[dpdk-dev] Coverity scan

[Thomas Monjalon]

11/03/2020 18:34, Aaron Conole:
> Thomas Monjalon <thomas at monjalon.net> writes:
> 
> > We have a public Coverity scan triggered by John for the community:
> > 	https://scan.coverity.com/projects/dpdk-data-plane-development-kit
> > Note there is a tool to help with this task:
> > 	http://thyrsus.com/gitweb/?p=coverity-submit.git;a=shortlog;h=refs/tags/1.13
> >
> > I see two issues with this scan:
> > 	- it is run manually
> > 	- not all code is scanned currently
> >
> > Note that we should be able to run one scan per day for free:
> > 	https://scan.coverity.com/faq#frequency
> >
> > With David, we looked at automating the Coverity scan,
> > with the help of Travis automation:
> > 	https://scan.coverity.com/travis_ci
> > Such automation cannot be configured on the existing Coverity project.
> 
> Why not?

Because Coverity does not allow it.
Travis integration is possible only if the project was created with GitHub credentials.


> > I tried to open a new Coverity project connected to our GitHub.
> 
> I don't know that it will work.  Either you'll need a separate GitHub,
> or you'll need to use a special branch.
> 
> > I have a very poor confidence in Coverity/Travis/GitHub integration.
> > I will explain below why.
> 
> Hrrm.. lots of projects use it.  And they do just what you prescribe
> below (skipping jobs/builds when on the coverity branch).

Which project is using Travis integration of Coverity?
How do they automatically update the specific branch without conflict?


> > 1/ The instructions were wrong. In this command, there are two mistakes:
> > 	openssl s_client -connect https://scan.coverity.com:443 |
> > 	sed -ne '/-BEGIN CERTIFICATE-/,/-END CERTIFICATE-/p' |
> > 	sudo tee -a /etc/ssl/certs/ca-
> > For the record, a proper a simpler command is:
> > 	true | openssl s_client -connect scan.coverity.com:443 |
> > 	openssl x509 |
> > 	sudo tee -a /etc/ssl/certs/ca-certificates.crt
> 
> Okay, that's fixable.
> 
> > 2/ The coverity scan is triggered as a job addon.
> > The rest of the job must be cancelled with this tricky patch:
> >
> > -script: ./.ci/${TRAVIS_OS_NAME}-build.sh
> > +script: if [ "${COVERITY_SCAN_BRANCH}" != 1 ] ; then ./.ci/${TRAVIS_OS_NAME}-build.sh ; fi
> 
> More than that, because we probably also want:
> 
> if ([[ "${TRAVIS_JOB_NUMBER##*.}" == "1" ]] && [[ "${TRAVIS_BRANCH}" == "coverity_scan" ]]); then ./.ci/${TRAVIS_OS_NAME}-build.sh ; fi
> 
> That will only do one job (which solves 3/ below)

OK good

> > 3/ We need only to prepare the source code once per day.
> > But our .travis.yml has many jobs which must be dropped or ignored.
> >
> > 4/ A big encrypted token must be added in the configuration:
> > 	# encrypted COVERITY_SCAN_TOKEN
> > 	- secure: "VgRYG9N5adKkM9/QpPgswn1c+VXS1mFVN0vgdjuC/bDv2x4u...etc..."
> 
> Why it's a problem?

It's not a problem. I explained all steps in this email, that's why it's here.


> > 5/ The addon is triggered when pushing to a specific branch
> > (adding config for the record):
> > 	coverity_scan:
> > 		project:
> > 			name: "DPDK/dpdk"
> > 		notification_email: test-report at dpdk.org
> > 		build_command_prepend: "meson build -Dexamples=all"
> > 		build_command: "ninja -C build"
> > 		branch_pattern: coverity_scan
> >
> > 6/ This attempt failed with this log (no more information):
> > 	$ export PROJECT_NAME=DPDK/dpdk
> > 	Coverity Scan analysis selected for branch coverity_scan.
> > 	Coverity Scan API access denied. Check $PROJECT_NAME and $COVERITY_SCAN_TOKEN.
> 
> Probably there is an issue with the token + PROJECT_NAME.

Probably. How can I debug it?


> > So I am giving up with Travis+Coverity.
> > The only benefit of Travis is to have a central build configuration.
> > So when a driver is enabled in Travis, it would be scanned in Coverity.
> > Note: Coverity does a build step to prepare the sources.
> 
> I can try to assist with this if you've not completely abandoned the idea.

OK, feel free to ping me for troubleshooting.


> > Now the question: how can we better configure the community Coverity scan?
> > I propose to set it up in our community lab.
> > Comments? Suggestions?
> 
> Since we do have something working, but it's manual, is there a way to
> at least make it happen automatically?  Maybe some cron job?

Yes a cron job, but where? I proposed a server of the community lab.