mesonbuild: update to 0.51.2

[InuSasha]

since meson 0.51.0, a host build environment ist needed for cross-compiling (instead of target build environment)

build and boot-tested on RPi2

[MilhouseVH]

Wow thanks for this, what a PITA! 😄

I find the reference to generic in setup_toolchain generic target a little confusing - I realise it could be anything (other than a valid toolchain ie. meson) but wonder if we were to flip the args around, maybe we can let the second argument (ie. toolchain) become optional, ie.

setup_toolchain target
setup_toolchain host
setup_toolchain target meson

What do you think? This would mean the case statement in setup_toolchain() becomes:

  case ${1}:${2} in
    target:meson)
    target:*)
    host:*)
  esac

does that help improve anything, as we no longer need dummy values? I don't know if we will ever need to support additional toolchains (ie. cmake etc.) like this in future (hope not!)

Also, we could add a small proxy helper create_meson_conf() since we already pass the target:

create_meson_conf()
{
  case ${1} in
    host|bootstrap) create_meson_conf_host ${1} ${2};;
    init|target) create_meson_conf_target ${1} ${2};;
  esac
}

then just call create_meson_conf ${TARGET} ${MESON_CONF} from scripts/build, it doesn't need to know/care about the host or target differences.

[InuSasha]

setup_toolchain
the parameter order of setup_toolchain change be changed.

create_meson_conf
i dropped the proxy, because we do not need this extra step. The proxy will only increase the compexity of code, imho. and we need it, only on one position.

[MilhouseVH]

the parameter order of setup_toolchain change be changed.

Thanks.

The proxy will only increase the compexity of code, imho. and we need it, only on one position.

Ok, thanks.

[InuSasha]

dropped wrongly changed gtk3

[MilhouseVH]

RPi/RPi2/Generic images and all addons built OK with meson-0.51.2. The resulting images all boot and run correctly (as far as I can tell).

I've compared the last published meson-0.49.1 images (#0827) with the new meson-0.51.2 test images (#0828) and there are some fairly significant file size changes (ignore the pvr.mythtv noise that crept into the overnight #0828 test build):

RPi: http://ix.io/1TJe (5.3MB increase with meson-0.52.1)
RPi2: http://ix.io/1TJf (5.3MB increase with meson-0.52.1)
Generic: http://ix.io/1TJg {9.6MB reduction with meson-0.52.1)

For RPi/RPi2 most of the increase comes from systemd and libinput, and we don't appear to be stripping these systemd/libinput files when building with meson-0.51.2

As for Generic, I can't explain the almost 10MB size reduction... as far as I can tell the Intel i915 driver is working correctly in a Skylake i5 NUC when built with meson-0.51.2. Maybe a different -On optimisation to what we were using before...?

[InuSasha]

look like the striping was not done

[5schatten]

Compiles & runs extended builds for Generic & RPi2 fine for me 👍

[InuSasha]

strip is working, now.
ready to push
(rabased to master, for better comparision)

[MilhouseVH]

Thanks @InuSasha - I'll re-run my RPi/RPi2/Generic tests.

[MilhouseVH]

Looks good now - many thanks!

RPi: http://ix.io/1Uix
RPi2: http://ix.io/1Uiy
Generic: http://ix.io/1Uiz

(the leftmost #0904 column is for the new meson-0.51.2 build without GLVND, the rightmost is meson-0.49.1 with GLVND)

We still get a very nice reduction in the Generic graphic drivers.

I'm going to build all the add-ons for RPi/RPi2/Generic overnight, just to make sure they all build OK.

[MilhouseVH]

All add-ons successful!

[5schatten]

@InuSasha could you rebase the mesa package please?

[InuSasha]

rebased