Curiosity-driven android app automatic testing method

A curiosity-driven Android app automatic testing method, comprising following steps:

• 1) a left part representing a pre-processing assembly for achieving abstraction of an Android app state based on an Android interface structure, wherein the pre-processing assembly extracts the Android interface structure using an automatic framework UIautomator, maps the Android interface structure into the Android app state, and extracts an executable event set;
• 2) a right part being a curiosity-driven reinforcement learning module for maintaining a historical access state set and optimizing an exploration policy constantly to guide a test to find more new states under a guidance of a reward function based on a curiosity thought;
• and
• 3) a middle part representing an advanced guidance module of a deterministic finite automaton (DFA), wherein the DFA is constructed during operation of the advanced guidance module of the DFA to record all access states and frequencies thereof;
• and if the new states are not explored within a given time budget, an AndroidExplore selects the most curious state as a starting point of a next exploration according to global information of the DFA, so as to prevent from falling into a local optimum and improve an exploration probability of a deep function;
• wherein an Android app under test (AUT), a testing time, a maximum step number of an action sequence and a similarity threshold are taken as inputs;
• then, the Android AUT is constantly explored within a given testing time, the maximum step number of the action sequence is set, and when the step number is the maximum, the AUT is rebooted to start a new round of the test;
• after the test is started, the AndroidExplore maintains a historical state set M to store an explored state and is hereby guided to explore some different states, wherein, curious states;
• a pre-processing module abstracts an initial Android app state st and an executable action set At from the AUT;
• each round of the test is started in the initial Andriod app state st, an action at is selected and executed according to a policy, and the AUT responds to the action and converts the action into another state;
• if the AUT is converted into a fault state, a current action sequence is added into a fault case set, and this round of the test is finished;
• otherwise, the pre-processing module abstracts a next state st+1 and an action set At+1;
• and if within the given time budget, the AndroidExplore has not explored the new states, the DFA constructed during the operation provides advanced guidance, selects the most curious state as the starting point of the next exploration, and updates the action sequence;
• and
• when the next state st+1 is reached, the AndroidExplore compares the next state st+1 with the explored state in the historical state set M according to an Android app state similarity calculation method, and calculates a reward rt according to similarity values thereof and updates the historical state set M;
• then, the AndroidExplore trains a policy π according to conversion information (st, at, st+1) and the reward rt;
• in addition, during the operation, the DFA is dynamically constructed according to the conversion information (st, at, st+1) during the operation, and an access number of times of each conversion is updated;
• and finally, after the test is finished, the AUT is analyzed to generate a code coverage rate of this exploration;
• wherein the interface structure of a current Android app is analyzed using the automatic framework UIAutomator to acquire Activity and information of various view components, so as to abstract the initial Android app state st=(act, e1, e2, . . . , en), wherein act represents the Activity of a current Android App interface, ei={pathi, classnamei, xi, yl, widthi, heighti} represents the view components in the interface, the view components being arranged in a tree structure, wherein pathi represents a path of an element e;
• in the Android interface structure, classnamei represents a class name of e1, xi represents a horizontal ordinate of ei in the interface, yi represents a longitudinal ordinate of e;
• in the interface, widthi represents a width of ei in the interface, and heighti represents a height of ei in the interface;
• and
• the pre-processing module takes the AUT as an input and outputs the current state and an action set At;
• during the test, the AndroidExplore maintains the historical state set M to store an accessed state;
• first, the pre-processing module acquires the current Android interface structure using the automatic framework UIAutomator, analyzes the Activity and the information of various view components comprised in the current Android interface structure, and generates the current state according to the information of various view components;
• then, the pre-processing module traverses and analyzes attributes (e.g., clickable, scrollable) of each view component to infer and generate the executable action set At;
• in addition, a system-level event of mobile equipment is added into the executable action set At to expose more app faults;
• and finally, the current state is compared with the accessed states by adopting the Android app state similarity calculation method, and if the similarity is greater than a set threshold, the states are merged, and otherwise, the state is returned in form of a new state;
• wherein the AndroidExplore dynamically and incrementally constructs the DFA during the Android app test;
• the AndroidExplore dynamically constructs the DFA according to the conversion information (st, at, st+1) during the test first;
• and during the test, if there are no new states found within the given time budget, a conversion with highest curiosity is selected from the DFA based on the curiosity thought;
• during the Android app test, the AndroidExplore records a number of times of each conversion, defined as N(s, a, s′);
• one conversion (st, at, st+1) is executed, so the DFA and N(s, a, s′) are also updated, i.e., δ=δU{((st, at, st+1)} and N(st, at, st+1)=N(st, at, st+1)+1;
• and during the test, if an Reinforcement Learninq (RL) has not found the state within the given time budget, the RL selects a conversion of the conversion information (st, at, st+1 with the highest curiosity from the DFA based on the curiosity thought;
• and
• by taking the DFA constructed during the operation and a set N(s, a, s′) recording the number of conversions of a number of state transitions as inputs, a path capable of reaching the conversion with the highest curiosity is outputted;
• first, a state conversion (sk, ak, sk+1) with the least executing number of times is selected according to the set N(s, a, s′), then, the DFA is traversed according to a shortest path algorithm to seek for a shortest path tr capable of reaching the conversion;
• and finally, the shortest path tr is returned.