www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - why simple hello world binary is ~700KB

reply Bedros Hanounik <2bedros NOSPAMgmail.com> writes:
why when I compile this program under gdc I get 697KB of binary and when I
compile similar C code in gcc I get 6.5KB (both under Linux (GCC) 4.1.2
20070214 (  (gdc 0.24, using dmd 1.020))



import std.stdio;

int main ()
{
	writefln("Hello World!");
	    return 0;
}

how can I get small footprint binaries out of any D compiler? any difference
between gdc and dmd?
Oct 04 2007
parent reply Bedros Hanounik <2bedros NOSPAMgmail.com> writes:
Ok, I tried dmd 2.0 with the same code and got 271KB binary out of it. still
huge compared to 6.5KB generated by gcc.


Bedros Hanounik Wrote:


 why when I compile this program under gdc I get 697KB of binary and when I
compile similar C code in gcc I get 6.5KB (both under Linux (GCC) 4.1.2
20070214 (  (gdc 0.24, using dmd 1.020))
 
 
 
 import std.stdio;
 
 int main ()
 {
 	writefln("Hello World!");
 	    return 0;
 }
 
 how can I get small footprint binaries out of any D compiler? any difference
between gdc and dmd?
Oct 04 2007
parent reply Sean Kelly <sean f4.ca> writes:
Bedros Hanounik wrote:

 Ok, I tried dmd 2.0 with the same code and got 271KB binary out of it. still
huge compared to 6.5KB generated by gcc.



About the smallest "hello world" binary you'll get from a recent version 
of D is 100k.  Prior to 1.0, it was possible to get a 70k "hello world" 
program.  In every case however, cstdlib is statically linked to the app.


Sean
Oct 04 2007
next sibling parent reply Gregor Richards <Richards codu.org> writes:
Sean Kelly wrote:

 Bedros Hanounik wrote:

 Ok, I tried dmd 2.0 with the same code and got 271KB binary out of it. 
 still huge compared to 6.5KB generated by gcc.



 About the smallest "hello world" binary you'll get from a recent version 
 of D is 100k.  Prior to 1.0, it was possible to get a 70k "hello world" 
 program.  In every case however, cstdlib is statically linked to the app.
 
 
 Sean



Um, I hope you mean the /D/ stdlib? The C stdlib is never statically linked.

In fact, glibc is incredibly resistant to being statically linked in any 
scenario.

  - Gregor Richards
Oct 04 2007
parent Sean Kelly <sean f4.ca> writes:
Gregor Richards wrote:

 Sean Kelly wrote:

 Bedros Hanounik wrote:

 Ok, I tried dmd 2.0 with the same code and got 271KB binary out of 
 it. still huge compared to 6.5KB generated by gcc.



 About the smallest "hello world" binary you'll get from a recent 
 version of D is 100k.  Prior to 1.0, it was possible to get a 70k 
 "hello world" program.  In every case however, cstdlib is statically 
 linked to the app.



 Um, I hope you mean the /D/ stdlib? The C stdlib is never statically 
 linked.
 
 In fact, glibc is incredibly resistant to being statically linked in any 
 scenario.



You're right.  I meant the D standard library.  But Phobos' use of 
fprintf creates quite a few glibc dependencies, be they dynamically or 
statically resolved.  I'll admit that Tango does use fprintf with DMD 
for coverage and profiling output as well--this is a dependency I've 
been a tad too lazy to eliminate :-p


Sean
Oct 04 2007
prev sibling parent =?ISO-8859-1?Q?Anders_F_Bj=F6rklund?= <afb algonet.se> writes:
Sean Kelly wrote:

 Bedros Hanounik wrote:

 Ok, I tried dmd 2.0 with the same code and got 271KB binary out of it. 
 still huge compared to 6.5KB generated by gcc.



 About the smallest "hello world" binary you'll get from a recent version 
 of D is 100k.  Prior to 1.0, it was possible to get a 70k "hello world" 
 program.  In every case however, cstdlib is statically linked to the app.



s/cstdlib/dstdlib/

Last time I tested, D program was right inbetween the C and C++ programs
(assuming that static libstdc++ library was used, to make things "fair")

So while it was 10x the size of C, it was still half the size of C++...

Even more interesting, the Java version was 10x smaller* than C even :-)
And if you want a *really* small hello world, then /bin/sh is good too.

--anders

* not counting the JRE, that is. it was 100x slower to start, though...
Oct 05 2007