Re: FYI: rpmbuild gets slightly stricter, files listed twice now terminates build

On Jan 24, 2012, at 4:10 PM, Jeffrey Johnson wrote:

> 
> "works" yes
> 
> But lets get the trash out of *.spec please:
> 	deleting %dir is KISS, no fuss, no muss
> 
> Note that you are almost certain to hit a real "brain fart"
> in that code soon. I've gotten burned with
> 	implementations not meeting expectations
> every time I've touched the code. I'm sure you will *ahem* step in something shortly.
> 
> And proper engineering with explicit regression tests under buildbots
> before rebuilding RPM is the best fix there: nice and quiet, not disruptive
> to anyone else.
> 
> Try-and-see please … or learn the hard way, that "works" too ;-)
> 

Just saying, not bitching.

But since you are currently active in this code, lets
try to get a RFE employed (and with an RPM blueprint at launchpad,
this is just preliminaries on e-mail)

There's a recurrent RFE for multiple %{buildroot}'s that is tied
into the Fts(3) traversal and pattern matching.

Drilling multiple %{buildroot} into rpm build isn't hard: Fts(3)
will easily traverse multiply tooted tree's.

There are several usage cases for multiple build root's

1) building kernels (and other complex source trees) for multiple
versions of CFLAGS (which can include -m32 etc) in a single build.

2) building multiple versions (like python-[23].[0123456789]) in a single build.

The hard part here is designing the syntax -- without using macro magic --
that causes the least amount of breakage to "legacy compatibility".

The actual ability to traverse multiply rooted trees became possible
when Fts(3) replace few(3)/nftw(3) nearly a decade ago.

Gnaw on that bone for a bit please: make yourself useful!

Just in case: I'm teasing w/o a smiley. About "useful": the rest is an actual RFE.

73 de Jeff