Hacker News Reader

Comment by rogerbinns | Hacker News Reader

The big change hinted at in the documentation around the 3.5 era is that SQLite switched from a stack based VM to a register based one. Note that the implementation details are not exposed to a developer using SQLite and there is a massive test suite so the change had no visible impact.

What isn't mentioned is why SQLite is using a virtual machine in the first place. The reason is that SQLite only calculates the next row of results when you ask - it does not calculate all result rows at once in advance. Each time you ask for the next row of results it has to resume from where it last left off and calculate that next row. The virtual machine is a way of saving state between those calls for the next row (amongst other things).

This is also why there isn't a method in SQLite DB adapters to get the number of result rows. SQLite has no idea other than actually calculating them all which is the same amount of effort as getting all of them.

View on HN

>Each time you ask for the next row of results it has to resume from where it last left off and calculate that next row. The virtual machine is a way of saving state between those calls for the next row (amongst other things).

They could have achieved the same effect by implementing it in a language with continuations or coroutines, or maybe just iterator functions. That might be viable today if the language is fast enough and callable from C.

C# and LuaJIT spring to mind.

But in the end, you're just trading performance for abstractions. There's not really a functional difference between writing the DB in a language running on a VM implemented in C; or writing a DB with a VM implemented in C.

If performance and portability are your goals, C is going to be the logical choice.

You already have the abstractions, they're just C abstractions.

"They could have achieved the same effect by implementing it in a language with continuations or coroutines"

They did. They just implemented their own minilanguage instead, that (AFAIK) simply has no conventional syntax-based serialization. It's still a language, though. Given their tight focus, it is completely plausible that they can create a little VM that will run far, far better than any general-purpose language VM could.

SQLite predates both; the first official release is from August 2000.

Also, one of the goals of SQLite is to be available everywhere. It's available as super-standard-super-clean C code in one .c file, that works on just about every architecture. Had they used C# or LuaJIT, that could not be achieved.

I'm not sure if you are aware, but SQLite is what powers almost every app on Android or iPhone that needs tabular storage, including the built in ones like Contacts.

SQLite is used everywhere. Embedded devices, all sorts of different architectures, etc.

C# and LuaJIT are not options. I don't really understand what your point is, because they implemented their own VM instead.

Fair enough, but I was thinking "how would I implement it today" rather than "how should they have implemented it at the time."

Implementing it today would have the same issues: SQLite runs on more platforms than even LuaJIT; and even then I still don't see that a whole general purpose language, no matter how small or embeddable (mostly thinking of Lua here) would compete against SQLite's "be good at SQL-y computation" VM. The main argument for Lua is having humans write it; but the human interface to SQLite is supposed to SQL. I'm not sure I understand the benefit of compiling the SQL down to Lua so it can then be interpreted by the standard Lua implementation...

Thanks for the insight. Do you know the reasons why they switched from using stack to registers in their VM implementation?

I do not know, but if I had to guess, it is for performance.

I am no expert in this field, but AFAIK http://static.usenix.org/events/vee05/full_papers/p153-yunhe... (discussed on http://news.ycombinator.com/item?id=704576) has not really been refuted.

I base that on the fact that both Dalvik (http://davidehringer.com/software/android/The_Dalvik_Virtual...) and Lua 5.0 (http://www.lua.org/doc/jucs05.pdf) use register-based VMs.

The Parrot VM is also register based: http://en.wikipedia.org/wiki/Parrot_virtual_machine

We found it is much easier to handle exceptions without risking stack leaks using a 3-address register machine. Code generation is also simplified in a register machine compared to a stack machine (at least in the case of a VM designed specifically to run SQL statement).

The VM used by SQLite is different from the VM used by things like Javascript or Python in that SQL is not a general-purpose programming language. Most of the opcodes in the SQLite VM are heavy-weight concepts such as "insert a key/value pair into a b-tree" or "decode the N-th column of a table row". These opcodes are implemented with hundreds or thousands of lines of C code. The overhead of instruction dispatch is insignificant in comparison. And so there isn't much of a performance advantage one way or the other between a stack machine and a register machine in SQLite. The choice of a register machine is for maintainability, reliability, and testability.

Reply on news.ycombinator.com