Re: [Letux-kernel] [BUG v5.2-rc1] ARM build broken

[H. Nikolaus Schaller]

Hi Kees,

> Am 22.05.2019 um 08:02 schrieb H. Nikolaus Schaller <hns@xxxxxxxxxxxxx>:
> 
> It turns out that HOSTCC and HOSTCXX are a gcc-4.9.4 installed through MacPorts.
> And CC is the self-bootstrapped cross-gcc-4.9.2 toolchain for arm.
> 
> The problem is likely that they do not know of each other, i.e. the required
> include and library search paths. Therefore HOSTCXX can't build plugins compatible
> with CC because it does not even know its existence. Or the gcc-4.9.4 from MacPorts
> is missing the gcc-plugin library to link against which would explain the HOSTLLD
> error message as well.
> 
> This seems not to be found by the tests of scripts/gcc-plugin.sh. I have to check why...

Well, here are some findings:
* gcc-plugin.sh does not check if the g++ is really capable of linking a gcc-plugin
* it only syntax-checks if g++ supports the designated initializer GNU extension
* my gcc from MacPorts passes this test
* but seems not to be able to properly build any gcc-plugin.so
* I also get the similar linker errors when trying this gcc-plugin example:
  https://thinkingeek.com/2015/08/16/a-simple-plugin-for-gcc-part-1/
* the step that fails with the MacPorts gcc is linking the gcc-plugin.so from the gcc-plugin.o file
* I have not found any hint where this step should get the missing symbols from
  There is no lgcc or similar available

So I'd suggest to expand gcc-plugin.sh to not only syntax-check but also test if a
plugin.so can be successfully linked.

Maybe something like

	$2 -std=gnu++98 -shared -o /tmp/plugin.so -I"${srctree}"/gcc-plugins -I"${gccplugins_dir}"/include 2>&1

Maybe the test source code should reference plugin_default_version_check() to trigger
the linker.

BR and thanks,
Nikolaus