7 ms·

Terra: A low-level counterpart to Lua

140 points by charlieegan3 11y ago

laarc 11y ago
Terra seemed promising when I ran across it a month ago. Does anyone know if it's possible to use LuaJIT's ffi module while writing Terra code? What I mean is, there is a large body of existing code that I'd like to run alongside Terra. But if the choice is Terra xor LuaJIT, this becomes difficult.
- an_ko 11y ago
  You might be looking for http://terralang.org/api.html http://terralang.org/api.html § Managing Terra Values from Lua. Quoting: > We provide wrappers around LuaJIT’s FFI API that allow you to allocate and manipulate Terra objects directly from Lua.
- corysama 11y ago
  You can use Lua, Terra or both at runtime. The Lua part is LuaJIT including the ffi. The Terra part is LLVM and can link directly to C (maybe C++?). Lua can be used as a metaprogramming system to produce a static Terra .o or exe. Or Lua can be used as a runtime that can JIT and invoke Terra functions on the fly.
andrewmcwatters 11y ago
So this is LuaJIT's ffi with some bastardized syntax that immediately becomes incompatible with any other Lua syntax parser, because "pretty." Just no.
- mpdehaan2 11y ago
  It's been a long time since I've done Lua (2005/2006 IIRC?), but I remember Lua's C embedding API being remarkably excellent already. C calling Lua, calling back into C, back and forth across the boundary was absolutely painless. Compared with experiencing that in many other languages, at the time, I was convinced that was as easy as it is. (Plus, the size of the lua runtime made it great for embedded projects, whereas now embedded computing has gotten so much more powerful you can really include almost any language stack if you really want to) As a result, I'm not positive why I need a C-like layer on top for cases where I probably want to write C.
  - andrewmcwatters 11y ago
    Exactly. Anyone who has ever actually done embedded work with Lua knows you can hardly get any closer in terms of "low-level" outside of the C API besides using LuaJIT's ffi. If anything this is actually a high level abstraction on top of LuaJIT's ffi, so this is far from low level. God forbid anyone criticize anything on HN.
    - haberman 11y ago
      I think Terra functions are AOT compiled and have explicit strong types. So it's quite a different thing that LuaJIT. For example, you could use Terra to actually write a JIT.
      - andrewmcwatters 11y ago
        Why would you use Terra to write a JIT? It uses LLVM. To write JIT in JIT?
        haberman 11y ago
        Think of it as a programming environment for generating LLVM IR that is a lot more convenient than using LLVM's C++ API.
        coldtea 11y ago
        Sort of like people use Java to write JVM languages?
    - vbit 11y ago
      The front page describes three distinct use cases for Terra. While criticism is welcome, it certainly appears that you are missing the uses for Terra. Consider the third case - you can write Terra functions (cleaner syntax than C) and use Lua as a meta-programming language that produce more Terra functions (vastly superior to C macros) which then compile down via LLVM to machine code. Your final code has no Lua pieces.
- mpdehaan2 11y ago
  It's been a long time since I've done Lua (2005/2006 IIRC?), but I remember Lua's C embedding API being remarkably excellent already. C calling Lua, calling back into C, back and forth across the boundary was absolutely painless. Compared with experiencing that in many other languages, at the time, I was convinced that was as easy as it is. (Plus, the size of the lua runtime made it great for embedded projects, whereas now embedded computing has gotten so much more powerful you can really include almost any language stack if you really want to) As a result, I'm not positive why I need a C-like layer on top for cases where I probably want to write C.
an_ko 11y ago
As an experiment, I've tried replacing C with Terra for a few projects. It feels like C++, except without the constant guilt. Having a proper hygienic macro system (instead of C's very limited string-replacement "macros" or C++ template hell) is a really helpful abstraction, and the compile-time stuff you can do can basically act as a build system. So I got excited and planned to use it to make a game, but although it can import existing C libraries [1], it can't import global variables, constants, or macros that take arguments, which puts a whole lot of existing libraries just out of reach. This doesn't seem to have stopped Leonard Ritter from using it for Nowhere anyway, though being a crazy genius, he seems to have written his whole environment from scratch. [2] [3] [1]: http://terralang.org/api.html#c-backwards-compatibility http://terralang.org/api.html#c-backwards-compatibility [2]: http://blog.duangle.com/2015/01/conspire-programming-environment-for.html http://blog.duangle.com/2015/01/conspire-programming-environ... [3]: https://bitbucket.org/duangle/nonelang/src https://bitbucket.org/duangle/nonelang/src
- nickpsecurity 11y ago
  "Having a proper hygienic macro system (instead of C's very limited string-replacement "macros" or C++ template hell) is a really helpful abstraction, and the compile-time stuff you can do can basically act as a build system." Exactly. I long ago came to the conclusion that any modern systems language should have a hygienic, macro system to balance productivity/maintenance against performance. Just makes things so much easier. That morons can abuse it is no reason not to have it in the language as they abuse everything haha.
- shahbazac 11y ago
  How is terra's performance? You sort of compared it to C++, can it actually be compared with C++ in terms of latency performance?
  - an_ko 11y ago
    Very good. I didn't benchmark it myself, but the language is designed for high-performance computing, and the authors' PLDI paper [1] shows it is competitive. Given that at the lowest levels, Terra basically reimplements C, and both compile to LLVM IR, similar performance is unsurprising to me. The main point that gives Terra an edge is that its macro system makes implementing and testing potential optimisations easier. [1]: http://terralang.org/pldi071-devito.pdf http://terralang.org/pldi071-devito.pdf § 6. Evaluation
haberman 11y ago
Previous discussion: https://news.ycombinator.com/item?id=5702793 https://news.ycombinator.com/item?id=5702793
- rch 11y ago
  Have a quick look at the 'past' link below the article title. It's a handy new feature.
  - nickpsecurity 11y ago
    Didn't know it was there. Thanks for the tip!
  - haberman 11y ago
    Interesting, but I don't think it's useful because it doesn't actually tell you if previous discussion exists or not until you actually click it. I'm not going to click "past" on every story for the 1% chance that it will actually have results. For it to be useful (to me at least), it should only show up if previous discussion exists, and it should be more prominent.
    - mturmon 11y ago
      Insightful comment. I thought the "past" link was cool when it was pointed out to me. Sometimes past discussions are superior to the current one. But in practice, I almost never use "past", except when I'm pretty sure I have seen the piece before. Your suggestion might fix the issue.
nickpsecurity 11y ago
Site is down right now. Here's Wayback Machine: https://web.archive.org/web/20150424222319/http://terralang.org/ https://web.archive.org/web/20150424222319/http://terralang....
badwolf93 11y ago
there's nothing wrong with Lua
Sleaker 11y ago
This looks very cool, my only concern is that luajit is lagging behind LUA and is in need of a new maintainer, how bound to luajit is terra for JIT compilation features? Is it something the terra developers are going to help with?

140 points by charlieegan3 11y ago

laarc 11y ago

Terra seemed promising when I ran across it a month ago. Does anyone know if it's possible to use LuaJIT's ffi module while writing Terra code? What I mean is, there is a large body of existing code that I'd like to run alongside Terra. But if the choice is Terra xor LuaJIT, this becomes difficult.

an_ko 11y ago
You might be looking for http://terralang.org/api.html http://terralang.org/api.html § Managing Terra Values from Lua. Quoting: > We provide wrappers around LuaJIT’s FFI API that allow you to allocate and manipulate Terra objects directly from Lua.
corysama 11y ago
You can use Lua, Terra or both at runtime. The Lua part is LuaJIT including the ffi. The Terra part is LLVM and can link directly to C (maybe C++?). Lua can be used as a metaprogramming system to produce a static Terra .o or exe. Or Lua can be used as a runtime that can JIT and invoke Terra functions on the fly.

andrewmcwatters 11y ago

So this is LuaJIT's ffi with some bastardized syntax that immediately becomes incompatible with any other Lua syntax parser, because "pretty." Just no.

mpdehaan2 11y ago
It's been a long time since I've done Lua (2005/2006 IIRC?), but I remember Lua's C embedding API being remarkably excellent already. C calling Lua, calling back into C, back and forth across the boundary was absolutely painless. Compared with experiencing that in many other languages, at the time, I was convinced that was as easy as it is. (Plus, the size of the lua runtime made it great for embedded projects, whereas now embedded computing has gotten so much more powerful you can really include almost any language stack if you really want to) As a result, I'm not positive why I need a C-like layer on top for cases where I probably want to write C.
- andrewmcwatters 11y ago
  Exactly. Anyone who has ever actually done embedded work with Lua knows you can hardly get any closer in terms of "low-level" outside of the C API besides using LuaJIT's ffi. If anything this is actually a high level abstraction on top of LuaJIT's ffi, so this is far from low level. God forbid anyone criticize anything on HN.
  - haberman 11y ago
    I think Terra functions are AOT compiled and have explicit strong types. So it's quite a different thing that LuaJIT. For example, you could use Terra to actually write a JIT.
    - andrewmcwatters 11y ago
      Why would you use Terra to write a JIT? It uses LLVM. To write JIT in JIT?
      - haberman 11y ago
        Think of it as a programming environment for generating LLVM IR that is a lot more convenient than using LLVM's C++ API.
      - coldtea 11y ago
        Sort of like people use Java to write JVM languages?
  - vbit 11y ago
    The front page describes three distinct use cases for Terra. While criticism is welcome, it certainly appears that you are missing the uses for Terra. Consider the third case - you can write Terra functions (cleaner syntax than C) and use Lua as a meta-programming language that produce more Terra functions (vastly superior to C macros) which then compile down via LLVM to machine code. Your final code has no Lua pieces.
mpdehaan2 11y ago
It's been a long time since I've done Lua (2005/2006 IIRC?), but I remember Lua's C embedding API being remarkably excellent already. C calling Lua, calling back into C, back and forth across the boundary was absolutely painless. Compared with experiencing that in many other languages, at the time, I was convinced that was as easy as it is. (Plus, the size of the lua runtime made it great for embedded projects, whereas now embedded computing has gotten so much more powerful you can really include almost any language stack if you really want to) As a result, I'm not positive why I need a C-like layer on top for cases where I probably want to write C.

an_ko 11y ago

As an experiment, I've tried replacing C with Terra for a few projects. It feels like C++, except without the constant guilt. Having a proper hygienic macro system (instead of C's very limited string-replacement "macros" or C++ template hell) is a really helpful abstraction, and the compile-time stuff you can do can basically act as a build system. So I got excited and planned to use it to make a game, but although it can import existing C libraries [1], it can't import global variables, constants, or macros that take arguments, which puts a whole lot of existing libraries just out of reach. This doesn't seem to have stopped Leonard Ritter from using it for Nowhere anyway, though being a crazy genius, he seems to have written his whole environment from scratch. [2] [3] [1]: http://terralang.org/api.html#c-backwards-compatibility http://terralang.org/api.html#c-backwards-compatibility [2]: http://blog.duangle.com/2015/01/conspire-programming-environment-for.html http://blog.duangle.com/2015/01/conspire-programming-environ... [3]: https://bitbucket.org/duangle/nonelang/src https://bitbucket.org/duangle/nonelang/src

nickpsecurity 11y ago
"Having a proper hygienic macro system (instead of C's very limited string-replacement "macros" or C++ template hell) is a really helpful abstraction, and the compile-time stuff you can do can basically act as a build system." Exactly. I long ago came to the conclusion that any modern systems language should have a hygienic, macro system to balance productivity/maintenance against performance. Just makes things so much easier. That morons can abuse it is no reason not to have it in the language as they abuse everything haha.
shahbazac 11y ago
How is terra's performance? You sort of compared it to C++, can it actually be compared with C++ in terms of latency performance?
- an_ko 11y ago
  Very good. I didn't benchmark it myself, but the language is designed for high-performance computing, and the authors' PLDI paper [1] shows it is competitive. Given that at the lowest levels, Terra basically reimplements C, and both compile to LLVM IR, similar performance is unsurprising to me. The main point that gives Terra an edge is that its macro system makes implementing and testing potential optimisations easier. [1]: http://terralang.org/pldi071-devito.pdf http://terralang.org/pldi071-devito.pdf § 6. Evaluation

haberman 11y ago

Previous discussion: https://news.ycombinator.com/item?id=5702793 https://news.ycombinator.com/item?id=5702793

rch 11y ago
Have a quick look at the 'past' link below the article title. It's a handy new feature.
- nickpsecurity 11y ago
  Didn't know it was there. Thanks for the tip!
- haberman 11y ago
  Interesting, but I don't think it's useful because it doesn't actually tell you if previous discussion exists or not until you actually click it. I'm not going to click "past" on every story for the 1% chance that it will actually have results. For it to be useful (to me at least), it should only show up if previous discussion exists, and it should be more prominent.
  - mturmon 11y ago
    Insightful comment. I thought the "past" link was cool when it was pointed out to me. Sometimes past discussions are superior to the current one. But in practice, I almost never use "past", except when I'm pretty sure I have seen the piece before. Your suggestion might fix the issue.

nickpsecurity 11y ago

Site is down right now. Here's Wayback Machine: https://web.archive.org/web/20150424222319/http://terralang.org/ https://web.archive.org/web/20150424222319/http://terralang....

badwolf93 11y ago

there's nothing wrong with Lua

Sleaker 11y ago

This looks very cool, my only concern is that luajit is lagging behind LUA and is in need of a new maintainer, how bound to luajit is terra for JIT compilation features? Is it something the terra developers are going to help with?