J2ME/MIDP was a fun platform to really cut your teeth on, and performed far better than many like to claim. For example, an old colleague of mine had written a Java emulator of a Game Gear, and it ran very well on J2ME, yet Android was horrific. Curiously the MIDP implementations out of Sun were also awful, and it was Nokia and some other European VM providers that did way better.
By far the most annoying aspect of J2ME were the jar size limits at the lower end, meaning to squeeze every byte (and this was real) you had to reduce the number of classes in the jar. Around 4-5 was typical. For many of the games on that page you'd be looking at 1 class of about 30000 lines, and a few others for device type specific abstractions, for example, maybe Sprite routines using Nokia DirectGraphics. This made teamwork on these games an absolute nightmare, especially in a pre-Git source code management era. (CVS/SVN/P4 were not exactly designed for working like that).
Later on there were some incredibly good studio specific optimization tools that handled transformations automatically, and outperformed humans, but these came into play exactly as iOS absolutely exploded.
They mention Mascot Capsule, which was a reasonably successful proprietary 3D engine pre-bundled on some devices. The other was the "standard" JSR 184, and I worked with at least two separate dev studios that had each implemented that in OpenGLES/C++ so that porting their 3D titles from J2ME to iOS/Brew could be reduced to an almost copy/paste business. This lasted maybe about two years before bothering with J2ME output was abandoned, but a side effect is the scene graphs for a lot of old iOS "premium" games look a lot like the one in JSR 184.
At least on Nokia's S60, the J2ME implementation was so good that it made Opera Mini feel almost native, and definitely more snappy than the actually native Opera Mobile (arguably mostly due to the server-side rendering).
It definitely took Android a while to get to that level of performance; the first few versions didn't even have JIT for Dalvik, while I think even Sun's MIDP implementation had it.
OpenGL also only became available on Android 1.6, while J2ME devices usually had either Mascot Capsule or JSR 184 as you say (sometimes backed by software rendering, but some S60 devices had a mobile GPU backing it, I believe).
> meaning to squeeze every byte (and this was real) you had to reduce the number of classes in the jar
I suppose Google did something right with Dalvik (other than improving their "It's not Java!" stance for copyright reasons), i.e. introduce a bytecode file format that can share constants across Java classes :)
No, they were proprietary, and their exact details remain trade secrets.
The best I knew of performed several tricks that the public research community assumes to be impossible. The big problem we had was making it totally deterministic as part of ensuring we had reproducible builds.
Of the public obfuscators in that era the most useful came from a company called dasho.
Ugh, really all the device specific fiddling with memory constraints or bugs always made me wonder why people complained about CSS incompabilties,etc :P
Really not missing doing J2ME development (even though at the time I heard that straight up Symbian was far-far worse).
J2ME is an unprecise term. I'd guess they mean J2ME/MIDP. The other profiles did even fly less probably.
I remember from the early 2000s you could get railway time tables from the German railways for your selected pair of stations as a midlet. That was truly useful.
I also used a mobile browser frontend. The data was rendered by the backend and transferred in compressed form. That was very usable at 2G speeds. Of course JavaScript was rare at the time. But I don't think the product was any commercial success.
Of cause their were (mostly toy) games. But in general the technology was probably 10+ years too early for the market.
I vaguely remember PP (Personal Profile IIRC). That was probably CDC/PP. Nokia either put it into one of the communicators or at least intended to so. 9210 I'd guess. But apps were extremely rare.
No, it was another company, I don't think a well-known name.
Opera Mini still works today on S60 phones from 2006ish. I regularly see web server log entries from a relative on my private server. But I believe it's a native Symbian app (.sis) not a Midlet. (I have working phones, but I don't have Opera Mini, so I cannot verify)
I think Opera Mini did much more than compression.
I seem to remember them actually rewriting (rendering?) the web page server side, and then sending an optimized mobile friendly version to make it more readable for pages that were originally designed for desktop.
It worked great on my phone and I even used it when the phones were more cabable, because web pages looked better and it saved a lot of bandwidth.
They rendered the site server-side (including JavaScript execution for a couple of seconds and all) and then sent the rendered page in some binary markup language to the client, with images heavily compressed.
I actually still sometimes use it on iOS! The app is no longer available in most (all?) App Store regions, but I still have it on my account and can redownload it. The servers seem to still be there!
IIRC, other browsers (the chinese UC browser and Nokia's OVI browser) had compression too. But opera was most popular (and perhaps most mature as well).
Opera Mini went way beyond just compression. It effectively rendered the website server-side and then sent optimized markup to the client (so it's much closer to "print site to SVG" on the server than "Content-encoding: gzip").
IIRC for WAP the only way to use form elements was JavaScript as there was no equivalent of HTML <form> that would just submit the request automatically.
J2ME was my first experience with mobile app development. It was very direct and easy to program for! But it definitely needed testing on the different devices. We had an array of different phones to test on. And some definitely had a nicer implementation than others.
The Java it supported was very old fashioned, with no generics. Which was a pain at times.
If you mean Chinese or Russian OEMs with their own version of ASOP, well that's not really Android anymore and is not using the Play app store anyway.
Western devices have to go through certification to use the name Android, so its not the wild wild west you think it is (although obviously some bugs may fall through).
I certainly mean Android device I can buy in any European country from the lame 100 euros pre-paid throwaway phone up to 2000 euros three ways foldable phone, plus everything else running Android on cars, kiosks, TVs, handhelds and everywhere else Android gets installed as alternative to GNU/Linux by OEMs.
CTS only covers a subset of what everyone is doing with Android.
And it will never cover everything, because as the Android team likes to point out, they don't want to stiffle the innovation of their partners.
I also have experience across Sharp, Nokia and Sony Ericson devices from J2ME days, and whatever brands ship Android phones, and outside flagship Android models, it is a jungle out there in drivers, AOSP customisations, vendor specific APIs, so I am of different point of view.
The most infuriating thing was extreme fragmentation when it came to support of various APIs in devices, even devices from the same vendor. Nokia was prominently culpable of this. The Finnish giant vomited out an enormous stream of S40 and S60 phones whose support for APIs was all over the map, which meant producing a shitload of JARs depending of what you needed, and the code was cluttered with constant checks of what is supported and what not. What an irony given Java's official motto of "Write once, run anywhere". Just freaking no.
Being a developer for Nokia was its own kind of hell. They never understood what API standardization is for and didn't care about your time and effort.
Hey, here are 12 different devices for this year with different hardware, screen size and other equipment, and you'd better get your hands on all of them to make sure that your app really works and looks acceptably, because an emulator only gets you half way. Also, any firmware update from us can kick your house of cards apart, and there will be like three of them coming in the next year. Happy programming!
> whose support for APIs was all over the map, which meant producing a shitload of JARs depending of what you needed
I wouldn't be surprised if this was actually one of the reasons why Sun fought MS and Google when they tried to make their own "Java" versions (embrace+extend). They didn't want a repeat of the J2ME situation.
I actually made money from selling a midlet via the Singtel marketplace. Not a lot but it was a good experience actually getting something to production outside of normal work.
Oh the special hell that were JSRs. As others mentioned, buying a crapload of multi-vendor devices to deploy specific JARs in hope to cover a sizeable customer base (not games, but public transport related).
so we've had some success getting libbluray with java working on android (i.e. needed for bluray's java support), but that requires an entire jvm built for android (heavy). I've always wondered if a j2me emulator (as bluray's java spec is built on top of j2me as I understand it), would be an easier way.
I got super interested with these in high school, because I finally got a Nokia N73 and that phone had the best and most complete J2ME implementation.
But I only had a netbook (atom cpu, 1gb ram, rotational hard disk) so I ended up coding J2ME using Emacs, a poorly written ant buildfile (due to my poor understanding of Ant) and the J2ME javadoc in a browser.
J2ME/MIDP was a fun platform to really cut your teeth on, and performed far better than many like to claim. For example, an old colleague of mine had written a Java emulator of a Game Gear, and it ran very well on J2ME, yet Android was horrific. Curiously the MIDP implementations out of Sun were also awful, and it was Nokia and some other European VM providers that did way better.
By far the most annoying aspect of J2ME were the jar size limits at the lower end, meaning to squeeze every byte (and this was real) you had to reduce the number of classes in the jar. Around 4-5 was typical. For many of the games on that page you'd be looking at 1 class of about 30000 lines, and a few others for device type specific abstractions, for example, maybe Sprite routines using Nokia DirectGraphics. This made teamwork on these games an absolute nightmare, especially in a pre-Git source code management era. (CVS/SVN/P4 were not exactly designed for working like that).
Later on there were some incredibly good studio specific optimization tools that handled transformations automatically, and outperformed humans, but these came into play exactly as iOS absolutely exploded.
They mention Mascot Capsule, which was a reasonably successful proprietary 3D engine pre-bundled on some devices. The other was the "standard" JSR 184, and I worked with at least two separate dev studios that had each implemented that in OpenGLES/C++ so that porting their 3D titles from J2ME to iOS/Brew could be reduced to an almost copy/paste business. This lasted maybe about two years before bothering with J2ME output was abandoned, but a side effect is the scene graphs for a lot of old iOS "premium" games look a lot like the one in JSR 184.
At least on Nokia's S60, the J2ME implementation was so good that it made Opera Mini feel almost native, and definitely more snappy than the actually native Opera Mobile (arguably mostly due to the server-side rendering).
It definitely took Android a while to get to that level of performance; the first few versions didn't even have JIT for Dalvik, while I think even Sun's MIDP implementation had it.
OpenGL also only became available on Android 1.6, while J2ME devices usually had either Mascot Capsule or JSR 184 as you say (sometimes backed by software rendering, but some S60 devices had a mobile GPU backing it, I believe).
> meaning to squeeze every byte (and this was real) you had to reduce the number of classes in the jar
I suppose Google did something right with Dalvik (other than improving their "It's not Java!" stance for copyright reasons), i.e. introduce a bytecode file format that can share constants across Java classes :)
> Later on there were some incredibly good studio specific optimization tools that handled transformations automatically
Do you know of any of those tools that are available? ProGuard Obfuscator is the only tool I know of to optimize your jar.
No, they were proprietary, and their exact details remain trade secrets.
The best I knew of performed several tricks that the public research community assumes to be impossible. The big problem we had was making it totally deterministic as part of ensuring we had reproducible builds.
Of the public obfuscators in that era the most useful came from a company called dasho.
> European VM providers
Now I can’t remember the name of that company from Switzerland…
They did some early work on dalvik but it didn’t fly commercially iirc…
Esmertec?
I think they've pivoted into microcontroller stuff.
Ugh, really all the device specific fiddling with memory constraints or bugs always made me wonder why people complained about CSS incompabilties,etc :P
Really not missing doing J2ME development (even though at the time I heard that straight up Symbian was far-far worse).
J2ME is an unprecise term. I'd guess they mean J2ME/MIDP. The other profiles did even fly less probably.
I remember from the early 2000s you could get railway time tables from the German railways for your selected pair of stations as a midlet. That was truly useful.
I also used a mobile browser frontend. The data was rendered by the backend and transferred in compressed form. That was very usable at 2G speeds. Of course JavaScript was rare at the time. But I don't think the product was any commercial success.
Of cause their were (mostly toy) games. But in general the technology was probably 10+ years too early for the market.
> J2ME is an unprecise term. I'd guess they mean J2ME/MIDP. The other profiles did even fly less probably.
Java’s attempt to make the GNU/Linux terminology nitpicking portable? ;)
Wasn’t there also CLDC, CDC etc.? I could never figure out what the hierarchy there was – was MIDP above or below C(L)DC?
All devices I ever encountered were of the MIDP and CLDC type, in any case.
Right those existed, too.
I think bottom up it's Java, J2ME, CLDC, MIDP.
So the programmer really works with MIDP.
I vaguely remember PP (Personal Profile IIRC). That was probably CDC/PP. Nokia either put it into one of the communicators or at least intended to so. 9210 I'd guess. But apps were extremely rare.
Was this Opera Mini? I remember installing that on my Nokia many years ago. It used compression as well. https://www.coderanch.com/t/229735/Opera-Mini-browser
No, it was another company, I don't think a well-known name.
Opera Mini still works today on S60 phones from 2006ish. I regularly see web server log entries from a relative on my private server. But I believe it's a native Symbian app (.sis) not a Midlet. (I have working phones, but I don't have Opera Mini, so I cannot verify)
I think Opera Mini did much more than compression.
I seem to remember them actually rewriting (rendering?) the web page server side, and then sending an optimized mobile friendly version to make it more readable for pages that were originally designed for desktop.
It worked great on my phone and I even used it when the phones were more cabable, because web pages looked better and it saved a lot of bandwidth.
They rendered the site server-side (including JavaScript execution for a couple of seconds and all) and then sent the rendered page in some binary markup language to the client, with images heavily compressed.
I actually still sometimes use it on iOS! The app is no longer available in most (all?) App Store regions, but I still have it on my account and can redownload it. The servers seem to still be there!
IIRC, other browsers (the chinese UC browser and Nokia's OVI browser) had compression too. But opera was most popular (and perhaps most mature as well).
Compression is inherent feture of WAP and some WAP browsers could use the WAP infrastructure to transfer not only WML, but also well-formed XHTML.
Opera Mini went way beyond just compression. It effectively rendered the website server-side and then sent optimized markup to the client (so it's much closer to "print site to SVG" on the server than "Content-encoding: gzip").
It still does! Ancient versions still work absolutely fine on old phones.
> Of course JavaScript was rare at the time.
IIRC for WAP the only way to use form elements was JavaScript as there was no equivalent of HTML <form> that would just submit the request automatically.
Very sad, that current dumbphones does not support J2ME. It is almost impossible to write own application for most recent Nokia keypad phones.
Don't they run some version of Firefox OS these days?
Of course, doing things equivalent to what J2ME did on kilobytes of RAM and storage requires hundreds of megabytes for webapps now...
J2ME was my first experience with mobile app development. It was very direct and easy to program for! But it definitely needed testing on the different devices. We had an array of different phones to test on. And some definitely had a nicer implementation than others.
The Java it supported was very old fashioned, with no generics. Which was a pain at times.
Android is no different, even though Google used to say otherwise to push Android over J2ME.
https://engineering.fb.com/2016/07/13/android/the-mobile-dev...
Even iOS exist in many different sizes and configurations these days. If you look closer at the pictures in that post you will see multiple iphones.
I think it is fair to say that no serious developer will publish an app without testing on at least 2-3 physical devices.
True, but iOS doesn't have OEMs having their own forks, which takes this to the same level as J2ME used to be.
Whatever they come up in GUI features, changes to AOSP behaviours, custom drivers, their own extended SDKs on top, hardware form factors,...
If you mean Chinese or Russian OEMs with their own version of ASOP, well that's not really Android anymore and is not using the Play app store anyway.
Western devices have to go through certification to use the name Android, so its not the wild wild west you think it is (although obviously some bugs may fall through).
See for example
https://source.android.com/docs/compatibility/cts
I certainly mean Android device I can buy in any European country from the lame 100 euros pre-paid throwaway phone up to 2000 euros three ways foldable phone, plus everything else running Android on cars, kiosks, TVs, handhelds and everywhere else Android gets installed as alternative to GNU/Linux by OEMs.
CTS only covers a subset of what everyone is doing with Android.
And it will never cover everything, because as the Android team likes to point out, they don't want to stiffle the innovation of their partners.
Having experience with both platforms I'd say it's not comparable. J2ME was much worse in this regard.
I also have experience across Sharp, Nokia and Sony Ericson devices from J2ME days, and whatever brands ship Android phones, and outside flagship Android models, it is a jungle out there in drivers, AOSP customisations, vendor specific APIs, so I am of different point of view.
Perhaps I'm spoiled from working mostly with React Native nowadays and have forgotten the pain :)
I have a few J2ME apps under my belt as well.
The most infuriating thing was extreme fragmentation when it came to support of various APIs in devices, even devices from the same vendor. Nokia was prominently culpable of this. The Finnish giant vomited out an enormous stream of S40 and S60 phones whose support for APIs was all over the map, which meant producing a shitload of JARs depending of what you needed, and the code was cluttered with constant checks of what is supported and what not. What an irony given Java's official motto of "Write once, run anywhere". Just freaking no.
Being a developer for Nokia was its own kind of hell. They never understood what API standardization is for and didn't care about your time and effort.
Hey, here are 12 different devices for this year with different hardware, screen size and other equipment, and you'd better get your hands on all of them to make sure that your app really works and looks acceptably, because an emulator only gets you half way. Also, any firmware update from us can kick your house of cards apart, and there will be like three of them coming in the next year. Happy programming!
> whose support for APIs was all over the map, which meant producing a shitload of JARs depending of what you needed
I wouldn't be surprised if this was actually one of the reasons why Sun fought MS and Google when they tried to make their own "Java" versions (embrace+extend). They didn't want a repeat of the J2ME situation.
I actually made money from selling a midlet via the Singtel marketplace. Not a lot but it was a good experience actually getting something to production outside of normal work.
Oh the special hell that were JSRs. As others mentioned, buying a crapload of multi-vendor devices to deploy specific JARs in hope to cover a sizeable customer base (not games, but public transport related).
so we've had some success getting libbluray with java working on android (i.e. needed for bluray's java support), but that requires an entire jvm built for android (heavy). I've always wondered if a j2me emulator (as bluray's java spec is built on top of j2me as I understand it), would be an easier way.
This brings back memories, as someone that start developing for mobile phones targeting Sharp GX 10 for a Vodafone competition.
https://www.mobilephonemuseum.com/phone-detail/vodafone-gx10
Here's my fork of FreeJ2ME, the desktop equivalent with m3g/mascot capsule support: https://github.com/zb3/freej2me
I remember one of my senior colleagues (I was just an internship) making a J2ME app to calculate who should pay the coffee for others.
What a great tool!
Does anybody here know how it works at a high level? Does it implement a JVM, or is the bytecode recompiled to target Android’s Dalvik?
From peeking through the code, it seems the latter.
It manually implements j2me api (javax.microedition.* packages and extensions) over the android api.
Ah yes, and there also seems to be a DEX (Dalvik's bytecode format) library included: https://github.com/nikita36078/J2ME-Loader/tree/master/dexli...
Oh man I played a lot of J2ME games in the past.
Fishlabs games, Attack Chopper, various Gameloft games...
Oh the memories, J2ME.
I got super interested with these in high school, because I finally got a Nokia N73 and that phone had the best and most complete J2ME implementation.
But I only had a netbook (atom cpu, 1gb ram, rotational hard disk) so I ended up coding J2ME using Emacs, a poorly written ant buildfile (due to my poor understanding of Ant) and the J2ME javadoc in a browser.
Those were the times, for me.
Oh to be young again...
You think you're old? I was part of the team at the other end of that stack building and debugging the Java runtime for the Nokia Symbian phones :)
Thank you, you did an amazing job! My hot take is that J2ME on Symbian was better than the native API/runtime in some aspects :)
> You think you're old?
This wasn't meant to start a competition on who's older. I was just meaning to express the fondness of my memories about J2ME and those times.
One of my very first ventures into programming as a little child was trying to figure out how to write my own games for my nokia phone.
I remember being so hopeless staring into the decompiled java code for my favorite games trying to learn something from it and spin into my own game.
I'm now an (aspiring) android dev, and looking back at the past I can only say "Damn, it has gotten so much better since then"