Hacker News Reader

Hacker News

Comment by cft | Hacker News Reader

Parent Full thread

cft

Choosing Java and Dalvic was probably a mistake for Android.

View on HN

If they had to redo it today, what could it be instead?

* Javascript -- I hate the language, it has enough warts to make it unpalatable

* Dart -- I can see it, that would be great, still kind of new though

* Python -- I understand they could have done it, but it was a political decision not to (at least according to Guido's remark during one of the PyCon's), performance might not be up to par on some devices

* Go -- too low level perhaps, but it would be cool

Go, C/C++, but in the ideal world- Python (they would have to very seriously tweak the interpreter)- imagine how easy app creation would have become with Python? They would have a huge advantage over iOS by lowering the entry thresholds for developers.

Python is neither fast enough or robust enough to use as the foundation for an OS. You need static typing and strong tooling both for performance and for correctness. Those that disagree with this need only point me to a successful counter-example.

Google has modified Java for Android. They could have instead modified Python. There have been academic papers that have shown that Python can be made to be statically typed (ignoring one or two keywords, like eval).

Assuming that Google had done these, there should be no reason that Python would not be fast or robust enough.

Yeah Guido was basically asked this at a PyCon in the past. He was working for Google and at some point he probably had brought up the question. Someone in the audience asked about it and his comment was that it was "political". In other words technically possible but powers that be decided against it for other reasons. It is also important how he said it, he was very calm before and after, but when answering that question he was visibly upset. You could tell it was a sore topic for him that made him pretty angry.

LuaJIT is a performance counterexample for dynamic typing, being around the same as Java.

LuaJIT may make the performance of Lua better or even just acceptable in some cases, but it does nothing to help ensure correctness. Correctness is almost always more important than performance. It most cases it is useless to do something quickly, but to also be doing it incorrectly.

I was responding to the OP who said static typing was necessary for performance, by providing a counterexample.

And "better or just acceptable" is rather dismissive of the facts.

Strongtalk is another example that can be given of dynamic performance.

"They would have a huge advantage over iOS by lowering the entry thresholds for developers."

Low entry thresholds for developers is nothing but poison for a platform.

The advantage the DVM has is it's a baseline that developers can work with instead of worrying about hardware.

This was probably very desirable to Google in 2005 as it gave them the chance to court mobile manufacturers with the promise that their operating system can be deployed across a range of devices and theoretically all apps etc should remain functional, regardless of hardware as it's the VM that's the deployment target, not the phone.

Remember that back in those days the mobile OS world was god awful, Symbian was abysmal, JavaME was broken and Windows CE....well let's not go there.

Did they really have to pick one language to rule them all?

It seems odd to me to design a platform that only supports one programming language. What happens when people find better ways of expressing their intent?

A phone in either Smalltalk or Lisp would be awesome IMO.

Not buying Sun was the mistake. $7B is all that would've taken (less if they sold the os and hardware division) to make the computing world a better place.

Not necessarily. If the judgement that APIs are not copyrightable is upheld (which is a big if), the computing world will be a better place than it would be if Google had managed to avoid the question.

the edge goes both ways - if it somehow (hopefully, now its false) judged that API's _are_ copyrightable, then what?

7B is pocket change to them, completely agree. I was shocked to hear they let it pass iirc. I figured they were playing hardball and it just blew up in their face.

Or did they make the right decision? Just because they have the money doesn't mean they need to spend it. Did litigating cost them 7 billion? The legal win shows they made the right choice in not spending 7 billion on what they are allowed to use for "free".

IMHO you're over simplifying. Oracle was better positioned to capitalize on Sun assets, so they were likely to pay more than Google. Microsoft or Apple may have decided to bid to prevent Google from acquisition. Merging with a huge enterprise is always painful.

I'm not so sure given the results. Google (with selling the hardware / chip division to Fujitsu) would have done pretty well given all the tech projects Sun had.

If Apple had bid to prevent (patents I suppose), I wonder if ZFS would have finally been on OS X. Java would have been the truly odd man out in that acquisition. I could see Java being sold to IBM.

zfs actually was on a few leopard betas.

Not that it worked well mind you, it made Finder even more useless than before. But it did "work" for varying definitions of the word.

Yeah, I was more thinking production ready default and integrated into the shipped programs (e.g. Disk Utility).

Oh I agree, what was there was nice too by the way. Shame it never made it though.

At least we have dtrace!

While you're obviously right (Oracle _did_ pay more that whatever Google offered (if anything), in retrospect perhaps blinkingled's assertion that $7B might have been a reasonable amount for Google to have offered seems at least arguable. In one sense perhaps you are oversimplifying - sure Oracle was in a better place to use some of Sun's assets, but Google is clearly now in a dangerous place having a competitor like Oracle owning major parts of technology Google's mobile strategy depends on. Whether that's seven billion dollars worth of danger I have no idea - but I suspect there are people at Google regretting the decision _not_ to buy Sun (or more specifically, buy Java and get all the rest of Su for free) back when they had the opportunity.

Google is basically a java shop. Are you saying they are unable to capitalize of Sun assets ? they choosed java as Dalvik first class language , what more do you need to capitalize of Sun's asset. It is a mistake and google will pay a huge price in the future.

Question: will this cost Google $7B?

I've thought this myself lately but what alternatives did they really have? The only other real alternatives are C++ and C#, which are both problematic for different reasons.

I wouldn't mind seeing a new mobile OS build around Go but it's way too late for a change that fundamental for Android. For better or worse Java & Android are married now.

"what alternatives did they really have?"

JavaScript

I really hope you're joking. No dynamic language is suitable for software engineering at that scale and particularly not one as badly flawed as JS.

I'm not a huge fan of JavaScript (to say the least) but I'm not joking either. Most Android apps aren't exactly "software egineering at scale". Certainly no more so than a modern web app.

The first Dalvik versions didn't even have a JIT, so performance apparently wasn't the main concern. They could have added optional static typing to their JavaScript VM as well.

I think Google chose Java because many people know it and because the Java faction inside Google is powerful. But many people know JavaScript as well, and arguably the typical Java developer isn't exactly a UX guru. So JavaScript would have been a better fit for Android even though I (and you) don't like it.

Actually a lot of apps are every bit as complex as a typical desktop app. I'm not sure if you've done much app development but many of the apps I've worked on have been pretty large.

Unless you propose writing the entire OS in JS then there's also a huge advantage in using the same language for core OS and APIs and for application development.

In any case as some of the more modest Java replacements like Xtend and Kotlin mature I expect to see Java the language recede in importance for Android.

"there's also a huge advantage in using the same language for core OS and APIs and for application development."

I don't disagree, but the core OS is written in C, a language that scares off a lot of people.

Sure but the bulk of the code above Linux + userland is Java and all the APIs are Java. In hindsight it might have been a better choice to just buy Mono but that ship has long since sailed, for better or for worse.

Java might have been, but I am not sure about Dalvik. Without the Java language conversion, I doubt Oracle would have ever bothered to attack Dalvik.

On the other hand, so far this case has been nothing but a boondoggle for Oracle. It seems that is unlikely to change at this point.

Reply on news.ycombinator.com