Basics of gcc

[MartinW]

Apologies if this is covered somewhere - I may have even asked myself before

Is there somewhere that explains what all the different options like '-noiexemul' and the like do? And why you might want one over the other or add one in some cases but not others?

It feels a bit like this is stuff that you're meant to just know. Maybe it's because I've pretty much always used IDEs when it comes to modern development and maybe they shield you from this kind of stuff?

[Cowcat]

With gcc in amiga land "-noixemul" makes a turn regarding posix functions: Instead of using ixemul library it uses at link time libnix (no unix): Small size and less overhead, but with the possibility of missing some gnu unix functions. Or something like that.

[jotd]

on latest gcc 6.5 by beebo, if you don't use "-noixemul", it may crash randomly. Didn't happen with 6.4. But 6.5 is much better. So I strongly recommend using the switch, always.

I use IDEs to develop, but I always choose options myself.

my base options: -Wall -Wextra -Werror -noixemul

(all warnings, no ixemul)

debug mode: -O0 -g

(don't optimize, add debug symbols)

release mode: -O2 -DNDEBUG

standard optim switches+define NDEBUG so "assert" checks and other stuff are eliminated from build, making it smaller, faster, and unsafe

I still use real time exceptions (C++). I'd like to know if there's a switch that drastically reduces code size, as my program still takes 1.2MB (release mode, no debug), and gets down to 1.0MB when using "strip" on it (even if I never use -g ..., strange). Using powerpacker "reduces" executable to 400MB on disk...

[MartinW]

Thanks for the info guys.

[alkis]

Originally Posted by jotd

I still use real time exceptions (C++). I'd like to know if there's a switch that drastically reduces code size, as my program still takes 1.2MB (release mode, no debug), and gets down to 1.0MB when using "strip" on it (even if I never use -g ..., strange). Using powerpacker "reduces" executable to 400MB on disk...

try adding -s on release builds to have it stripped

[jotd]

Great it works: "-Wl,--strip-debug" failed, but -s worked...

[thyslo]

For a release build I use

-Os -Wall -Wno-unused-function -fomit-frame-pointer -fno-rtti -fno-exceptions -noixemul -Dlinux

And then strip it, might think about adding the -s directly instead of calling strip in the makefile.

Produces a 65k binary for 10.000 lines c++ code:-)

[MartinW]

Why -Dlinux ?

[jotd]

-Wall -Wno-unused-function are warnings options. They don't affect the code generation.

-Os: optimize for size (not necessarily for speed)

-fno-exceptions reduces by 20% in my case. But you have to remove/#ifdef all the try/catch statements.

-noixemul is just MANDATORY. Without it, it just locks up randomly on system calls.

-fno-rtti doesn't save a lot of space, and you probably cannot use dynamic_cast with that.

I'll try the other options (not -Dlinux )

[thyslo]

Originally Posted by MartinW

Why -Dlinux ?

I had nearly forgotten it, but a look in the code revealed: It defines a macro which is needed because sometimes I also build the project with StormC4.

The macro is used to #include <new> when build in in Linux/gcc6.5, whereas <new.h> must be included for StormC4.

[MartinW]

OK, so that one is pretty specific to your project, or use case. Understood