svn commit: r216473 - head/sbin/geom/class/eli

[Nathan Whitehorn]

On 12/16/10 13:04, Robert Watson wrote:
> On Thu, 16 Dec 2010, David O'Brien wrote:
>
>>>> Log:
>>>>   Bump WARNS to 6.
>>>>
>>>> Modified:
>>>>   head/sbin/geom/class/eli/Makefile
>>>
>>> FYI, this broke the tinderbox on arm, ia64, mips, and sparc64.
>>
>> Errr.  Reverted.  I built it on the architectures I had access to...
>
> For WARNS-related changes, I generally use "make universe" to test 
> across architectures.  This builds all of our architectures world + 
> all available kernels, and seems the most effective way to avoid the 
> above situations. (I've fallen into exactly the same trap...)
>
> The one thing to be cautious about is that make universe won't fail if 
> an individual build fails, so you need to check the logs to make sure 
> everything actually succeeded.

The trouble with make universe is that it has been broken for months and 
months now. ARM and powerpc64 are disconnected from the build entirely, 
as are big-endian and 64-bit MIPS, and an increasing number of ARM and 
PowerPC kernels depend on FDT tools not built by default, and so do not 
build. Build infrastructure changes also make it appear that the PowerPC 
GENERIC64 kernel is broken when it is not. This severely reduces the 
coverage of make universe for problems like this.

I have a patch at http://people.freebsd.org/~nwhitehorn/universe.diff 
that fixes both of these problems, by teaching the universe rule in 
src/Makefile about MACHINEs with multiple MACHINE_ARCHs and by enabling 
the build of the FDT tools by default, which adds about 300K to world. 
The way these are done is probably not optimal, but it is a better than 
the current situation and is a good stopgap. With the patch, all 
architectures succeed except for the ARM AVILA kernel, which seems 
genuinely broken, and the various 64-bit MIPS kernels, since 64-bit MIPS 
is not hooked up to the build yet. If I don't hear any objections, I 
would like to commit it on Wednesday the 22nd.
-Nathan