Merge lp:~vanvugt/mir/fix-1358210 into lp:mir

[Daniel van Vugt]

DemoRenderer: Don't try to create a texture of width zero. libGL might crash
and definitely won't produce anything useful. This simple regression was
caused by accidentally changing float corner_radius = 0.5f into an int in
revision 1845. (LP: #1358210)

Apparently nobody did manual testing of r1845 before approving it. We should watch that...

[PS Jenkins bot (ps-jenkins)]

FAILED: Continuous integration, rev:1848
http://jenkins.qa.ubuntu.com/job/mir-team-mir-development-branch-ci/2424/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-utopic-i386-build/1365
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-utopic-amd64-build/1371
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-mediumtests-utopic-touch/1348/console
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-team-mir-development-branch-utopic-amd64-ci/946
        deb: http://jenkins.qa.ubuntu.com/job/mir-team-mir-development-branch-utopic-amd64-ci/946/artifact/work/output/*zip*/output.zip
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-team-mir-development-branch-utopic-armhf-ci/947/console
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-mediumtests-builder-utopic-armhf/285/console

Click here to trigger a rebuild:
http://s-jenkins.ubuntu-ci:8080/job/mir-team-mir-development-branch-ci/2424/rebuild

review: Needs Fixing (continuous-integration)

[Kevin DuBois (kdub)]

The line is under test in the unit tests, and the branch was manually tested, although if its causing this problem, the problem slipped through. This line was changed to avoid the jenkins valgrind problem above:
==31577== valgrind: Unrecognised instruction at address 0xabe15d.
I'll investigate another solution

[Daniel van Vugt (vanvugt)]

"unhandled instruction: 0xEEBE 0x0ACC" is a valgrind bug. It almost certainly just means Valgrind has't implemented support for all the instruction types gcc is outputting.

So we need to avoid executing such code under Valgrind/tests until Valgrind support for armhf is improved a little.

[Daniel van Vugt (vanvugt)]

P.S. I tried using pragmas. GCC tells me that optimization pragmas are not supported for ARM :(

So that's why we have this workaround instead.

[PS Jenkins bot (ps-jenkins)]

FAILED: Continuous integration, rev:1850
http://jenkins.qa.ubuntu.com/job/mir-team-mir-development-branch-ci/2433/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-utopic-i386-build/1377
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-utopic-amd64-build/1383
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-mediumtests-utopic-touch/1360/console
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-team-mir-development-branch-utopic-amd64-ci/955
        deb: http://jenkins.qa.ubuntu.com/job/mir-team-mir-development-branch-utopic-amd64-ci/955/artifact/work/output/*zip*/output.zip
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-team-mir-development-branch-utopic-armhf-ci/956
        deb: http://jenkins.qa.ubuntu.com/job/mir-team-mir-development-branch-utopic-armhf-ci/956/artifact/work/output/*zip*/output.zip
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-mediumtests-builder-utopic-armhf/297
        deb: http://jenkins.qa.ubuntu.com/job/mir-mediumtests-builder-utopic-armhf/297/artifact/work/output/*zip*/output.zip
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-mediumtests-runner-mako/2441/console
    SUCCESS: http://s-jenkins.ubuntu-ci:8080/job/touch-flash-device/11713

Click here to trigger a rebuild:
http://s-jenkins.ubuntu-ci:8080/job/mir-team-mir-development-branch-ci/2433/rebuild

review: Needs Fixing (continuous-integration)

[Alan Griffiths (alan-griffiths)]

An ugly necessity I guess

review: Approve

[Alan Griffiths (alan-griffiths)]

[libprotobuf ERROR google/protobuf/descriptor_database.cc:57] File already exists in database: mir_protobuf_wire.proto
[libprotobuf FATAL google/protobuf/descriptor.cc:954] CHECK failed: generated_database_->Add(encoded_file_descriptor, size):
terminate called after throwing an instance of 'google::protobuf::FatalException'
what(): CHECK failed: generated_database_->Add(encoded_file_descriptor, size):
Aborted (core dumped)

lp:1358698

[Kevin DuBois (kdub)]

lgtm, sorry for the regression when the code was integrated. Hopefully the demo shell server can be tested the same was as the basic server in CI.

review: Approve

=== modified file 'examples/demo-shell/demo_renderer.cpp'
--- examples/demo-shell/demo_renderer.cpp	2014-08-13 12:20:38 +0000
+++ examples/demo-shell/demo_renderer.cpp	2014-08-19 07:11:37 +0000
@@ -77,12 +77,20 @@
     return corner;
 }
 
-GLuint generate_frame_corner_texture(int corner_radius,
+GLuint generate_frame_corner_texture(float corner_radius,
                                      Color const& color,
                                      GLubyte highlight)
 {
     int const height = 256;
-    int const width = height * corner_radius;
+/*
+ * GCC 4.9 with optimizations enabled will generate armhf NEON/VFP instructions
+ * here that are not understood/implemented by Valgrind (but are by hardware),
+ * causing Valgrind to crash:
+ *   eebe 0acc       vcvt.s32.f32    s0, s0, #8
+ * So this clumsy expression below tricks the compiler into not using those
+ * optimized ARM instructions that Valgrind doesn't support yet:
+ */
+    int const width = height / (1.0f / corner_radius);
     Color image[height * height]; // Worst case still much faster than the heap
 
     int const cx = width;