Doh!

        gesture = "two_finger_tap";
        specifiers =     {
            coordinate =         (
                100,
                100
            );
        };
        "{}" =     {
        };
    }

I have visually confirmed that building from Xcode and building from scripts causes the special configuration file to be the most recent.

• launching with :xcodebuild
• launching with :xctestctl

To test, remove the delay in cucumber/features/steps/shared.rb#wait_for_app.

The static analyzer is picking up a couple of errors. We should be running in CI and failing on non-zero exit.

@sapieneptus One of the problems is in CBXDevice#dictionaryRepresentation; I seem to recall that you had a macro or there was a builtin macro for ensuring object or NSNull? Am I wrong?

What

Spikes for integrating CalabashXCUITestServer with the rest of calabash ecosystem.

Spike Acceptance Criteria

• Get a calabash cucumber test running (with the following limited API)
• Have some sort of API for the following actions:
• tap
• swipe
• type_text

Spike 1: Self Contained XCUITest API

Description

This API should query CalabashXCUITestServer directly and perform gestures directly.

Projects Touched:

• Need to tweak existing Calabus engine to handle requests with index
• Need to create an API in calabash-ios

Extra TODOs

• query(ID or Marked, { index: optional } )

CC: @krukow @jmoody I would love feedback as to whether this is a sane way to go about this.
@jraczak

I expected return an HTTP error.

May 17 19:46:38 aft XCTRunner[33527]: POST /1.0/gesture {
        gesture = tap;
        options =     {
        };
        specifiers =     {
            coordinate =         {
                x = 50;
                y = 50;
            };
        };
    }

See the discussion in #103

RunLoop::XCUITest::HTTPError: Could not parse response 'Unhandled endpoint: POST /1.0/tap -- http://127.0.0.1:27753/
Params: {
    wildcards =     (
        "1.0/tap"
    );
}
Body: {
    id = General;
}'; the app has probably crashed

I could check the response first but I would rather reason with and report a JSON object in the error message.

The xcodebuild process running the DeviceAgent is blocking the Xcode.app from performing indexing which in turn blocks commands like indexing.

This is not a good experience for the user. I am not sure yet if this affects projects other than the DeviceAgent project. If it does, this effectively means the end of Calabash as a BDD tool.

There are two cases to understand and deal with:

1. Orphaned xcodebuild processes. I took a shot at managing the xcodebuild process in RunLoop earlier and the results were not good. Sending TERM to xcodebuild threw a bunch of UI crash reports. Users may or may not see these - my machine is configured to generate a UI crash report for every crash. This is not the default.
2. User is developing an app using Calabash BDD. A Calabash test has run (the AUT is still open) or tests are running. The user goes back to Xcode to make some change and finds Xcode is not in a good state for development (stuck indexing). The user will have to quit the Calabash test.

ATTN @krukow This needs to be considered when deciding how to distribute/launch the runner.

On iOS Simulators, we can detect that the app has launched using ps.

Can the server itself tell us that the app has been launched?

This Scenario is failing

Scenario: app with bundle id is not installed
Given com.example.MyApp is not installed on target device
And I call /session with com.example.MyApp
Then an error is raised

What actually happens is that /session replies that it has launched the app.

I want urls like this to work:

http://denis.local:27753

I think the repo should be named: ios-device-agent. This is in line with the other names in the Calabash organization.

If and when an Android agent is introduced, it should be called android-device-agent.

I have confirmed that the server responds to /health for at least a second after the /shutdown route is called.

This causes problems when launching on physical devices because the server is available even while the xcodebuild is in progress.

I prefer this pattern in format strings to avoid compiler warnings:

NSInteger my_int = 1;
CGFloat my_float = 0.1;
[NSString stringWithFormat:@"%@ %@", @(my_float), @(my_int)];

@sapieneptus

I saw in a recent PR that you tested rotate against Google Maps and a drawing application.

I would like to stand up some kind of integration tests in this repo.

I have some recommendations, but am interested to know if you have any suggestions.

I want to make a decision on this quickly because I want to have cucumbers for the 2-finger tap gesture.

UnitTestApp

This is what I do in the LPServer. I use the LPTestTarget as the unit test target and cucumber test target. These cucumber tests are transient - they are eventually deleted when the feature stabilizes. Most of the time I move the feature to Permissions, CalSmokeApp, or CalWebApp. Sometimes I just delete the feature (if it is not interesting).

There are some drawbacks to this approach.

1. I end up duplicating the feature and the implementation (e.g. a view that responds to double tap) later in another app.
2. The UnitTest app has two roles - unit testing and integration testing. I have never had a problem with this on the LPServer, but the unit test coverage is light in that repo.

The benefit to this approach is that there is no other target to maintain. It also supports a BDD workflow. I want to implement 2 finger tap? I write the cucumber, I add the server side implementation, and then I write the application code to add a view that demonstrates 2 finger tapping.

(Yet) Another App Target

This is only interesting if there is some problem using the UnitTest app as the integration test target.

The drawback here is that the we have to manage (yet) another target in the DeviceAgent target. This is not terrible, just something to consider.

CalSmoke-cal.app

We do like UITest does - we add a CalSmoke-cal.app directly to this repo. The benefit is that the CalSmoke-cal.app has views that cover the majority of gestures.

The drawback here is that maintaining the CalSmoke-cal.app. The static binary will need to be updated everytime there is a release or a change in the app. There is already a lot of work around what happens after a Calabash iOS release. And the CalSmokeApp will require more updates.

I would prefer not to use the CalSmokeApp in this way. This app has a clear role these days: demonstrate good Cucumber and that the Calabash API works. It should always be passing and always in the best possible state.

CalSmoke git submodule

I don't want to do this. I mention it as an option so you can argue for it if you want.

• two_finger_tap / multi_finger_tap // @jmoody
• two_finger_drag_coordinates / multi_finger_drag_coordinates // @sapieneptus
• rotate @sapieneptus

static int const CBX_MAX_NUM_FINGERS = 5; //TODO should this be 4?

I think 5 is right value.

Most of my work is done is done on the command line. The Server directory is a big bag of files which is hard to reason about from the Terminal.

These were added to the LPServer, but I don't recall why.

@krukow Should these be part of the DeviceAgent response headers? If so, assign to me.

 @"Access-Control-Allow-Methods" : @"GET, POST, PUT, DELETE, OPTIONS",
 @"Access-Control-Allow-Origin" : @"*",
 @"Access-Control-Allow-Credentials" : @"true",
 @"Access-Control-Max-Age" : @"3000",

Requires that the Calabash.keychain be stood up

The server API should not have compound gestures.

This enter-text-in requires a touch, a check for the keyboard, and then a keyboard interaction.

There are at least three possible error conditions here:

1. can't find the thing to touch
2. can't touch
3. keyboard doesn't appear
4. text does not get entered.

The client will have to handle these cases to produce a meaningful error for the user. This is on top of the implementation for enter_text.

I think it is easier for enter_text_in to be implemented on the client side.

• rotate_device
• screenshot
• send_app_to_background

Optional

• shake

What about

• set_location ?

  # DoubleTap.m
   //TODO: should this throw an exception?
    if ([self duration] >= 0.5) {
        NSLog(@"WARNING: Having a tap duration greater than 0.5 will trigger a LongPress, not a tap.");
    }

It makes sense to confine :duration to a range. Would it make more sense to drop the :duration key? The tap gestures have a fixed duration. Everything else is a long press?

Regardless of the outcome of this discussion, an invalid duration should return an :error response to the client.

It gets a little murky here when the server could launch more than one application.

Should the client be keeping track of the sessions that were created and be responsible for deleting each one?

The problem with the client keeping track is that when attaching to an existing CBXRunner (from the console), the client cannot know what apps have been launched with the CBXRunner. Does that make sense?

Implementing the same logic twice is a pain in the ass, especially when you don't know what implementation will be used.

- (XCSynthesizedEventRecord *)eventWithCoordinates:(NSArray<Coordinate *> *)coordinates {
- (XCTouchGesture *)gestureWithCoordinates:(NSArray<Coordinate *> *)coordinates {

/Users/moody/git/calabash/xcuitest/Server/AutomationActions/Gestures/EventSynthesis/CBXTouchEvent.m:19:22: implicit conversion loses integer precision: 'long long' to 'NSInteger' (aka 'int') [-Wshorten-64-to-32]

    te.orientation = path.orientation;
                   ~ ^~~~~~~~~~~~~~~~

• Turn on warnings fail the build for the cli unit test runner
• Turn on warnings fail the build for the CBXRunner build

@jmoody

Given the architecture we have decided, calabash-ios-server (henceforth CBServer) will need to POST some stuff to calabash-xcuitest-server (henceforth XCUITServer). Therefore, CBServer needs to know XCUITServer's port.

Approaches

1. XCUITServer has a static port. This is hardcoded into CBServer.
2. XCUITServer checks a range of ports and uses the first available. It must then POST its port to CBServer to be stored for the remainder of the session.
   As a result, since CBServer's port is technically dynamic as well, the calabash-ios-client will need to read the CBServer's port from its Info.plist and write that into XCUITServer's info plist prior to installation on device. This of course requires resigning.
3. User is required to add CBServer's port to XCUITServer's info.plist (somehow). Then XCUITServer can POST its port to CBServer for registration, however it may be determined.

For now, I'm using approach 1 since it's the easiest and quickest.

CC: @jmoody @krukow

Is there an reason not use

- (double)twoFingerTapInRect:(CGRect)arg1 
                              orientation:(UIInterfaceOrientation)arg2 
                                   handler:(XCEventGeneratorHandler)arg3;

Remove.

Use the same formatter as the LPServer to create consistent logs.

We want to enforce that the documentation is complete.

calabash/calabash-ios-server#349

      @"Access-Control-Allow-Methods" : @"GET, POST, PUT, DELETE, OPTIONS",
      @"Access-Control-Allow-Origin" : @"*",
      @"Access-Control-Allow-Credentials" : @"true",
      @"Access-Control-Max-Age" : @"3000",
      @"Content-Type" : "application/json; charset=utf-8"

Must supply a device id with -d (or plug in a device)
ERROR: Unable to resign.
make: *** [runner] Error 4

Why?

codesign does not need to have a device connected to sign an ipa.

And I have a device connected?

<string>file:///Users/chrisf/Library/Developer/Xcode/DerivedData/CBXDriver-erdosoyfksqnruaufqghalxwyuud/Build/Products/Debug-iphoneos/CBX-Runner.app/PlugIns/CBX.xctest/</string>
        <dict>

Should we replace this with a bogus path?

I am concerned about the use of int, float, and double primitive types.

We had many headaches upgrading the LPServer between Xcode 5 and 6 because we used primitives instead of the NS and CGFloat types. This was the switch between NSInteger being an int and a long (et. al.).

I propose we convert all int to NSInteger and unsigned int to NSUInteger. We should also use NSUInteger whenever we expect an unsigned integer.

We had a number of problems in the LPServer with long double type encoding and double vs. float values. If we use CGFloat everywhere we can avoid these problems.

I think I want this.

It would make extracting touch coordinates from query results easier on the client side.

I was testing cucumbers on a physical device this morning and had a FaceTime signing dialog pop up, then the UnitTestApp crashed.

2016-04-19 08:54:14.454 XCTRunner[1377:766729] POST /1.0/query {
    id = "gesture performed";
}
    t =     5.24s     Find the Application "sh.calab.UnitTestApp" 0x136e56150
    t =     6.25s         Find the Application "sh.calab.UnitTestApp" 0x136e56150 (retry 1)
    t =     7.26s         Find the Application "sh.calab.UnitTestApp" 0x136e56150 (retry 2)
Provided dSYM: [/var/containers/Bundle/Application/AE6506D1-22FF-432C-B3B0-98555202252E/CBX-Runner.app/PlugIns/CBX.xctest.dSYM/Contents/Resources/DWARF/CBX] does not match symbol owner 0x136f82da0
    t =     7.53s         Assertion Failure: UI Testing Failure - Application is not running, unable to get Accessibility data. Did you call -launch?
<unknown>:0: error: -[XCUITestDriver testRunner] : UI Testing Failure - Application is not running, unable to get Accessibility data. Did you call -launch?
    t =     7.54s     Get all elements bound by index for: Elements matching predicate '"gesture performed" IN identifiers'
2016-04-19 08:54:16.768 XCTRunner[1377:766729] (
    0   CoreFoundation                      0x0000000183e6ae50 <redacted> + 148
    1   libobjc.A.dylib                     0x00000001834cff80 objc_exception_throw + 56
    2   CoreFoundation                      0x0000000183e6ad80 <redacted> + 0
    3   XCTest                              0x00000001000680cc +[_XCTestCaseInterruptionException interruptTest] + 44
    4   XCTest                              0x000000010007c760 -[XCAXClient_iOS screenshotData] + 412
    5   XCTest                              0x0000000100099c28 __43-[XCUIElementQuery allElementsBoundByIndex]_block_invoke + 64
    6   XCTest                              0x000000010006c8f0 -[XCTestCase startActivityWithTitle:block:] + 364
    7   XCTest                              0x0000000100099bd0 -[XCUIElementQuery allElementsBoundByIndex] + 176
    8   CBX                                 0x00000001023bc624 -[Query execute] + 612
    9   CBX                                 0x000000010237b7d0 __24+[QueryRoutes getRoutes]_block_invoke_2 + 332
    10  CBX                                 0x00000001023b1488 __39+[CBXRoute handleRequestAndExceptions:]_block_invoke + 140
    11  CBX                                 0x00000001023866e4 -[RoutingHTTPServer handleRoute:withRequest:response:] + 376
    12  CBX                                 0x0000000102387584 __72-[RoutingHTTPServer routeMethod:withPath:parameters:request:connection:]_block_invoke + 5

The client replied with:

   Expected status code < 200, found 500.

      Server replied with:

      {"error"=>"Interrupting test"}
       (RunLoop::XCUITest::HTTPError)

                     "y" => 408.9999866485596,
                     "x" => -6.210803985595703e-05,
                "height" => 258.0000754594803,
                 "width" => 375.0000754594803

I believe that the JSON parser that UITest uses will explode on the "x" value.

WTF?

Instance variable used while 'self' is not set to the result of '[(super or self) init...]'

I cannot make heads or tails of what is going on in that method.

This will make the client work easier.

At the moment, I have to branch on the health route.

• Remove 'CB' prefix wherever posible
• Remove copyright comments from tops of all files
• Ensure that fb driver files reference the correct license