Software commit risk level

Abstract

A risk level of a software commit is assessed through the use of a classifier. The classifier may be generated based on attributes pertaining to previous commits and used to determine a risk level for deployment of a software commit into a production environment based on attributes extracted from the software commit.

Description

CROSS-REFERENCE TO RELATED APPLICATION

This application is a U.S. National Stage Application of and claims priority to international Patent Application No. PCT/US2013/067488, filed on Oct. 30, 2013, and entitled “SOFTWARE COMMIT RISK LEVEL,” the entire content of which is hereby incorporated in its entirety.

BACKGROUND

With the advent of on-line services, executable software often is available 24 hours per day, every day. Any changes or improvements to the software should function correctly when released into the production version of the software. The continuous availability of software code creates a stress point on the ability to test the code.

BRIEF DESCRIPTION OF THE DRAWINGS

For a detailed description of various examples, reference will now be made to the accompanying drawings in which:

FIG. 1 illustrates a software generation cycle in accordance with an example;

FIG. 2 shows a system for generating a classifier for use in assessing a risk level of a software commit in accordance with an example;

FIG. 3 illustrates a data structure in accordance with an example;

FIG. 4 illustrates another system for generating a classifier for use in assessing a risk level of a software commit in accordance with an example;

FIG. 5 shows a method for generating the classifier in accordance with an example;

FIG. 6 illustrates a system for assessing a risk level of a software commit in accordance with an example;

FIG. 7 illustrates another system for assessing a risk level of a software commit in accordance with an example;

FIG. 8 illustrates a method for assessing risk in accordance with an example; and

FIG. 9 illustrates another method for assessing risk in accordance with an example.

DETAILED DESCRIPTION

As used herein, the term “code” refers to all files related to an executable application. For example, code may be executable software itself (e.g., source or object code) as well as various files used by the executable code such as images, documentation, etc.

A model of a software code's creation cycle is illustrated in FIG. 1. The code is developed by software programmers during a software development stage. Then, the code is “committed” which means that the software programmer releases the code for testing. A “software commit” (or simply “commit”) refers to code that has been written by a programmer, but has not yet been inserted into a production environment. The testing phase can be quite time consuming. After the code is tested and all, if any, bugs are fixed, the code is released into production. If the code is a patch or an enhancement of a feature to an existing program (e.g., a new feature provided to an on-line service), releasing the code into production includes inserting the new code into the existing program and making it available to the end users.

As noted above, the testing portion of the cycle can be very intensive and time consuming. The examples described herein provide a technique by which the testing phase may be eliminated in at least some situations (as indicated by the “X” in FIG. 1 through the testing phase).

The disclosed technique involves the use of a supervised machine learning approach to generate a classifier which predicts a risk level with merging the software commit in to the production environment. The disclosed machine learning approach assumes that there is a common denominator to a “bad” commit, that is, a commit that introduces a bug into the production environment. The common denominator can be learned by generating a classifier based on prior commits. If the classifier deems a particular commit to be a good commit (e.g., a commit that does not introduce a bug), then the testing phase may be skipped and the commit released into production. However, if the classifier deems a particular commit to be a bad commit, further testing may be performed to fix any bugs. The classifier generates a risk level for each commit to designate the likelihood that a commit is good (bug free) or bad (likely contains a bug).

Testing may be skipped for good commits and only performed for bad commits, thereby generally reducing the amount of testing needing to be performed.

As used herein, referring to a commit as “good” means the commit is bug-free. Referring to a commit as “bad” means the commit contains a bug. The term “risk level” may indicate whether the commit is risky or not risky. That is, in some implementations, a risk level only contains two levels (risky, not risky). In other implementations, more than two risk levels are provided.

FIG. 2 illustrates a system that includes a data structure 100 which includes a plurality of software commits 102. Each commit has already been deemed to be good (bug free) or bad (contains a bug) based on prior actual usage of the commit in the production environment. A classifier engine 110 accesses the data structure to generate a classifier 120 based on the software commits from the data structure 100. The classifier is generated based on prior commits known to be good or bad and thus the classifier is usable to classify future commits as good or bad before such commits are actually released into production and known to be good or bad.

FIG. 3 provides an example of the data structure 100. As shown in FIG. 3, the data structure 100 comprises a table that includes a plurality of entries (e.g., rows). Each entry is used to store information about a separate software commit. Commits 1, 2, 3, . . . , n are illustrated in FIG. 3. Each software commit represents a piece of code that has been written by a software developer. The commits included in the data structure 100 have been previously released into production. As a result, it has been determined whether a given commit was good or not. For example, a commit for which a bug was later detected in the production environment may be deemed to have been bad while a commit for which no bug was detected may be deemed to have been good. A label 106 is assigned to each commit to indicate the success level of that commit. The label 106 may be designated as good or bad, 1 (e.g., good) or 0 (bad), or any other type of designation to indicate the success level of the commit as being good or bad. In some implementations, the label 106 may be one of three values +1, −1, or 0. The label value +1 indicates a good commit. The label value −1 indicates a bad commit. The label value 0 indicates a commit for which a technical problem prevented an assessment of the commit as being good or bad. The labels may be ascertained for a commit after its release into production and enough time has elapsed to allow an assessment of the commit as good or bad.

The commits in a given data structure 100 may be commits that pertain to different software projects. In other implementations, however, the data structure 100 may contain commits that al pertain to the same software project. For example, an on-line web service may have undergone multiple software updates. The various updates (commits) for the on-line web service are all stored in one data structure for subsequent use in estimating the risk level of a future commit for the same project. In such implementations, a separate data structure of software commits may be provided for each software project.

Each commit in the data structure 100 includes a plurality of attributes 104. The attributes include at least one attribute about the software commit. In at least some implementations, the attributes 104 do not pertain to the functionality or logic implemented by the code itself. Instead, the attributes characterize the environment associated with the software commit. Any number of attributes 104 is possible.

In some implementations, the attributes may be categorized in three classes. A first class of attributes includes attributes based on the commit itself without considering the entire project history (source control). Source control attributes include, for example, the number of days that have elapsed since the last commit for the project (project age). In general, there may be a correlation between the frequency with which a software project is updated and the likelihood of a bug. A software project that has not needed an update in a long time is generally less likely to experience a bug when an update is made as compared to a software project that has experienced a higher frequency of software updates. Another example of a source control attribute is the time of day when the commit was made. In general, commits made during normal business hours may be more reliable than commits made during off-hours.

A second class of attributes includes attributes based on the labels of the prior commits (previous labels). As explained above, a “label” is a characterization as to whether a software commit proved to be, for example, good (bug free) or bad (had a bug). The previous labels attributes include attributes pertaining to labels of prior commits such as, for example, whether the last commit had a “good” or “bad” label (label of last 1 commits), the average of the last three commits (label of last 3 commits), the minimum number of days that have elapsed since the last “good” commit of each file in the commit (last good commit days min), etc.

A third class of attributes includes attributes pertaining to the complexity of the committed code (code complexity). In general, a modification made to more complex code is more likely to result in a bug than a modification made to simpler code. Any of a variety of techniques for measuring code complexity may be used. One suitable complexity attribute, for example, is the number of lines of source code (SLOG). Other complexity metrics may include the cyclomatic complexity, Halstead complexity measures, and maintainability metrics. Cyclomatic complexity measures the number of linearly independent paths through the source code. Cyclomatic complexity may be computed using a control flow graph of the source code whose nodes correspond to indivisible groups of commands and whose directed edges connect two nodes if the second command might be executed immediately after the first command. Halstead complexity produces a difficulty measure that is related to the difficulty of the program to write or understand. The maintainability index is computed based on lines of code measures, Halstead complexity measures and possibly different or other measures. Additional or different complexity metrics are possible as well.

Table I below provides examples of various attributes, some of which are discussed above. Other implementations may use a different set of attributes.

TABLE I
Attribute	Description	Type
project age	Days passed since the last commit	source
		control
time of day	The hour when the commit was made.	source
sin	Because we want it to be circular,	control
	calculated as: sin(hour × 2 × Pi/23),
	where hour is an integer between 0 and 23
time of day	The hour when the commit was made.	source
cos	Because we want it to be circular,	control
	calculated as: cos(hour × 2 × Pi/23),
	where hour is an integer between 0 and 23
day of week	The day of the week when the commit was	source
sin	made. Because we want it to be circular,	control
	calculated as: sin(day × 2 × Pi/6), where
	day is an integer between 0 and 6
day of week	The day of the week when the commit was	source
cos	made. Because we want it to be circular,	control
	calculated as: cos(day × 2 × Pi/6), where
	day is an integer between 0 and 6
lines added	How many lines were added in total	source
		control
lines removed	How many lines were removed in total	source
		control
minimal added	Out of all the files in the commit, what is	source
lines	the minimal number of lines added	control
maximal added	Out of all the files in the commit, what is	source
lines	the maximal number of lines added	control
average added	Out of all the files in the commit, what is	source
lines	the average number of lines added	control
maximal	Out of all the files in the commit, what is	source
removed lines	the minimal number of lines removed	control
minimal	Out of all the files in the commit, what is	source
removed lines	the maximal number of lines removed	control
average	Out of all the files in the commit, what is	source
removed lines	the average number of lines removed	control
number of files	How many files were changed in this	source
	commit	control
number of	How many files in this commit are in the	source
sensitive files	“usual suspect” list	control
minimal age	How old (in days) is the newest file in the	source
	commit	control
maximal age	How old (in days) is the oldest file in the	source
	commit	control
average age	What the average age (in days) of the files	source
	in the commit	control
minimal times	How many changes were made to the file	source
changed	that changed the least	control
maximal times	How many changes were made to the file	source
changed	that changed the most	control
average times	How many changes on average were	source
changed	made to the files in this commit	control
minimal	What is the lowest change frequency in	source
changed	this commit. Change frequency is	control
frequency	calculated as: timesTheFileChanged/
	fileAge
maximal	What is the highest change frequency in	source
changed	this commit. Change frequency is	control
frequency	calculated as: timesTheFileChanged/
	fileAge
average	What is the average change frequency in	source
changed	this commit. Change frequency is	control
frequency	calculated as: timesTheFileChanged/
	fileAge
length of	How many characters in the comment for	source
comment	this commit	control
days since last	How many days past since the last commit	source
commit	to this project	control
label of last 1	Was the last commit labeled as “Good” or	previous
commits	“Bad”	labels
label of last 3	What is the average of the labels of the	previous
commits	last 3 commits	labels
label of last 5	What is the average of the labels of the	previous
commits	last 5 commits	labels
last good	Calculate how many days past since the	previous
commit days	last “Good” commit of each file in this	labels
min	commit, and return the minimum
last good	Calculate how many days past since the	previous
commit days	last “Good” commit of each file in this	labels
max	commit, and return the maximum
last good	Calculate how many days past since the	previous
commit days	last “Good” commit of each file in this	labels
avg	commit, and return the average
last good	Calculate how many commits past since	previous
commit count	the last “Good” commit of each file in this	labels
min	commit, and return the minimum
last good	Calculate how many commits past since	previous
commit count	the last “Good” commit of each file in this	labels
max	commit, and return the maximum
last good	Calculate how many commits past since	previous
commit count	the last “Good” commit of each file in this	labels
avg	commit, and return the average
last bad	Calculate how many days past since the	previous
commit days	last “Bad” commit of each file in this	labels
min,	commit, and return the minimum
last bad	Calculate how many days past since the	previous
commit days	last “Bad” commit of each file in this	labels
max	commit, and return the maximum
last bad	Calculate how many days past since the	previous
commit days	last “Bad” commit of each file in this	labels
avg	commit, and return the average
last bad	Calculate how many commits past since	previous
commit count	the last “Bad” commit of each file in this	labels
min	commit, and return the minimum
last bad	Calculate how many commits past since	previous
commit count	the last “Bad” commit of each file in this	labels
max	commit, and return the maximum
last bad	Calculate how many commits past since	previous
commit count	the last “Bad” commit of each file in this	labels
avg	commit, and return the average
label streak,	If the label of the last commit is X, how	previous
	many commits past since the last time the	labels
	label wasn't X
dependency	The number of modules that depend on	Code
graph	the modules that were changed	complexity
SLOCP total	Total number of source lines of code in all	Code
	modules of a commit	complexity
SLOCP avg	Average number of source lines of code	Code
	among the modules of a commit	complexity
SLOCP max	Max number of source lines of code in a	Code
	module of a commit	complexity
SLOCL total	Total number of source lines of code in all	Code
	modules of a commit	complexity
SLOCL avg	Average number of source lines of code	Code
	among the modules of a commit	complexity
SLOCL max	Max number of source lines of code in a	Code
	module of a commit	complexity
JSHINT total	Run a code analysis tool, see how many	Code
	errors it finds	complexity
JSHINT avg	Run a code analysis tool, see how many	Code
	errors it finds on average	complexity
JSHINT max	Run a code analysis tool, see what's the	Code
	maximal number of errors it finds	complexity
cyclomatic (M)	Total for all modules of commit,	Code
total	computed as (number of edges of control	complexity
	flow graph) − (number of nodes of
	graph) + (2)(number of connected
	components); connected points are exit
	nodes of the graph
cyclomatic (M)	Average cyclomatic measure across the	Code
avg	modules of the commit.	complexity
cyclomatic ( )	Maximum cyclomatic measure across the	Code
max	modules of the commit.	complexity
Halstead length	Total number of operators + total number	Code
(N) total	of operands for all modules of commit	complexity
Halstead length	Average length across the modules of the	Code
(N)_avg	commit.	complexity
Halstead length	Maximum length across the modules of	Code
(N) max	the commit.	complexity
Halstead	Total number of distinct operators and	Code
vocabulary (η)	distinct operands in all modules of commit	complexity
total
Halstead	Average number of distinct operators and	Code
vocabulary (η)	distinct operands across the modules of	complexity
avg	commit
Halstead	Maximum number of distinct operators	Code
vocabulary (η)	and distinct operands across the modules	complexity
max	of commit
Halstead	Difficulty measure for all modules of	Code
difficulty (D)	commit computed as (total number of	complexity
total	distinct operators)/2 × (total number of
	operands)/(total number of distinct
	operands)
Halstead	Average difficulty measure across the	Code
difficulty (D)	modules of commit	complexity
avg
Halstead	Maximum difficulty measure across the	Code
difficulty (D)	modules of the commit	complexity
max
Halstead	Program length (total number of operators	Code
volume (V)	and operands) × log2(program vocabulary)	complexity
total	computed for all modules of commit
Halstead	Average volume across the modules of	Code
volume (V)	the commit.	complexity
avg
Halstead	Maximum volume across the modules of	Code
volume (V)	the commit.	complexity
max
Halstead (E)	Difficulty metric × Volume, computed for	Code
effort total	all modules of commit.	complexity
Halstead effort	Average Effort computed across the	Code
(E) avg	modules of the commit.	complexity
Halstead effort	Maximum Effort computed across the	Code
(E) max,	modules of the commit.	complexity
Halstead bugs	Effort2/3/3000 or V/3000	Code
(B) total		complexity
Halstead bugs	Average number bugs across the modules	Code
(B) avg	of the commit	complexity
Halstead bugs	Maximum number bugs across the	Code
B) max	modules of the commit	complexity
Halstead time	E/18 seconds	Code
(T) total,		complexity
Halstead time	Average Time (T) across the modules of	Code
(T) avg	the commit	complexity
Halstead time	Maximum Time (T) across the modules of	Code
(T) max	the commit	complexity
maintainability	Total maintainability metric computed for	Code
total	all modules of the commit.	complexity
maintainability	Average maintainability metric across	Code
avg	the modules of the commit.	complexity
maintainability	Maximum maintainability metric across	Code
max	the modules of the commit.	complexity

FIG. 4 is an illustrative implementation of the system of FIG. 2. FIG. 4 includes a processor 130 coupled to a non-transitory computer-readable storage device 140. The storage device 140 includes any suitable type of non-volatile storage such as random access memory (RAM), compact disk read only memory (CDROM), a hard disk drive, etc. The storage device 140 includes a classifier module 142 that includes code that is executable by the processor 130 to implement the functionality of the classifier engine 110 of FIG. 2. The storage device 140 also includes the data structure 100 in which the attributes of about software commits 144 are stored for analysis by the classifier module 142. Any functionality described herein attributed to the classifier module 142 is applicable as well to the classifier engine 110, and vice versa.

FIG. 5 illustrates a method that may be implemented by the system of FIGS. 2, 4. At 150, the method includes adding attributes for a software commit to a data structure (e.g., data structure 100). The attributes to be added may be any or all of the attributes listed in Table I or different attributes as desired. For each commit whose attributes are added to the data structure, the method also includes adding a label for the commit. The label may be, for example, good (e.g., 1) or bad (e.g., 0) indicating whether or not the commit had a bug. Attributes for any number of commits can be added to the data structure.

At 154, the method includes generating a classifier based on the commits' attributes and corresponding labels in the data structure. In some examples, operation 154 is implemented as an off-line learning operation in which the data structure with the commits' attributes and labels are provided to the classifier engine 110. The classifier engine 110 may implement any suitable classifier technique such as a “leave-1-out” cross validation to generate the classifier 120. For example, for each commit in the data structure, the method removes that commit from the data structure, calculates the classifier on the remaining commits, run the newly computed classifier on the commit that was removed, and compare the risk level label returned by the classifier to the actual label of the removed commit.

Once the classifier 120 is generated, the classifier 120 can be used to assess a risk level for a given commit. The generated classifier 120 is usable to classify a new commit (i.e., a commit that is not already in the data structure or for which a label has not already been determined) as good or bad. FIG. 6 illustrates an example of a system for risk assessment of a commit. The system includes an attribute extraction engine 160 and a risk assessment engine 162. Attributes about a new commit to be analyzed are extracted by the attribute extraction engine. The extracted attributes may include at least some or all of the attributes listed in Table I. The extracted attributes are then provided to the risk assessment engine 162 which runs the previously generated classifier on the newly extracted attributes to classify the new software commit (e.g., as good or bad).

FIG. 7 is an example of an implementation of the system of FIG. 6. FIG. 7 shows a processor 170 coupled to a non-transitory computer-readable storage device 172. The storage device 172 includes any suitable type of non-volatile storage such as random access memory (RAM), compact disk read only memory (CDROM), a hard disk drive, etc. The storage device 172 includes an attribute extraction module 174 and a risk assessment module that comprise code that is executable by the processor 170 to implement the functionality of the attribute extraction engine 160 and risk assessment engine 162, respectively, of FIG. 6. The storage device 172 also includes a data structure 178 in which the attributes of software commits 180 are stored for use by the risk assessment module 176 to classify a new commit. Any functionality attributed to the attribute extraction module 174 and risk assessment module 176 is applicable as well to the corresponding attribute extraction engine 160 and risk assessment engine 162, and vice versa.

FIG. 8 shows an example of a method for assessing the risk level of a software commit. The method shown includes extracting attributes from a software commit (190). For example, some or all of the attributes listed in Table I can be extracted. At 192, the method further comprises determining a risk level of a software commit based on the extracted attributes. The classifier generated per the method of FIG. 5 may be run on the newly extracted attributes to generate the risk level.

FIG. 9 shows another method in which attributes for a software commit are extracted (200) as explained above. The attributes are provided at 202 to a classifier (e.g., the classifier generated in FIG. 5). At 204, the classifier is run on the newly extracted attributes to assess the risk level for the software commit.

The above discussion is meant to be illustrative of the principles and various embodiments of the present invention. Numerous variations and modifications will become apparent to those skilled in the art once the above disclosure is fully appreciated. It is intended that the following claims be interpreted to embrace all such variations and modifications.

Claims

1. A method, comprising: generating, by a processor, a classifier for attributes of previous software commits, wherein the previous software commits are code that has already been deployed, and wherein generating the classifier includes classifying the attributes of the previous software commits into three classes: a first class based on a frequency of update of the previous software commits, a second class based on successful levels of the previous software commits, and a third class based on code complexity of the previous software commits;extracting, by the processor, attributes pertaining to a new software commit, wherein the new software commit is code that has not been deployed; andrunning, by the processor, the classifier on the extracted attributes of the new software commit to determine a risk level for deployment of the new software commit into a production environment to determine whether or not to skip a testing phase,wherein generating the classifier of the previous software commits includes: removing one of the previous software commits from the data structure,calculating a classifier on the previous software commits remaining in the data structure,running the calculated classifier on the removed software commit, andcomparing a risk level label returned by the calculated classifier to an actual risk level label of the removed software commit.

2. The method of claim 1, wherein the attributes of the new software commit include an attribute that provides information about the new software commit, an attribute that provides information about a label for the new software commit, and an attribute indicative of the code complexity of the new software commit for which the risk level is determined.

3. The method of claim 1, wherein the risk level includes a plurality of levels.

4. The method of claim 1, further comprising: storing the previous software commits in a table of the data structure, the table including a column of the attributes of the previous software commits and a column of labels indicating the successful levels of the previous software commits.

5. A non-transitory computer-readable storage device containing instructions that, when executed by a processor, cause the processor to: generate a classifier for attributes of previous software commits, wherein the previous software commits are code that has already been deployed;determine a plurality of attributes pertaining to a new software commit, wherein the new software commit is code that has not been deployed; anduse the classifier to classify the attributes of the new software commit to determine a risk level for deployment of the new software commit into a production environment to determine whether or not to skip a testing phase,wherein, to generate the classifier, the instructions are to cause the processor to: remove one of the previous software commits stored in a data structure,calculate the classifier on the previous software commits remaining in the data structure,run the calculated classifier on the removed software commit, andcompare a risk level label returned by the calculated classifier to an actual risk level label of the removed software commit.

6. The non-transitory computer-readable storage device of claim 5, wherein, to generate the classifier for the attributes of the previous software commits, the instructions are to cause the processor to: classify the attributes of the previous software commits into three classes: a first class based on a frequency of update of the previous software commits, a second class based on successful levels of the previous software commits, and a third class based on code complexity of the previous software commits.

7. The non-transitory computer-readable storage device of claim 6, wherein, when executed, the instructions are to cause the processor to validate the classifier.

8. The non-transitory computer-readable storage device of claim 5 , wherein the attributes of the new software commit include an attribute that provides information about the new software commit and an attribute that provides information about a label for the new software commit.

9. The non-transitory computer-readable storage device of claim 5, wherein the attributes of the new software commit include an attribute indicative of code complexity of the new software commit for which the risk level is determined.

10. The non-transitory computer-readable storage device of claim 5, wherein the new software commit includes an update to an existing software application.

11. The non-transitory computer-readable storage device of claim 5, wherein the data structure includes a plurality of entries that store the attributes of the previous software commits and labels indicating successful levels of the previous software commits.

12. A system, comprising: a data structure to store a plurality of software commits that have already been deployed, each software commit in the data structure to include a plurality of attributes and a label indicating a success level of the software commit;a processor; anda memory storing instructions that when executed cause the processor to: generate a classifier for the plurality of attributes of the software commits from the data structure,determine an attribute of a new software commit that has not been deployed, andrun the classifier to classify the attribute of the new software commit to determine a risk level for deployment of the new software commit into a production environment to determine whether or not to skip a testing phase,wherein, to generate the classifier, the instructions are to cause the processor to: remove one of the previous software commits from the data structure,calculate the classifier on the previous software commits remaining in the data structure,run the calculated classifier on the removed software commit, andcompare a risk level label returned by the calculated classifier to an actual risk level label of the removed software commit.

13. The system of claim 12, wherein the attribute of the new software commit includes a measure of code complexity of the new software commit for which the risk level is determined.

14. The system of claim 12, wherein to generate the classifier, the instructions are to cause the processor to classify the plurality of attributes of the software commits into three classes: a first class based on a frequency of an update of the software commits, a second class based on successful levels of the software commits, and a third class based on code complexity of the software commits.