Slashdot Mirror
Simpler "Hello World" Demonstrated In C
An anonymous reader writes "Wondering where all that bloat comes from, causing even the classic 'Hello world' to weigh in at 11 KB? An MIT programmer decided to make a Linux C program so simple, she could explain every byte of the assembly. She found that gcc was including libc even when you don't ask for it. The blog shows how to compile a much simpler 'Hello world,' using no libraries at all. This takes me back to the days of programming bare-metal on DOS!"
Interesting, but she does sort of sidestep the whole 'Hello World!' part of a hello world program.
FYI, Steve Jobs came up with the idea for the "Hello World" app.
He also holds the design patent on the touch wheel interface for it.
I'm sure "SlashdotMedia" will improve on all the wonders that Dice Holdings blessed us all with
http://web.archive.org/web/19991128041233/http://muppetlabs.com/~breadbox/software/tiny/teensy.html
November 1999. Slow news day much?
This post expresses my opinion, not that of my employer. And yes, IAAL.
Ok, this is wicked great in theory. Our programs have become bloated. We do have them taking up too much RAM, HD space, and CPU time. But after reading through this in-depth analysis I have to wonder if it's all worth it.
If we're willing to leave behind all pretenses of portability, we can make our program exit without having to link with anything else. First, though, we need to know how to make a system call under Linux.
Or I can just write it the old way, making the file size larger and not have to concern myself with portability and how to make system calls under Linux. After all that's what the whole point of this all was right?
45 bytes, huh? I can do it in....
#!/bin/sh
exit 42
18 bytes and it's portable across all Unices. Maybe the assembler version is faster, though?
Since when does a Hello World program not actually output anything?
I think you missed the point of the article.
The author is trying to highlight that amount of bloat in modern programs is so rampant that even "Hello World" is excessively over sized for what it accomplishes. How can we as programmers expect fast, efficient, lightweight code when our compiler (even ones as popular as gcc) are bloating the program without being asked to?
Why doesn't it fit in TFS?
If God forks the Universe every time you roll a die, he'd better have a damned good memory.
As to the point of this... we recently had a story about how computers had gotten "too big to understand".
And here we have a program, 45 bytes long, for which every single byte has a well-explained purpose. It's getting back to the bare metal and that's what makes it interesting. =)
Mainframers have been using this most simple of all utilities for decades - literally. The Wikipedia entry on it has a good write-up about this (literal) do-nothing program. It's whole purpose is to provide a mechanisim to to exploit the various functions contained in JCL to create, delete, and otherwise manipulate datasets on mainframes.
The wikipedia entry is here: http://en.wikipedia.org/wiki/IEFBR14
Ken
At the end, the code was assembler, and the compiler wasn't even called - just the linker. I can't say for sure where a C program ends and an assembler program begins, but I'm fairly certain that the last few iterations are assembler, based on the "let's do away with the compiler" suggestion.
Also, "Hello World" programs have to, you know, actually display the message "Hello World" - this is a program that isn't written in C, and doesn't write "Hello World" - care to revisit the title of this entry?
Ken
I understand the point of the article, and everything else mentioned here. I just think that the amount of time spent eliminating 11k from a program in this case is irrelevant because any real application is going to need libc. It's not like she needed to strip it out so it would fit inside a tiny corner of an embedded processor - she's probably running it on a PC with anywhere from 1GB - 4GB of RAM.
Thank God we have finally crossed this hurdle. The baffling complexity of helloworld.c is no longer an obstacle to world domination.
I think we can now finally say once and for all that 2010 will be the year of Linux on the desktop.
The whole point was learning ELF structure and why things were they way they were. Didn't you ever wonder why a "hello world" program took over 4000 bytes on a modern computer, when in 1980 a Commodore VIC-20 managed to play games in less than 4K of available memory? This wasn't a waste of time.
OK, when I first read this, I thought to myself, "now why in the hell would anyone care to do this?"
Then it dawned on me. One stoned programmer said to another....Yeah, that's probably how it went down. Both now, and back in 1979, when you could still smoke in the Data Center...
But my stupid build process that generates the bloated Hello World is much more maintainable. Now get off my lawn.
Colorless green Cthulhu waits dreaming furiously.
"An 11k app is not going to make me, or my computer, say 'Good Bye World'"
It is if your computer is a 38-cent Atmel AVR tiny 10, which only has enough space for 512 12-bit instruction words. This chip is about half the size of a sunflower seed, but is faster, and, in several ways, more powerful, than the original $5000 IBM PC from 1981.
Get away from the idea of Gigahertz desktops and $1000 laptops and join the real computer revolution!
For me, if it costs more that $5, it's not a computer that I take seriously. It's just a 20th-century digital processing appliance.
Shouldn't the linker remove unreferenced functions?
I've had this problem with gcc for a while, with C++ code. I was writing some embedded code, and I wanted to use some simple C++. Just by adding a #include of one of the stream libraries. the executable grew by 200k, even though none of it was referenced. The C++ code in iostream is template-generated anyway, so even if the compiler wanted to include the code, it can't until I instantiate it.
But it really is much simpler. The reason your 'average first year comp sci student' might find it less understandable is because they dont actually understand the bloated version either. Using a high-level language doesnt reduce complexity, quite the opposite in fact, it greatly increases actual complexity. It simply makes it easier to get something done without understanding it, and thus makes it easier to kid yourself into thinking you know what you are doing, when you dont.
=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Friends don't let friends enable ecmascript.
Yeah, but the 45-byte program doesn't say "Hello World". In fact, there's no example that I can find in TFA that outputs that message or any other. So the summary is incorrect on its face. TFA doesn't show a simpler "Hello World" program; it doesn't show any sort of "Hello World" program at all.
I feel cheated, and tricked into reading an article that didn't do what was advertised.
(It's not the author's fault, of course; the author didn't claim to be writing the sort of program that the summary talked about. Though I was a bit disappointed that only the first few examples were in C. The article was almost entirely about assembly-language programs. So again, I was a bit disappointed, since I was hoping to learn something about making C programs smaller. This was done only in the first example, and it was made smaller by removing its call on write() so it didn't output anything at all. I already understood that I can make programs smaller by removing all functionality. ;-)
Those who do study history are doomed to stand helplessly by while everyone else repeats it.
No Steve Jobs designed "iHello World", which is actually one byte larger than the standard hello world app, but he's litigating against everyone who creates "Hello World" since 100% of it is quite obviously a subset of "iHello World".
I drink to make other people interesting!
Maybe thinking like that is why we have to get 4 gigs of ram to run without slowing down lately. I bet every executable on the hard drive has an extra 11k that somebody thought was insignificant.
This is basic knowledge that ANYONE using c should know - that the startup library is linked to so it can find main.
This is almost as lame as their previous slashvertisement/product_whoring - where they claimed to have gotten around the Mythical Man-Month and quadrupled output - and it turned out that neither claim was true.
And their lame excuse, which I derided in this comment:
I'm sure we're not the only ones to have used embedded assembler in c programs.
I wasted too much time reading this one... nothing surprising about what I found in it. Step one, don't write it in C. Step two, stop linking to things that aren't needed. Step three, perform the functions contained in the library omitted manually. Step five, start cheating in the elf binary format.
The only thing interesting about it was that the article pointed out an interesting fact -- Linux will run inappropriately formatted binaries. BAD. Linux kernel people? Are you reading this? Fix it before someone figures out how to use this in making and executing more exploits.
The fact that helloworld.c compiles to 11k has less to do with bloat than it has to do with people generally not caring about 11k. You could get rid of that 11k, but to do so, you'd have to make trade offs that either make real programs either slower or bigger, or make compilation slower. Very few people would make those trade offs in the other direction. Those that do either use special purpose compilers or (more likely) write in assembly.
The cake is a pie
You're right! I'm going to throw my laptop out the windows right now! Reading slashdot will be so much more fun on a computer smaller than a sunflower seed.
"A week in the lab saves an hour in the library"
These days, costs of development and deployment, not runtime memory usage, are the limiting factors in software development.
http://outcampaign.org/
Guy reminds me of an old joke.
What's the difference between a bitch and a whore?
A whore fucks everyone. A bitch fucks everyone but you.
If you're actually programming on "bare metal", you're not really using DOS, are you? After all, DOS is an operating system -- a layer between your code and the hardware.
Most of the microprocessors in the world today have less than a few 10's of kilobytes of RAM. They tend to do useful things most of the time.
I am becoming gerund, destroyer of verbs.
One of the earliest processors I used had only 256 bits of RAM (Yes, that's right 256 B I T S of memory)
#hello world tiny program
.data
.text
.equ SYSCALL, 0x80
.equ SYS_EXIT, 1
.equ SYS_WRITE, 4
.equ STDOUT, 1
.section
hello:
.ascii "hello world!\n"
.section
.globl _start
_start:
movb $SYS_WRITE, %al #put write syscall in eax
movb $STDOUT, %bl #set stream to stdout
movl $hello, %ecx #give address of start of buffer to print
movb $13, %dl #how many characters of buffer to print
int $SYSCALL
movb $SYS_EXIT, %al
int $SYSCALL
The above is a tiny hello world program i wrote myself, it's worth noting that even the resulting binary is larger than it needs to be, I wound up with a 133 byte binary by moving the text string into the ELF header via hex editor, and changing the instruction data to point to the new addresses.
Kind of hard to get it smaller than that while keeping it in ELF format, considering the actual object code in the binary was something like 15 bytes with the data illegally in the header.
TFA explains it: main() isn't the true start of the program, _start is. That resides in ctrl.o, which fires off a bunch of setup stuff before calling __libc_start_main, which in turn kicks off main(), and off your program goes.
To put it as a car analogy: What she found is that turning the key to start doesn't just activate the starter, it also activates the airbag system, the traction control, and the radio too. And if all you want to do is start the engine to prove that it runs (ala Hello World!), then it's kind of silly to lug around all that extra "unnecessary" crap too.
Or something like that. Sadly i'm a better mechanic than a programmer (4yrs vs 1yr), but i'm working on fixing that. :)
Mod parent up. This is all a semantic game about where significant portions of functionality are stored (and thus counted or not). After all, back in the "pre bloatware" days, you'd have had to manage all of the complexities of machine management and I/O yourself. The assembly would have been much larger to achieve the same effect.
Yes, you can make the argument that Linux comes with screen I/O, a scheduler, memory management, etc. already, so that's just overhead, but as others have pointed out, you can say the same thing about bash. It comes everywhere and is just overhead.
STOP . AMERICA . NOW
c:\ xxx>debug :001D
-a
mov dx, 100
mov cx, 000D
mov bx, 1
mov ah, 40
int 21
mov ah, 4C
int 21
-f 111 "Hello World!"
-a100
mov dx, 0111
-r cx
-n c:\ xxx\ hello.com
-w
-q
c:\ xxx>hello.com
Hello World!
c:\ xxx>dir hello.com
03/18/2011 11:29 AM 29 HELLO.COM
Try programming a micro-controller and suddenly you'll be facing hardware limits that force you to favor small unreadable code over bigger more maintainable code. There is a solution for it though... comments! Lots of them :D
gcc for an AVR target doesn't make an 11k hello world, though.
Probably because that's an application where it matters, and a modern PC it doesn't matter at all.
Sent from my PDP-11
There are three links in the article summary. The first is to the Wikipedia entry for "Hello World"; the second is to an article about writing "Hello World" without libc; the third is to part II of the second, an examination of the ELF format and demonstrates the 45 byte program. The summary headline is rubbish. Whoever wrote it either (a); never read either article, or (b); deliberately sensationalized it by conflating the salient features of both articles, in which case they should be working for the tabloids.
Back in the early 1980s, I was doing development on MS-DOS 2.11 - the first real working version of MS-DOS that resembled Xenix more than CP/M.
I was using a combination of Lattice C and assembly language to do my day job. But I was upset about the libc bloat that Lattice C would drag into the program. Over the Christmas break, I sat down and wrote a tiny version of libc, with the 60% of the calls I actually used. Most of them were either thin wrappers on top of MS-DOS Int21 calls, assembly language implementations (the string functions), or reduced functionality (printf didn't handle strange alignments, floats or doubles), and custom startup/exit code. I also structured the library so that the linker would only link in functions that were actually used. For simple executables, I saw the on-disk file size drop from 10KB-20KB down to 400-600 bytes. Another thing that reduced on-disk file size was to create .com programs, rather than .exe programs.
I was also writing the handful of unix commands that I couldn't do without (ls, cat, cut, paste, grep, fgrep, etc). Since I was implementing dozens of Unix commands, each statically linked to libc, it was very important to reduce the over-all size of each executable. Most of the smaller trivial commands were less than 1KB in size. I think the largest was 4KB. I also had an emacs clone* that was 36KB when compiled and linked against my tiny lib.
For the longest time, I carried around a bootable MS-DOS 2.11 floppy, with my dozens of Unix commands, an emacs-like editor, Lattice C compiler, tiny libc, and some core MS-DOS programs. It allowed my to have my entire development environment on a floppy that I could stick in anyone's machine and make it usable.
* We had a source license for Mince, orphaned by Mark of the Unicorn, a tiny emacs-clone that ran on CP/M, MS-DOS, and Unix. We had enhanced it significantly.
I bet every executable on the hard drive has an extra 11k that somebody thought was insignificant.
So if you have, say, 1000 open processes, that means your computer is wasting 11 MB of RAM. Such inefficiency!
Actually, the reason you need 4 GB of RAM is because the programs you're using are far more complex than the ones that people were using when 256 MB was top-of-the-line. You may say, "But all I need is to read e-mail and browse the web!" -- except that nowadays those tasks involve rendering GUIs with Javascript, streaming and playing HD video in realtime, and doing constant full-text indexing in the background so that you can quickly search anything for any phrase. On top of that, in the background your operating system is trying to predict what you'll do next and prefetching blocks from your hard drive into RAM so that they'll already be cached when you actually need them.
Some of that RAM is honestly being taken up by insignificant chunks of data, but most of it really is being used.
Karma: Terrifying (mostly affected by atrocities you've committed)
OOP makes people lazy and gives them less of an understanding of what's actually going on.
All that OOP code you write gets translated back into something procedural, you know.
Edward@Tomato - /home/Edward/ man woman
man: no entry for woman in the manual.
"Qua!?"
It's funny that this always come up in conversations about bloat, because not everyone has to program for embedded code, because not everyone is programming embedded devices. It's almost like you guys are a subsubculture of programmer, to the point where many of you guys come off with the general attitude of being superior, when in fact, neither approach is superior, just different based on the situation.
/rant
What a shock this comes from Kdawson. I'm about one more kdawson article away from dumping slashdot. I can't imagine that all the people in the slashdot batcave aren't laughing at this tool.
I sometimes wonder if he just goes out and gets completely hammered at lunch then comes back and picks a few articles.
Hmmz, I was hoping my post was without any judgement about what is 'better' and more '133t' coding. Sorry you think otherwise :(
Some people like their code to run on OSes for grownups.
The fact that people would even still use C at all for anything anywhere ever shocks me.
I started writing device drivers in Ruby, and have never looked back.
In order to get Ruby to run on my system, I run it in an interpreter. The interpreter is written in Java, which is a much faster language and therefore more suitable as an interpreter.
The JVM on my system is written in C#. I know that C# is comparable to Java in terms of efficiency, but since this is a Windows machine, I figure it's "closer to the metal."
The implementation of the .NET framework on my computer (and the Windows operating system itself) is written in Ruby. Since I already have a Ruby interpreter on my system, this presents no problems.
I don't believe in time. It's a grand conspiracy designed to sell watches.
Not only more maintainable, but filesystems should use 4k per sector, specially on raid's for performance stuff discussed on this post. This means that in a decently configured modern system, anything under 4k will still occupy 4k on disk.
- Human knowledge belongs to the world
Apple defines system call APIs at the top of libc ...no static linking allowed.
This annoys people who like to link things statically, and those who want to make their own libc equivalents for things like embedded language interpreters and don't want to have to figure out vtables and dynamic linking.
But it also makes everyone else who likes binary compatibility, and Mac OS X historically getting faster with every release, extremely happy, by allowing the interface between the kernel and libc to be changed, without breaking their applications.
If you statically link, you can't do that. That's great, if your OS has pretty much no real commercial application base, and you are a technical enough person to "just recompile everything from source", but it's not so good when you are talking about an OS where commercial software is very important to customers. Customers who are either non-technical, or who are technical, but think recompiling something that was working just fine before the OS update is a complete waste of time. Lump me in with these last people: I don't believe in "bit rot", I just believe in lazy engineers not maintaining their code or defining their interfaces properly.
Yeah, if you want fast LMBench results on a null system call -- which keeps changing its definition so that it can't be gamed, exactly the same way you'd game it if you were a commercial application developer needing higher performance -- static linking seems great. But practically, most modern software is either CPU bound or I/O bound. If it's CPU bound, it spends all its time in user space, not making system calls. If it's I/O bound, it spends all its time waiting for whoever is on the other end of the network to send it more bytes. Either way, null system call performance is, frankly, unimportant to almost every possible application.
So static linking, and writing your system calls at the trap/sysenter/syscall level (with no way to change them when Intel or another chip vendor introduce a "new! optimized method of making system calls!", as has already happened twice in the past) is generally a pretty useless exercise.
-- Terry
Hello (World). I am from the future. The code you have provided a link to will not run on Windows 9.
It's too bad with all these things you "heard" that you didn't happen to hear that programs are written for environments other than Windows (or Linux, Mac OS, etc), and for devices other than PCs. It's unfortunate that you are so in the dark that you don't realize that there are entire industries that rely on devices that have tiny fractions of the memory and processor speed that you ignorantly assume that we all have access too. You probably have no idea how often you are affected by devices that run 100 times slower than the desktop PC you gave as an example, or also have 1,000 times less RAM. On some of these devices C is the most advanced language you can get short of writing a compiler or interpreter yourself.
Sure, pissing away storage space and waving a hand at execution efficiency is fine for some circumstances, but sometimes it's a luxury you can't afford. The world of software development is far bigger than the tiny little niche of programming you've been exposed to.
I suggest you use some "real" perspective, and reevaluate what a "real language" is.
"Single instance only" applied to all code - applications, libraries and OS. We often used to have 16 users on an 11/70 with 1MW of memory (ie 2MB) - all running the same program, so only one copy was resident. (or one of us was running the C or Fortran compilers :-)
isn't that how the BSDs do it today?
In RSX/11, programs could be "installed" so that they made their location on disk known to the OS, so when you ran a program, it was not necessary to search the file system for it. The location on disk, and offset to the entry point, was already known. A program could have multiple entry points (like grep, egrep etc), and libraries just used this mechanism. The dynamic linker stored the program (library) name and index into a table of entry points. I think early implementations statically linked the code to hack this stuff, before it became part of the OS.
If the search of the program you asked for found nothing, then the directories were searched. It was laziness of the users that meant the default strategy was used for most applications. I think VMS continued to support the install option, but I cant remember. I don't think Un*x ever did.
Programs only loaded the pages that were in use, and pages not in use were eventually swapped out. So huge programs did not take for ever to load - you loaded the first page, and jumped to it - then loaded which ever pages execution went to - so you did not need to spend years designing overlay strategies! This was possible because pages could load anywhere - the PDP11 supported "position independent code" (All modern 16 and 32 bit processors could still do this).
Don't you young people know anything? I know this, and I didn't even do computer science in college!
Get off my lawn.
Sent from my ASR33 using ASCII
I think VMS continued to support the install option, but I cant remember. I don't think Un*x ever did.
IIRC install in VMS was to register a privileged library with the OS. A library like that could do stuff the calling process couldn't do. Most likely it was install image.obj/priv=sysprv,setprv and so on.
My experience with RSX was with a traffic signal application called SCATS. I once interviewed for a job in a hospital where they supported ~60 users on a single machine, probably an 11/84.
Our SCATS systems had up to 16 DZ11 MUX cards for 128 serial lines. I have never seen a system which could handle that many interrupts and run so cool in the sense that it was always responsive regardless of load and it would chug away for years without showing any signs of stress.
BSD is as close as you would get to that with modern systems. Maybe QNX though I haven't worked with that OS.
http://michaelsmith.id.au
My God, are you saying that people should use the right tools and techniques for the job at hand, rather than applying the same limited ones to every problem they come across?
It's official. Most of you are morons.
*sigh*
Been there done that... on the PDP-11 in 1979.
And did you write up a nice article for other people to learn from what you had done?
I think the real value here is not that she did this, but that she wrote it up in a nice easy to read way so that you can follow her train of thought and get a feel for how one goes about tinkering with compilers and such.
This adds value for people like me who are not as smart as you. I could never have done this on a PDP-11 (although I did have access to one back in my days at university). I also previously would not have know enough to do this in Linux. But having read this article I feel I have learnt something and have a new insight into how linkers and libraries work. Who knows, maybe I will be able to do something similar myself after this learning experience, and for that I am grateful to Jessica for doing it, writing about it and (I'm guessing it was her) submitting it to /.
Now I shall respectfully step off your lawn.
I have a suggestion: If you write your JVM in Visual Basic instead of C#, it'll be portable, since most old microcomputers included BASIC in ROM. And, of course, .NET already brings Visual Basic.NET!
-dZ.
Carol vs. Ghost
...besides, high-level programmers often underestimate just how big a sector embedded programming is. The $IDIOTS_PET_LANGUAGE is for a PC. Now get me more RAM and better CPU for all the devices running embedded software, that are in my sight range as I look around:
- my cell phone.
- 6 different monitors (OSD doesn't happen magically. Something remembers the settings...)
- a videoserver
- 2 cheap switches
- a regulated power supply
- a heat-controlled soldering iron
- a regular phone
- 3 PC keyboards (hey, neither PS2 nor USB protocols happen by themselves)
- 3 computer mice (optical, meaning pretty advanced image analysis)
- my hand watch
- a battery charger
- a USB hub
- a security motion sensor
- an MP3 player
- a webcam
- a multimeter
- a car alarm remote
- a pendrive.
These all were programmed either in VHDL, Asembler, or C. The phone has some J2ME code too. Think of upgrading each of these devices so much that its firmware could be rewritten in, say, Perl. Or C#.
Also, think about how much embedded programming is in every PC. Each device controller has its own firmware... my bet is any average house contains more embedded programs (in embedded devices) than PC applications on the "family PC" and stored on media.
High-level programming languages are nice and have their place, but considering embedded "a niche not worthy of attention" is a bad mistake. The proportions between amounts of server:desktop:embedded software are much closer to 1:1:1 than most "high-level" programmers are willing to admit.
45 5F E1 04 22 CA 29 C4 93 3F 95 05 2B 79 2A B2
Was that in RSX11-M?
Version 6 UNIX. I didn't abuse a.out as badly as this example abuses elf, though.
Really, with a.out, it wasn't abuse. That format LIKED these kinds of games. When I was hacking in Forth I wrote a "snapshot" word that did something like
Where patch-to-executable took advantage of the fact that an a.out was basically a core dump with some extra segments... and you could leave those segments off if you needed to. :)
I did some nasty stuff on RSX-11, though. Portable file and terminal I/O was a pain in the butt, because text files were variant record files with each line having a count and a length and a record/line type field and IIRC occasional block alignment issues, so to read and write text files in Forth I had a FORTRAN main that called Forth through an assembly glue routine, then called back to FORTRAN for textfile I/O. Also got tired of FORTRAN formatted I/O so I wrote a version of sprintf for my RATFOR code that used assembly glue to implement varargs in FORTRAN. Ah, the good old days...
A better solution for everyone is to replace Kdawson with an editor who doesn't have shit for brains. This would give us more quality articles and less garbage without having to modify any settings at all.
Turning to a Linux advocate for thoughts on Microsoft is like asking Hitler how he felt about the Jews.
She found that gcc was including libc even when you don't ask for it.
This is basic knowledge that ANYONE using c should know - that the startup library is linked to so it can find main.
Okay, and where am I supposed to learn it from? That was new to me, after using gcc for a very long time.
I'm actually very happy that someone out there told me something that you think I should just know.
So it wasn't new to you? Don't read it.
How about 22 bytes?
:16
C:\>debug
-a 0100
0D39:0100 MOV AH,09
0D39:0102 MOV DX,0109
0D39:0105 INT 21
0D39:0107 INT 20
0D39:0109
-e 0109 'Hello world!$'
-r cx
CX 0000
-n hello.com
-w
Writing 00016 bytes
-q
C:\>HELLO.COM
Hello world!
C:\>dir hello.com
26/08/2009 10:48 22 HELLO.COM