Rewriting Old Solaris C Code In Python Yielded A 17x Performance Improvement

[carewolf]

-Os is generally terrible for performance compared with -O2, or -O3.

Terrible is a strong phrasing. Worse is a better description. Not bad, is another.

[vladpetric]

Considering compilers that only do C aren't really a thing anymore and pure C is mostly used in limited embedded systems (where Python obviously isn't an option) talking about C++ was implied.

The issue is not the compilers, it is the projects. There's still a huge number of projects that are pure C (e.g., bash). In fact, the vast majority of Linux commands are written in pure C, not C++.

You can use ldd to check whether a binary links to a C++ lib or not.

[tuxd3v]

Lua didn't exactly get everything right either... *cough* 1-based indexing *cough*

In a sequence you see:
1234567890,
You only need to look into your keyboard
Lua has started as a scientific Language for mathematicians, they don't like the 0 indexing as primary by default..

Actually 1-indexing start is a miss-conception... you can start in positions '-1000' if you want to, its a hash table..

mytable          = {}
mytable[ -1000 ] = "Lua"
mytable[ 0 ]     = "Lua"
mytable[ "abc" ] = "Lua"

[pal666]

They are, but that's not really meaningful to say. Their implementation is 2500 lines of C code (- some python specific stuff). It makes a lot more sense to use Python than to write your own implementation of sets in C...

sane choice would be to replace your own implementation of sets in c with standard implementation of sets in c++. as a bonus point then you really replace c implementation with other language, not one c implementation with different c implementations

[pal666]

Actually 1-indexing start is a miss-conception... you can start in positions '-1000' if you want to, its a hash table..

you can't, list interface has implicit keys and starts with 1

[pal666]

The issue is not the compilers, it is the projects. There's still a huge number of projects that are pure C (e.g., bash). In fact, the vast majority of Linux commands are written in pure C, not C++.

c++ was designed to allow mixing with c and gradual switch from c. unlike python

[vladpetric]

c++ was designed to allow mixing with c and gradual switch from c. unlike python

True, but it's not really happening within the linux tools space. Again, use ldd to see what is c++.

[pal666]

True, but it's not really happening within the linux tools space

aren't gdb and gcc within the linux tools space?

Again, use ldd to see what is c++.

that way you can only see what is linked with c++ standard library

$ echo 'int main ( ) { }' | g++ -o a -x c++ - -Wl,--as-needed ; ldd a
        linux-vdso.so.1 (0x00007ffc19beb000)
        libc.so.6 => /lib64/libc.so.6 (0x00007f6ad214c000)
        /lib64/ld-linux-x86-64.so.2 (0x00007f6ad2343000)

[vladpetric]

aren't gdb and gcc within the linux tools space?
that way you can only see what is linked with c++ standard library

$ echo 'int main ( ) { }' | g++ -o a -x c++ - -Wl,--as-needed ; ldd a
linux-vdso.so.1 (0x00007ffc19beb000)
libc.so.6 => /lib64/libc.so.6 (0x00007f6ad214c000)
/lib64/ld-linux-x86-64.so.2 (0x00007f6ad2343000)

What I'm suggesting is an approximation, yes. But most code is not linked with --as-needed (and also not statically linked, which can obscure libraries).

Yes, you're absolutely right that gdb and gcc contain c++ code. Most tools in /usr/bin and /bin are still pure C though.

I don't think we're disagreeing about much here. I would like more code to be migrated to C++. And as you said, it can be done somewhat incrementally.

[tuxd3v]

you can't, list interface has implicit keys and starts with 1

Implicitly you start at 1, but nothing prevents you from starting in any other positions, or even with a string, because a table is a hash-table..