5 ms·

Agreed; the criticism here is way too harsh. In particular, I think it's somewhat amusing that people are giving him grief for wanting to "jump to C as soon as

Agreed; the criticism here is way too harsh. In particular, I think it's somewhat amusing that people are giving him grief for wanting to "jump to C as soon as possible." Any rational kernel developer wants to jump to C as soon as possible -- and wants as much of the system to be in C as possible. (Indeed, this is the ethos that drove Unix -- and separated it from the horrifically unportable systems that predated it.[1]) Further, there are some good ideas here -- in particular, jumping directly to long mode allows one to avoid much of the replaying of the history of the PC that one is historically required to do to boot an x86. Most civilians (and I dare say, most of the negative commenters here) have NFI how ugly this process is and how much of a drag it is on future development. With the decline of the PC, it's reasonable to believe that a future is coming in which x86 becomes primarily a server-side microprocessor -- and begins to shed much of the baggage from the misadventures of the 1980s and 1990s. All that said: there is a certain arrogance of youth here, and one hopes that when reality has sandblasted it away, the resulting humility will find its way to a formal computer science education and ultimately into professional software engineering; our discipline needs more people who have dared to write an OS from scratch, not fewer. [1] http://cm.bell-labs.com/who/dmr/cacm.html http://cm.bell-labs.com/who/dmr/cacm.html

ChuckMcM 14y ago
Totally agreed here. This is one of the reasons I really like the "build your own OS" course for the Raspberry Pi as the ARM ecosystem doesn't suffer from this legacy of system design. That said, the current SoC's sort of assume you are building a phone, but perhaps that will change.
iRobot 14y ago
The BIOS on the original IBM PC totally made it easy to code a homebrew bare metal OS which could access all the peripherals. These BIOS calls exists today in i7 based motherboards allowing OS's coded in the 80's to still (mostly) function on a modern Motherboard, its biggest fault was being 16 bit which meant all newer OS's needed to write there own once they switched to 32/64 bit mode. The BIOS layer made the PC easier. I find it frustrating that almost every new SOC I get, even from the same manufacturer requires me to re-code all my IO routines. If a common BIOS existed across the ARM/x86 which emulated the simpleness of the original BIOS concept and not the API hell you usually get now, there would be a lot more adventurers in homebrew OS's
javajosh 14y ago
Very well said, particularly our discipline needs more people who have dared to write an OS from scratch, not fewer. Truth is that I've thought about doing what the OP is talking about, but quailed at the complexity of it. That said, I'd be interested in making an OS that gets to JavaScript as fast as possible. Yes the system would be a dog, but sure would be fun to build.
- spencera 14y ago
  And amazingly fun to hack on.
- stcredzero 14y ago
  Smalltalk used to be an OS, and there have been some things resembling ports of the Smalltalk image to JavaScript. Also, JIT compilation is good enough, that such an OS need not be a dog, provided a clean architecture. BeOS showed that to be the case on machines orders of magnitude slower than today's.
  - ptramo 14y ago
    BeOS was written in C++ and compiled with gcc 2. For all intends and purposes the whole operating system was native code, certainly not running over a VM.
    - stcredzero 14y ago
      Yes, but it could achieve feats of multitasking and multimedia display mainstream OS could only achieve on machines with much more power. It showed that a 10X increase in efficiency is possible though architecture.
- charliesome 14y ago
  I did this once: http://github.com/charliesome/jsos http://github.com/charliesome/jsos
- tptacek 14y ago
  At least when it comes to systems programming, but probably across our whole field, most seeming impossibility has more to do with fear than real difficulty. I've spent a lot of time clumsily trying to paint a target on exactly what that fear is, but I've learned by having a lot of crazy stuff shoved into my face as a security researcher that if you can just get yourself to focus on a task and start coding, a lot of contraptions that seem so hard to build as to be implausible are actually the work of just a day or two; at least, the kernel of those contraptions (emulators, hypervisors, compilers, debuggers, drivers, interpreters, routing protocols, static analyzers). If you have a basic aptitude for coding, not knowing how hard a task is isn't a liability; it's a powerful asset. If you haven't actually done real work on an OS, you wish you thought it was as easy as this guy does.
nitrogen 14y ago
Dead comment from iRobot: iRobot 10 hours ago | link [dead] The BIOS on the original IBM PC totally made it easy to code a homebrew bare metal OS which could access all the peripherals. These BIOS calls exists today in i7 based motherboards allowing OS's coded in the 80's to still (mostly) function on a modern Motherboard, its biggest fault was being 16 bit which meant all newer OS's needed to write there own once they switched to 32/64 bit mode. The BIOS layer made the PC easier. I find it frustrating that almost every new SOC I get, even from the same manufacturer requires me to re-code all my IO routines. If a common BIOS existed across the ARM/x86 which emulated the simpleness of the original BIOS concept and not the API hell you usually get now, there would be a lot more adventurers in homebrew OS's Note to iRobot: it looks like your comment 165 days ago about "Melissa[sic] Gates" got your account killed.

ChuckMcM 14y ago

Totally agreed here. This is one of the reasons I really like the "build your own OS" course for the Raspberry Pi as the ARM ecosystem doesn't suffer from this legacy of system design. That said, the current SoC's sort of assume you are building a phone, but perhaps that will change.

iRobot 14y ago

The BIOS on the original IBM PC totally made it easy to code a homebrew bare metal OS which could access all the peripherals. These BIOS calls exists today in i7 based motherboards allowing OS's coded in the 80's to still (mostly) function on a modern Motherboard, its biggest fault was being 16 bit which meant all newer OS's needed to write there own once they switched to 32/64 bit mode. The BIOS layer made the PC easier. I find it frustrating that almost every new SOC I get, even from the same manufacturer requires me to re-code all my IO routines. If a common BIOS existed across the ARM/x86 which emulated the simpleness of the original BIOS concept and not the API hell you usually get now, there would be a lot more adventurers in homebrew OS's

javajosh 14y ago

Very well said, particularly our discipline needs more people who have dared to write an OS from scratch, not fewer. Truth is that I've thought about doing what the OP is talking about, but quailed at the complexity of it. That said, I'd be interested in making an OS that gets to JavaScript as fast as possible. Yes the system would be a dog, but sure would be fun to build.

spencera 14y ago
And amazingly fun to hack on.
stcredzero 14y ago
Smalltalk used to be an OS, and there have been some things resembling ports of the Smalltalk image to JavaScript. Also, JIT compilation is good enough, that such an OS need not be a dog, provided a clean architecture. BeOS showed that to be the case on machines orders of magnitude slower than today's.
- ptramo 14y ago
  BeOS was written in C++ and compiled with gcc 2. For all intends and purposes the whole operating system was native code, certainly not running over a VM.
  - stcredzero 14y ago
    Yes, but it could achieve feats of multitasking and multimedia display mainstream OS could only achieve on machines with much more power. It showed that a 10X increase in efficiency is possible though architecture.
charliesome 14y ago
I did this once: http://github.com/charliesome/jsos http://github.com/charliesome/jsos
tptacek 14y ago
At least when it comes to systems programming, but probably across our whole field, most seeming impossibility has more to do with fear than real difficulty. I've spent a lot of time clumsily trying to paint a target on exactly what that fear is, but I've learned by having a lot of crazy stuff shoved into my face as a security researcher that if you can just get yourself to focus on a task and start coding, a lot of contraptions that seem so hard to build as to be implausible are actually the work of just a day or two; at least, the kernel of those contraptions (emulators, hypervisors, compilers, debuggers, drivers, interpreters, routing protocols, static analyzers). If you have a basic aptitude for coding, not knowing how hard a task is isn't a liability; it's a powerful asset. If you haven't actually done real work on an OS, you wish you thought it was as easy as this guy does.

nitrogen 14y ago

Dead comment from iRobot: iRobot 10 hours ago | link [dead] The BIOS on the original IBM PC totally made it easy to code a homebrew bare metal OS which could access all the peripherals. These BIOS calls exists today in i7 based motherboards allowing OS's coded in the 80's to still (mostly) function on a modern Motherboard, its biggest fault was being 16 bit which meant all newer OS's needed to write there own once they switched to 32/64 bit mode. The BIOS layer made the PC easier. I find it frustrating that almost every new SOC I get, even from the same manufacturer requires me to re-code all my IO routines. If a common BIOS existed across the ARM/x86 which emulated the simpleness of the original BIOS concept and not the API hell you usually get now, there would be a lot more adventurers in homebrew OS's Note to iRobot: it looks like your comment 165 days ago about "Melissa[sic] Gates" got your account killed.