Your Appium hive-log build fails to install on a real Android phone and on a real iPhone. How do you triage each?

## Separate the artifact from the target Install is where a mobile suite fails first, and almost every install failure is one of two things: the file is not what this target accepts, or the target is not in a state that will take it. Deciding which, before changing anything, is the whole method. Everything below is a way of answering that one question quickly on a beekeeping hive-log run. Start by pinning down the facts you actually have: - Which artifact path did the session use, and what shape is it in? - Which target is this — virtual or physical, phone or simulator? - Did the install command itself fail, or did it succeed and something later break? - Is an app with the same identifier already on the device? That last one is answerable directly with `mobile: isAppInstalled`, and it changes the triage path more than anything else you can check in one call. ## The Android checklist 1. **Check the shape of the build.** A single `.apk` and an `.apks` archive both go through `mobile: installApp`. Loose split files do not — they need UiAutomator2's `mobile: installMultipleApks`, and passing one of them alone produces a package that installs and then dies at launch. 2. **Check for an existing copy.** Call `mobile: isAppInstalled`. If the package is already present, a replace can fail for reasons that have nothing to do with your file. 3. **Remove and retry.** `mobile: removeApp` followed by a fresh `mobile: installApp` is the cheapest decisive test: if it now succeeds, the problem was the resident copy, not the artifact. 4. **Check the artifact suits this hardware.** A build that omits the phone's ABI, or a device with no free storage, fails at install time and reads as a driver error rather than as a build problem. 5. **Read the server log.** The Android install runs host-side through `appium-adb`, so the underlying installer message usually travels back into the log and names the cause outright. ## The Apple checklist On hardware the first suspect is different, because the XCUITest driver takes two routes behind one method: `simctl` for simulators and `devicectl` for real devices. - **Is the artifact a hardware build?** A `.app` bundle produced for the Simulator is not the file a real device accepts. This is the single most common cause of an install that works all week on simulators and fails the first time it meets a phone. - **Is the device reachable at all?** A device install depends on the driver reaching the device first, and newer real devices are reached over an optional remote-XPC tunnel package the driver declares. A failure here is a connectivity failure wearing an install failure's clothes. - **Is a copy already present?** `mobile: isAppInstalled`, then `mobile: removeApp` and retry, exactly as on Android. The commands are shared even though the machinery is not. - **Is this really a provisioning question?** If the artifact is the right kind and the device is reachable, and the platform still refuses the file, you are looking at signing and provisioning — a separate subject, and the point at which this step hands off. ## What each platform's failure tends to look like | symptom | Android | Apple platforms | |---|---|---| | install succeeds, launch crashes | only the base split landed | rare — the artifact is one archive | | install refused with a copy present | replace-time conflict; remove and retry | remove and retry likewise | | install refused on hardware only | artifact does not match the device | Simulator build handed to a real device | | failure before any transfer | host-side adb layer problem | device not reachable over its transport | ## Where this stops being an install problem Three honest hand-offs keep triage from wandering: - **Signing and provisioning** of the application artifact is its own subject. Recognise the boundary and stop rather than guessing at certificates. - **Whether the device is prepared to take a session at all** — its developer-side setup and its automation agent — is a different question from whether it will take this file. - **Which build the lane should be running** is a test-design decision, not something to solve by editing a path in the setup step at three in the afternoon. The cheapest habit is to make the suite say, in one log line per session, which target it used and which artifact it installed with which command. Most of the triage above then collapses to reading that line — and the cases that do not are precisely the ones that belong to someone else's subject.