Merge lp://staging/~raof/mir/prober-drm-device-probe into lp://staging/~mir-team/mir/trunk

This is better than what we've currently got, but I'd prefer not to merge it yet because what we currently have accidentally works on my hybrid laptop and this only works about half the time; when the intel card gets to be the first drm device.

Now actually selects a card that can possibly drive an output!

Makes sense

FAILED: Continuous integration, rev:768
http://jenkins.qa.ubuntu.com/job/mir-ci/813/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1021
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/906
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/51/console

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/813/rebuild

FAILED: Continuous integration, rev:769
http://jenkins.qa.ubuntu.com/job/mir-ci/815/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1024
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/909
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/53/console
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-vm-ci-build/./distribution=quantal,flavor=amd64/487

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/815/rebuild

201 + close(tmp_fd);

We also need to reset tmp_fd to -1. Otherwise we may end up returning an inappropriate fd instead of failing.

156 + udev_enumerate *enumerator = udev_enumerate_new(udev.ctx);
170 + udev_device *dev = udev_device_new_from_syspath(udev.ctx, sys_path);

Not a blocker for this MP, but ideally we would have RAII classes for udev resources to avoid the explicit new/unref matching.

Some style nits:

128 + const char *sys_path = udev_list_entry_get_name(device);

char const*

97 + char *busid = drmGetBusid(fd);
117 + udev_enumerate *children = udev_enumerate_new(udev.ctx);
248 + udev *ctx;
270 + int is_appropriate_device(UdevHelper const& udev, udev_device *dev);
... and at other places

'*' with type

+int mggh::DRMHelper::is_appropriate_device(UdevHelper const &udev, udev_device *drm_device)

'&' and '*' with type

124 + if (!udev_enumerate_scan_devices(children)) {
127 + udev_list_entry_foreach(device, devices) {
132 + if (strcmp(sys_path, udev_device_get_syspath(drm_device))) {
168 + udev_list_entry_foreach(device, devices) {
... and at other places

Opening brace should be on a separate line.

323 + .WillByDefault(DoAll(InvokeWithoutArgs([this]() {this->busid = (char *)malloc(10);}),

static_cast<char*>()

FAILED: Continuous integration, rev:770
http://jenkins.qa.ubuntu.com/job/mir-ci/816/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1026
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/911
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/54/console
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-vm-ci-build/./distribution=quantal,flavor=amd64/488

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/816/rebuild

@Alf: If you want to make a better UdevHelper class a blocker for the MP, I can certainly accommodate that ☺

Barring that (and getting the CI passing), stylistic comments should all be resolved.

FAILED: Continuous integration, rev:771
http://jenkins.qa.ubuntu.com/job/mir-ci/818/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1029
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/914/console
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/56/console
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-vm-ci-build/./distribution=quantal,flavor=amd64/489

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/818/rebuild

FAILED: Continuous integration, rev:773
http://jenkins.qa.ubuntu.com/job/mir-ci/819/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1031
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/916/console
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/57/console

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/819/rebuild

401 + va_list ap;
402 + mode_t mode;
403 + va_start(ap, flags);
404 + mode = va_arg(ap, mode_t);
405 + va_end(ap);
406 +
407 + return (*real_open)(path, flags, mode);

The C/C++ calling convention is to push parameters on the stack in reverse order. This means a function never needs to know exactly how many parameters the caller really passed. Yes, really.

So you should just implement the 3-parameter version of open and not bother with va_list stuff, I think.

Looks good as a first step. I am looking forward to:

1. 155 + // TODO: Wrap this up in a nice class

2. Tests to cover interaction with various udev configurations. If we continue with the link-time seam for injecting udev, perhaps it would be worth it to set up an infrastructure to create configurations easily (e.g. like we do with FakeDRMResources for KMS).

I think this is worth having integration or "acceptance testing" for. Especially if we expect our usage of udev to grow. It looks cumbersome to test, but will only become more so!

FAILED: Continuous integration, rev:774
http://jenkins.qa.ubuntu.com/job/mir-ci/833/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1048
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/933
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/71/console

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/833/rebuild

FAILED: Continuous integration, rev:775
http://jenkins.qa.ubuntu.com/job/mir-ci/835/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1050
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/935
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/73/console

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/835/rebuild

FAILED: Continuous integration, rev:776
http://jenkins.qa.ubuntu.com/job/mir-ci/836/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1053
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/938
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/74/console

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/836/rebuild

Abstain. There are enough issues blocking this without my opinion right now.

Please make umockdev optional:

-- checking for module 'umockdev-1.0'
-- package 'umockdev-1.0' not found
CMake Error at /usr/share/cmake-2.8/Modules/FindPkgConfig.cmake:279 (message):
  A required package was not found
Call Stack (most recent call first):
  /usr/share/cmake-2.8/Modules/FindPkgConfig.cmake:333 (_pkg_check_modules_internal)
  CMakeLists.txt:131 (pkg_check_modules)

Not just for the sake of raring, but particularly for other distros that don't have umockdev packages.

^^^ I mean just disable those tests when umockdev is missing. It shouldn't be a build failure.

FAILED: Continuous integration, rev:780
http://jenkins.qa.ubuntu.com/job/mir-ci/864/
Executed test runs:
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1098/console
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/984/console
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/102/console
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-vm-ci-build/./distribution=quantal,flavor=amd64/518

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/864/rebuild

FAILED: Continuous integration, rev:781
http://jenkins.qa.ubuntu.com/job/mir-ci/867/
Executed test runs:
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1102/console
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/987
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/105/console

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/867/rebuild

FAILED: Continuous integration, rev:783
http://jenkins.qa.ubuntu.com/job/mir-ci/874/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1113
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/998
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/112/console
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-vm-ci-build/./distribution=quantal,flavor=amd64/527

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/874/rebuild

FAILED: Continuous integration, rev:785
http://jenkins.qa.ubuntu.com/job/mir-ci/875/
Executed test runs:
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1114/console
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/999
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/113
        deb: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/113/artifact/work/output/*zip*/output.zip
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-vm-ci-build/./distribution=quantal,flavor=amd64/528

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/875/rebuild

FAILED: Continuous integration, rev:786
http://jenkins.qa.ubuntu.com/job/mir-ci/876/
Executed test runs:
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1115/console
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/1000
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/114
        deb: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/114/artifact/work/output/*zip*/output.zip
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-vm-ci-build/./distribution=quantal,flavor=amd64/529

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/876/rebuild

I think this is a step forward, but now I get all my tests skipped if I'm missing umockdev. I think that's extreme. We should only skip those tests (GBM?) that are not possible without umockdev.

I would like to see most of the tests still runnable without umockdev, as they are today in lp:mir.

I don't yet approve, but also understand that my vote is not a veto.

I think the test-suite should be an all-or-nothing affair - if valgrind didn't take ages I'd be agitating to drop the non-memcheck option. The more optional bits of the test-suite there are the more likely it is that you accidentally miss out.

Dropping the test-suite entirely when you don't have the prerequisites for all of it makes it very clear.

PASSED: Continuous integration, rev:787
http://jenkins.qa.ubuntu.com/job/mir-ci/877/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1118
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/1003
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/115
        deb: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-ci/115/artifact/work/output/*zip*/output.zip

Click here to trigger a rebuild:
http://s-jenkins:8080/job/mir-ci/877/rebuild

LGTM, the udev functionality should see some refactoring but that can happen in a follow-up MP.

FAILED: Autolanding.
Unapproved changes made after approval.
http://jenkins.qa.ubuntu.com/job/mir-autolanding/395/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1119
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/1004
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-autolanding/49
        deb: http://jenkins.qa.ubuntu.com/job/mir-saucy-amd64-autolanding/49/artifact/work/output/*zip*/output.zip