Re: [kvm-unit-tests PATCH v2 0/9] Support for out-of-tree builds

["Radim Krčmář" <rkrcmar@redhat.com>]

2017-05-12 17:39+0100, Alex Bennée:
> Hi,
> 
> Here is v2 of the out-of-tree build. There have been a number of
> changes following review and also inclusion of Thomas' ppc64 patch
> which completes the set.
> 
> The biggest change is the way I deal with creating build directories.
> There is now a make variable called OBJDIRS which sub-builds can add
> to. I had originally tried to be clever by expanding OBJDIRS into a
> bunch of templated mkdir's which could then be set as
> order-on-prerequisites as suggested by Drew. However it turns out to
> be very hard to add the directory of the target as a prerequisite even
> using hacks like GNU's secondary expansion. In the end I just created
> a directories target and made sure the all: target had it as the first
> thing. This breaks if someone tries to build an individual file
> without first building the tree but I assume most people don't build
> like that. If anyone else can come up with a neater solution I'm all
> ears ;-)

I noticed few cases that don't work:
 - SRC_DIR path contains whitespace
 - SRC_DIR was `./configure`d for another arch
 - SRC_DIR isn't clean (up-to-date results won't be rebuilt out-of-tree)
 - reconfiguring --arch in a out-of-tree directory

The in-tree build still works, out-of-tree in decent cases, Makefile
changes are reasonable, and reviewers seem content in front of a
first-grade bikeshedding material ...

Applied, thanks.

> Finally I've added a .travis.yml recipe. This really only works for
> github hosted repos but its better than nothing. Evidently the cross
> compilers complain about:
> 
>   lib/report.c:38:17: error: comparison between signed and unsigned integer expressions [-Werror=sign-compare]
>   assert_msg(len < sizeof(prefixes), "%d >= %lu", len, sizeof(prefixes));
> 
> and:
> 
>   lib/report.c:38:2: error: format '%lu' expects argument of type 'long unsigned int', but argument 6 has type 'unsigned int' [-Werror=format=]
>   assert_msg(len < sizeof(prefixes), "%d >= %lu", len, sizeof(prefixes));
> 
> But these can be addressed with separate patches.

(Should be fixed now.)