Re: defaulting to clang on FreeBSD 10.x

[Joerg Sonnenberger <joerg%britannica.bec.de@localhost>]

On Sun, Mar 22, 2015 at 08:01:33PM +0000, Jonathan Perkin wrote:
> * On 2015-03-22 at 18:12 GMT, Joerg Sonnenberger wrote:
> 
> > To add to this: It has been a long standing issues that most of the
> > TOOLS_PLATFORM.* stuff should not be hard-coded at all, but set by
> > bootstrap. One of the big reasons it is not done is because certain
> > people always insist that NetBSD must be handled specially without
> > bootstrap. The other big reason is noone has sit down to do it.
> > Stop adding any more complexity to the platform specific files. It
> > is a step into the wrong direction.
> 
> I disagree, this would make our illumos bootstrap kits incompatible
> across distributions, and may do the same with Linux.  It would be a
> big step backwards to lose the ability to distribute one set of
> packages and bootstrap kit which work across all variants of the same
> OPSYS, thanks to handling the few differences with some ".if exists()"
> logic.
> 
> It also doesn't seem to be a great win for user-friendliness moving
> over a hundred TOOLS_PLATFORM.* lines into mk.conf
> 
> And what happens when you want to introduce a new tool?  Force
> everyone to re-bootstrap?
> 
> It's quite possible I'm missing something here, as I see no benefit at
> all in doing this.

The way we handle it with exists() chain is a PITA and maintainance
nightmare. Heck, for Linux you likely need to search any combination of
/bin and /usr/bin just for the sake of someone being different.
I'm not saying they should be placed in mk.conf, but *somewhere* after
probing them. A package might be an option as well. That is kind of
missing the point I am trying to raise: the lists should be generated as
much as possible and we shouldn't have to maintain them by hand.

Joerg