[ https://issues.jboss.org/browse/ARQ-1319?page=com.atlassian.jira.plugin.s... ] Aris Tzoumas edited comment on ARQ-1319 at 3/3/13 2:21 PM: ----------------------------------------------------------- Well Lukáš, you just found a *Bug* in {{CommandEventBus}}... The case is that the Container may send a {{CoverageDataCommand}} to the client *after* a response has already been sent to the client ({{ServletResponse}} {{OutputStream}} is closed). As a result, client's CommandEventBus may have already been closed (the {{After}} event has fired on client side) and thus it doesn't receive the coverage command. Actually, there is a race condition between the timer thread on {{CommandEventBus}} and the {{@After}} event that cancels the timer. Sometimes the {{TimerTask}} manages to process the command, sometimes it doesn't. That's why I was getting inconsistent coverage reports from Jacoco's reporting task when testing. There are two possible solutions to this problem: * Add a 100ms delay to {{CommandEventBus}} shutdown process, so that the TimerTask has time to process the last Command from the container (...if it does send one). * Make sure that all container commands happen before the {{After}} event fires on client. Although the first solution may be easier to implement, the second one is the right thing to do. That's why I have prepared the following two commits: * https://github.com/atzoum/arquillian-extension-warp/commit/6d284459402ebf... This one closes the {{HttpServletResponse}} only afterwards {{Manager}} shuts down. So, {{CommandEventBus}} stays open for as long it is needed. You may need to review this one, since you have a better understanding of Warp's internals. I basically create a {{NonWritingResponse}} at the beginning of {{WarpFilter}}'s processing, which actually gets written and closed only at the end of the processing. * https://github.com/atzoum/arquillian-extension-warp/commit/0fa31972597c0b... This one is an updated Test that should fail if the HTTP Response gets committed before {{AfterSuite}} event is called in the container. The above two commits are inside my branch: https://github.com/atzoum/arquillian-extension-warp/tree/race-condition This branch runs all tests successfully and behaves as expected. I am also attaching a screenshot with a correct coverage report on {{ftest}} project (https://issues.jboss.org/secure/attachment/12361047/ftest-coverage.jpg). If you want to add arquillian coverage on *{{jsf-ftest}}*, please don't forget to add the following dependencies on {{pom.xml}} {noformat} <dependency> <groupId>org.jboss.arquillian.extension</groupId> <artifactId>arquillian-jacoco</artifactId> <scope>test</scope> </dependency> <dependency> <groupId>org.jacoco</groupId> <artifactId>org.jacoco.core</artifactId> <scope>test</scope> </dependency> {noformat} was (Author: atzoum): Well Lukáš, you just found a *Bug* in {{CommandEventBus}}... The case is that the Container may send a {{CoverageDataCommand}} to the client *after* a response has already been sent to the client ({{ServletResponce}} {{OutputStream}} is closed). As a result, client's CommandEventBus may have already been closed (the {{After}} event has fired on client side) and thus it doesn't receive the coverage command. Actually, there is a race condition between the timer thread on {{CommandEventBus}} and the {{@After}} event that cancels the timer. Sometimes the {{TimerTask}} manages to process the command, sometimes it doesn't. That's why I was getting inconsistent coverage reports from Jacoco's reporting task when testing. There are two possible solutions to this problem: * Add a 100ms delay to {{CommandEventBus}} shutdown process, so that the TimerTask has time to process the last Command from the container (...if it does send one). * Make sure that all container commands happen before the {{After}} event fires on client. Although the first solution may be easier to implement, the second one is the right thing to do. That's why I have prepared the following two commits: * https://github.com/atzoum/arquillian-extension-warp/commit/6d284459402ebf... This one closes the {{HttpServletResponse}} only afterwards {{Manager}} shuts down. So, {{CommandEventBus}} stays open for as long it is needed. You may need to review this one, since you have a better understanding of Warp's internals. I basically create a {{NonWritingResponse}} at the beginning of {{WarpFilter}}'s processing, which actually gets written and closed only at the end of the processing. * https://github.com/atzoum/arquillian-extension-warp/commit/0fa31972597c0b... This one is an updated Test that should fail if the HTTP Response gets committed before {{AfterSuite}} event is called in the container. The above two commits are inside my branch: https://github.com/atzoum/arquillian-extension-warp/tree/race-condition This branch runs all tests successfully and behaves as expected. I am also attaching a screenshot with a correct coverage report on {{ftest}} project (https://issues.jboss.org/secure/attachment/12361047/ftest-coverage.jpg). If you want to add arquillian coverage on *{{jsf-ftest}}*, please don't forget to add the following dependencies on {{pom.xml}} {noformat} <dependency> <groupId>org.jboss.arquillian.extension</groupId> <artifactId>arquillian-jacoco</artifactId> <scope>test</scope> </dependency> <dependency> <groupId>org.jacoco</groupId> <artifactId>org.jacoco.core</artifactId> <scope>test</scope> </dependency> {noformat} -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira