Re: Package download statistics

[John Marino <dragonflybsd@xxxxxxxxx>]

On 5/6/2013 14:52, Petr Janda wrote:
>> Now, the _very_ interesting (and unexpected) part is the less than 2% pkgsrc
>> packages usage.
>> Maybe it's just that pkgng is so much better, maybe it's that pkgsrc users
>> prefer to build software from source. Difficult to know.
>>
>> In any case, I don't see a reason to continue building pkgsrc packages with
>> these numbers.
>
> I wonder if this means clang should be included in base instead of GCC
> 4.4. Since Ports are a FreeBSD thing, and they have switched to clang,
> more and more ports are going to build with it.

I'm not making the connection.
For starters, just because FreeBSD has clang as a system compiler 
doesn't mean they use it to build packages.  It is similar to how most 
pkgsrc packages and dports will build with gcc 4.7, but we use gcc 4.4 
by default.  If a port doesn't build with clang, FreeBSD devs simply 
specify GCC and perhaps a minimum version of GCC depending on the port. 
  They are a long way off from eliminating GCC as a requirement to build 
ports.

Furthermore, even if they did, I don't make the connection why this 
would tip us towards clang.  Sure, it removes an obstacle, but in and of 
itself it's not a motivating factor.

Other than the fact that clang and llvm are written in C++ (addition 
bootstrap concerns), I've got nothing against it, but right now moving 
to clang is more of a liability than an asset.  GCC 4.7 is a nice little 
compiler and it supports OMP, something clang doesn't.

> Maybe it would be an interesting project to find out if GCC 4.7 builds
> more ports than clang or vice versa.

The gcc4.7 test would be easy enough.  Anybody can do it -- just add 
DRAGONFLY_CCVER=gcc47 to the /etc/make.conf and build away.  Clang 
probably would require some work with the mk/bsd.*.mk files since we 
wouldn't be using a base clang but a ports one.  I don't know if that's 
currently supported.

John