I am biased on the subject as I am the author or tss-react (the second way recommended by mui to migrate from v4 to v5) but I wouldn't advise to actually implement the method described here because:
If you are using typescript: It provides no types.
No SSR (Next.js) support ( styles will be loaded client side)
No nested selectors (the $ syntax of JSS)
No withStyles()
Hard to debug, all classes will be mui-xxxxx.
This useStyles() accept no prams.
It relies on @emotion/css which is different from @emotion/react. Mui depends on @emotion/react
No mention of how to prioritize one class over another.
I really don't want to dunk on an otherwise very good article. I just think the case against actually implementing it had to be made.
This way you can
a) debug ALL the class name in development and
b) you reuse classes in production. If you use the label / class names in production then you can not reuse them
If you need additional parameters, just add them to useClasses :) it is a simple function.
Yes it relies on @emotion/css, which relies on stylis, @emotion/serialize, @emotion/cache, @emotion/sheet, etc. ... basically exact the same base libraries on which relies @emotion/react :)
In general, this is a self made small function that can help someone (like us) to migrate easily from mui4 to mui5, without the need to install any third party library.
In our use-case the mui5 migration scripts did the most, the only problem was the makeStyles/useStyles hook, which could be easily solved with this simple 15 lines of code.
This function can easily be extended to use nested classes or typescript, or with any other functionality. Anyway if someone need more (complex) features or typescript I would also recommend to use tss-react :)
I even put right now this recommendation in my article! ;)
I even put right now this recommendation in my article! ;)
It's very nice of you! Much appreciated! :D
But surely it is not that big deal to adjust it :)
This is, infact, extremely hard to adapt this API to work with type inference. I'd say the types definitions represent 90% of the overall effort that went into developing tss-react.
Regarding SSR, you are not right, emotion/css is capable of SSR. You just need one or two more tweaks.
It's not just a few tweaks. It is very hard to get working properly. However, user that chose to implement your approach can always refer to TSS to setup SSR.
It is also not hard to debug. If you really need to debug, you can do following
Fair enough, it's also the approach implemented in TSS (except that we can optionally add the component name but you could easily add it as well).
If you need additional parameters, just add them to useClasses :) it is a simple function.
Fair enough
Yes it relies on @emotion/css, which relies on stylis, @emotion/serialize, @emotion/cache, @emotion/sheet, etc.
You are right, it isn't a very strong argument.
In our use-case the mui5 migration scripts did the most, the only problem was the makeStyles/useStyles hook, which could be easily solved with this simple 15 lines of code.
There is now a codemod in MUI for migrating from JSS to TSS. See doc
This function can easily be extended to use nested classes or typescript.
Again, with all due respect, no, I don't want people to think. "Well, if I ever need the nested selectors I'll just need to adapt my makeStyles a little bit". Getting selectors to work is a degree of magnitude harder than implementing the base hook.
Again. No hate whatsoever. Just helping people to make an informed decision.
I am biased on the subject as I am the author or tss-react (the second way recommended by mui to migrate from v4 to v5) but I wouldn't advise to actually implement the method described here because:
$
syntax of JSS)withStyles()
mui-xxxxx
.useStyles()
accept no prams.@emotion/css
which is different from@emotion/react
. Mui depends on@emotion/react
I really don't want to dunk on an otherwise very good article. I just think the case against actually implementing it had to be made.
I also would not recommend this approach when you want to use Nested Selectors or withStyles.
emotion.sh/docs/ssr#api
This way you can
a) debug ALL the class name in development and
b) you reuse classes in production. If you use the label / class names in production then you can not reuse them
If you need additional parameters, just add them to useClasses :) it is a simple function.
stylis, @emotion/serialize, @emotion/cache, @emotion/sheet, etc.
... basically exact the same base libraries on which relies @emotion/react :)In general, this is a self made small function that can help someone (like us) to migrate easily from mui4 to mui5, without the need to install any third party library.
In our use-case the mui5 migration scripts did the most, the only problem was the makeStyles/useStyles hook, which could be easily solved with this simple 15 lines of code.
This function can easily be extended to use nested classes or typescript, or with any other functionality. Anyway if someone need more (complex) features or typescript I would also recommend to use tss-react :)
I even put right now this recommendation in my article! ;)
It's very nice of you! Much appreciated! :D
This is, infact, extremely hard to adapt this API to work with type inference. I'd say the types definitions represent 90% of the overall effort that went into developing
tss-react
.It's not just a few tweaks. It is very hard to get working properly. However, user that chose to implement your approach can always refer to TSS to setup SSR.
Fair enough, it's also the approach implemented in TSS (except that we can optionally add the component name but you could easily add it as well).
Fair enough
You are right, it isn't a very strong argument.
There is now a codemod in MUI for migrating from JSS to TSS. See doc
Again, with all due respect, no, I don't want people to think. "Well, if I ever need the nested selectors I'll just need to adapt my
makeStyles
a little bit". Getting selectors to work is a degree of magnitude harder than implementing the base hook.Again. No hate whatsoever. Just helping people to make an informed decision.
The most frustrating thing here is
No nested selectors
indeed