page should simply call these functions. You shouldn't write business logic in the controller (or Liveview, or anywhere in the roundest_phoenix_web folder) to begin with.
I haven't needed to use a controller in Phoenix for well over a year at this point
This is actually really good to know thank you
And start_async instead of the task for the vote, and assign_async and corresponding component for that initial loading render for before the socket connects.
I haven't needed to use the connected check in a while
when do you use a controller then?
@@snowe.. expose API's, DeadView (traditional views)
As a Brazilian and a José, you NAILED the pronunciation at 17:33
The video makes no sense since you are obviously junior on several of these stacks. The same thing should be done but with senior developers, and they should have exactly the same specs
Fun fact about Rails is that u complain about things because u don’t know the framework. I don’t understand why u didn’t use turbo frame for your Rails demo when it is the default front framework for Rails… it’s like if u use Next without component .
You could have done something very similar to your Next cache in the Elixir version, by using a GenServer or an ETS table. I've found that Elixir/Phoenix projects can go crazy far w/o a database thanks to the OTP primitives.
i build my first real estate web app years ago with angular and rails. still works like a charm with very min maintenance
Thanks!
That's a strange Go stack.
I wouldn't really call it a fair test, when you are comparing a graphql backend against rest backends...
He is not testing stacks he tried the stacks that he used to work with back in the days
@@abdelazizlaissaoui9079 But then that was a strange stack choice previously. And when you go forward in time you should also revisit your choices. Idioms change and you should always pick the best and/or widely used technology.
He could’ve used binary serialization via gRPC. Keep tanstack-query and just use raw sql or sqlc. That probably would’ve been the fastest of these stacks in terms of network
Are you sure you used the proper Ruby/Rails extension? I use the Ruby LSP and it works just great! Do not encounter the type of issues you had
I am in the middle of my transition from JS dev to Elixir/Phoenix dev, I am glad that you like it too
I've been trying to transition to elixir as well, but it's not easy, any advice or resources
I really appreciate videos like this. Great to see your takes on these frameworks with a bit deeper understanding of some of the gripes, along with the things they get right
Bruh has to be the most jr developer if I had ever seen one
GraphQL implementation did Go dirty
I really dislike this trend of using pokemon as sample data, clearly the best solution will always be to simply store the dataset in memory and everything else is overengineered.
Doing something like returning a random wikipedia article would be way better.
Bro has no clue what he is doing.
😂
Most new users working with these technologies won't either. They'll just have 1 lang's exp.
Bro is building the same app with 5 frameworks. Not sure what yall were expecting
A round of applauses for the cookies hack, outstanding!
Thank you! I was proud of that one :)
Where was this?
@@t3dotgg lame
One thing you can do if you need to seed in prod is a migration. Seed scripts aren't idempotent.
You can remove unneeded boilerplate with options sent to rails new. You should consider using Phlex instead of erb.
14:40 fly is using MicroVM. They are unpacking docker but not using it for runtime.
React server component version seems a bit unfair in the comparison because you didn't use any database, where other projects did use it. Yes, it might not change the performance or anything in that regard, but since you ephesise on the line count - that's not a fair comparison. You could do the same thing on the GO and other versions to fetch data from the api directly in resolvers and that would decrease the number of lines in the project.
Moreover this comparison doesn't address the performance and cost, only DX. Would be interested to actually know how much it would cost to run these versions and how performant would they be. Maybe an idea for the next video )
For the Rails app it should be a single request. It's SSR, not JS slop.
12:45 there is nothing stopping you from writing explicit changes in the add_mons_field. Using the naming convention does it for you, though. You could copy what's in the conventionally named migration into the add_mon_field migration and it would do that exact same thing when ran.
Yeah right? Like the naming thing for g migration is a convenience feature, not a rule. They’re so easy to just write 😂
Do Gleam + Wisp + Lustre next! (If you want. No pressure. Love you Theo ❤)
Rails - 1:30
Elixir - 20:10
Go- 38:22
T3 - 56:05
Next - 1:08:30
A Rails developer would be able to implement a solution faster than the the others, including api caching etc. No contest.
So much explaining was required for all the JS versions. Even revalidatePage required a ton of explaining
Actually, fly doesn’t run docker. They released a really interesting video about this on their YT but TLDR they spin up a firecracker VM and then unpack all of the docker image layers onto that VMs filesystem. This gives them better isolation than just docker would while also providing better performance than VM + docker.
That cookie revalidation bit is pure gold 1:17:21 ⚜️
Cool! I wonder how HTMX + Go would do 🤔
Shity dx
@@okadz7037 what's shitty about it
@@LiveErrors Golang
@@okadz7037 i agree, the Templ lib to render html is very dirty. But is effective and fast, good to know if you already program in Go
@@jmlandi you don't need templ for 2 pages.. the build-in template engine is more then enough
Rails, convention over configuration. Understand the conventions.
good luck if you need something outside those conventions
Develop that using a more appropriate stack and connect the two using APIs or MQs. Rails doesn’t do everything, but what it does do has a convention for use.
@@martinlesko1521like what?
here here, I can't figure out where all of the hate for Rails comes from. Yeah that's true it's perfect for every single use case, but your car isn't a rocket. Does that mean you should get rid of it?
@@martinlesko1521 You can avoid convention whenever you want. There’s this whole section in the doctrine about it, “sharp knives” (I don’t love doctrines/manifestos btw) about it. The convention is there to make things quick. Ruby is a pretty powerful language and you can do whatever you want with rails. It’ll just be on you. Even then it’s actually not too bad and Laravel actually falls apart much quicker when you colour outside the lines. It’s a bit of a misconception.
I never faced the issues you mention with GraphQL. It’s a config issue in the way you defined the schema. And you can generate hooks for each query and mutation with codegen, all of which will be type-safe.
Great Idea - would appreciate more stack comparisons :)
Agreed! Would love to see PHP & vanilla JS. With the vanilla JS, I'd love to see the output lines of JS code.
Elixir/Phoenix also has async assigns - this could possibly speed things up as you wouldn't need to block on fetching the next pair. Or it would be quite trivial to have a small 'next' cache genserver or something that gives you back a pair and reloads the next one in the background. So Elixir was fast, but there is still room to make it even faster I think.
Not sure what was going on with Rails, but it is definitely not normal for a new Rails project to take 6+ minutes to set up, especially with Ruby and all the gems already installed. That should take seconds, not minutes.
It's not really fair to fault Rails for old Stack Overflow answers being wrong, though. That's going to happen for out-of-date answers on any topic in web development. Rails has just been around for a long time. The official setup guide is up to date, so I'd recommend using that.
That being said, a lot of your other criticisms were definitely valid.
I think it's fair to point out that many answers about rails are going to be out of date. It's not specific to rails, but it's something that should be mentioned since it's something developers will experience.
It takes maybe 5 minutes to setup rails, not sure what went wrong in his setup
He's blaming Rails for the limitations of his text editor...wtf? You guys do realize there's other text editors/IDEs than vscode, right? 🙄
@@smitjel3739 Devtools are definitely a good reason to dock points from a language. It's not like vscode is some niche editor with limited tooling, it's the most used and supported code editor. It's a massive hassle to have to change your entire dev env and/or use paid software to have a decent DX.
Turns out revalidatePath made your nextjs project refetch the whole database? Would generate a nice bill when deployed on Vercel... Kinda proves the point that there is a lot of magic going on in that framework. Having to rely on a personal connection to get it figured out is not the greatest look.
What are we supposed to learn from a comparison about 2 projects with postgres, 2 projects with graphql (just generate protobufs and it would save 85% of the frustration from the GO part) and projects with a cached database and a kv store?
does some weird graphql shit and complains go is bad lmao....
The way the ruby lsp extension also just sets up its own gemfile is a little psycho extra thing, thanks shopify
What is T6 ( 0:12 ) 😛
t3x2
Samsung SSD 😂
TBH this Elixir Phoenix code is elegant, and the ability to define same function multiple times make code even more elegant, Rails I'm sorry mate.
yeah, pattern matching in function signature is AMAZING. You just ditch if/switches from your code and have specialized methods to handle just that path that would be inside a conditional. It is something wonderful that I miss dearly.
@@ThugLifeModafocah simple function overloading, lost in the JS world. Remember learning about them and finding it so cool when learning Java in college lol
@@MKTV-1483 yeah, kind of overloading maybe linked with multimethod... I don't know. Anyways, it's too powerful because often, the overload happens on types, but in elixir you can do it types (using guards) and with map (struct) values (which I like more than guards). It is awesome.
Would love to see more of these videos. Python + FastAPI, maybe?
when i enter a try not to be unfair, inconsistent and biased and my opponent is theo
amazing job!!! thank u for spending so much time on this. and great overview! sure there are nuances but i love to see this 🔥
Super appreciative of the effort you guys put into this video! This was a great watch.
So I'm also building in Elixir and Liveview and really have to agree with your painpoints. Seeding and env files are weird.
I wrote the seedcode in a way so I can just connect to the server via IEX and copy paste it.
For the env files i wrote a .env.exs file and imported it in the config. That worked pretty well.
Regardless of that programming in Elixir is just wonderful and the responsiveness of Phoenix Liveview is insane.
Forgive me, I'm not a web dev, as an infra guy, couldn't you just dump your seed data from a dev db and import into prod? Is that not seen as elegant or am I missing something obvious?
if I was a private equity I would create a framework to gaslight and lure naive developers/managers into think 'this is easy' and play their ego into paying extreme markup for something you can host on a $5 VPS and then continue milking you.
Besides the total number of files and lines of code, it would also be nice to get an idea of how many lines of codes and files you have to create/edit yourself using something like a git diff. Although Elixir Phoenix has a large number of files and lines of code, the amount you have to write yourself is probably comparable or maybe even less than the other stacks.
I am happy to see that Phoenix is number 1 in terms of performance though.
Phoenix shares a lot of DNA with Rails. It’s going to be low and Phoenix encourages reusability.
Rails should also be very low. Part of its productivity is due to how little you actually need to write in order to build an application. That’s why there’s so much convention in place and a lot of the Rails “magic” is just good organisation and sensible defaults.
@@EightNineOne This is a feature of full-stack frameworks that are very opinionated imo. The same can be said about Laravel. JS frameworks lack in this regard because they are not opinionated enough.
@@Voidstroyeroh f’sure. Laravel is basically just rails in PHP. (Phoenix improves upon rails is a load more ways though IMO) I think what I’m driving at is even with how opinionated rails, django, Laravel etc are, rails still requires little boilerplate by comparison to get you up and running.
where can i find your cursor settings?
@Theo: The nullable-field issues with `| null` can be fixed by appending `!` to the end of each field on the GraphQL schema (look at the `field-nullability` page in the Apollo client docs)
Awesome to see Ben here!
Love Next RSC for sure
11:15 You can have that in Rust if you use Diesel, the Diesel CLI is used to run the migrations, and it syncs the macros with your schema. And thanks to the strong typing, if you forget about it, your code won’t compile until you update the structs that should have changed with the schema.
When and for how long did he use go? I remember prime was saying theo hasnt used go much
prolly like 2 weeks from this example
@@venomweed i saw one of his videos saying he used it for solid 2 months
@@AbhiShake-pl3cf close enough
1:09:01 really? Couldn't you just send back the data via the same POST request in rails?
Just started learning Elixir and man it feels good.
Why use graphql on the go project? You are just comparing random things.
It’s just a stack he’s familiar with bc of his early days at Twitch
@@Cuptial-ev9tb thats still weird bro there a like less that 10% of go projects that use bloated stuff like these just use normal go lang templates wth is this
What VSCODE theme is theo using? :o
I'm surprised you did not use dev-containers in vscode, you wouldn't need to install those things locally.
Long live Elixir! I hope to learn it soon..
laravel when? :-)
but find it a super unfair comparison IMO if you are caching all in nextjs and then comparing this to the other frameworks where you are saving in db? 😅
Sorry to ask but who won the match?
Definitely Tyson
@@t3dotgg hahaha but really who won?
No winner was announced? I’ll be back in the comments when someone announces it. Lol
Thanks. Great video. You might like SolidStart. I built the same project in Next and SolidStart, and I prefer SolidStart. Either way, this video helped me realize I picked great stacks. The others seem so unnecessarily complicated and verbose.
PLEASE THIS
Excellent video Theo!
I like RSC model, but I still prefer to fetch and mutate data with tanstack query.
it really feels like this entire video is just about how a framework-level caching abstraction makes things faster and easier to use?
Disappointed laravel didn’t make the cut, but it’s similar to rails. Would like to see your opinion on inertiajs
No laravel? 😢
Man, appreciate the work you put into this one, again !
@Theo, please INCREASE VOLUME, I maxed OUT on VOLUME and barelly hear you! Other YT videos and twitter vids are much louder!
The graphql nullability stuff is just an issue with the schema. Saw there was a PR open to fix that already.
Hey Theo, I have always wandered if web dev complexity should be ranked based on the following
1. How easy is it to setup auth
2. How easy is it to setup Db
3. How easy is it to deploy
What would you rate each of these rank in the above.
Will it be the new 1 billion row challenge ? Criteria: DevExp (How fast a new developer with some kind of experience can run pop into your approach?) + Less entropy possible + Max Performance possible
Phoenix mentioned! Let's go!
was t3 deployed on fly?
This is premium content. Thank you for that
Love technical breakdowns 🔥 versus the random dev Twitter drama stufff
Well obviously now we need Remix and SolidStart (same thing) versions
The support policy is so aggressive because Aspire supports the current LTS (.NET 8 at time of writing) and .NET 9 concurrently.
Sucks that modern t3 stack isn't aligned with the new patterns as much. Feels like its been a little abandoned.
We have things cooking! Sorry for the delays. You get most of the fun stuff if you use the app router config
Why is he not deploying the Nextjs version to fly??
Returning data on POST at 1:08:48 comes at a cost: refreshing the page unintentionally upvotes. Typically POST requests mutate server state and return a redirect code to tell the browser where to go next / fetch the updated state, and leave the browser in state such that hitting refresh will not cause unintended mutation.
Was curious to see Remix
The thing for me with elixir/phoenix is it adds far less churn in the codebase... So fed up with npm dependencies adding breaking changes or React throwing out new ways of doing stuff in every new version... React apps always end up adding massive refactoring churn as pen testing always forces you to upgrade everything to latest.
Cool vid. Ill try and do a Tanstack Start version.
The phoenix config files are very well made. I wish more franeworks adopted their separation (nestjs did!) however, yes it is intimidating the first time you see it in the same way that tailwind is intimidating.
As for seeds, lol you are never ever supposed to do what you did. Seeds are script meant to ease local development, nothing more. If you wanted to push data to production, you’d import it via psql or turso cli or whatever else it is you are using. 😢
Appreciate all the effort put into this. And ben as well. Can’t say I know too much about caching, but it seems that cache the whole db vs having to query it in the other stacks is pretty unfair?
I had the SAME issue trying to connect to my local postgres with RoR. I was a brand new dev at the time, but it took me about an entire day or three to figure it out 😄. I was too dumb to even know what to ask stack overflow. Ya know and asking 'Help I can't connect to my database in RoR' on stack overflow is a quick way to get roasted, and your account banned haha. Looking back, I get it , that can mean a million different things, but yeah I didn't know any better at the time.
Let me guess next JS is the best
Thanks
Ben should make a Svelte version!. Would love to see the comparison.