[petsc-dev] Using Clang static analyzer in nightly/daily tests (master and next)

[Barry Smith]

   Satish has kindly set up having the Clang static analyzer run on all PETSc code during our testing. You can find links to the output of the analyzer on the seventh row of the dashboard (accessible from ftp://ftp.mcs.anl.gov/pub/petsc/nightlylogs/index.html) as 

linux-analyzer_crush [B] [E]

The B points to the problems detected in the source code and the E to problems in the examples.

   Though some of you believe that static analyzers produce tons of false positives I have already looked through and fixed a multitude of problems (some that had been in the code for at least a decade) and I have detected very few false positives.

   Please look through the output and try to fix ones you see in portions of the code you frequent. By clicking on the 

View Report

buttons on the right side of the screen for each issue you can see exactly why the static analyzer thinks there is a problem.

Notes:  Please do not blindly initialize a variable to zero if the static analyzer states it is used uninitialized; this may actually conceal a bug where the value should have been set in the logic and was not. So make absolutely certain that it is a false warning before you do such an initialization.

"PETSc style" asserts can be used in the code to eliminate paths that you know will not exist in the code and remove clang analyzer warnings. For example you could add a

if (x < 0.0) SETERQ(...) 

if you know that x cannot be negative and the analyze gives you a warning when it is negative.

Since many issues are found at the moment by the analyzer the Dashboard does not automatically send email to let you know about them. The Dashboard will continue to send email about all issues found by the compilers (including seemingly a single false positive that goes to Toby). Once we have eliminated most analyzer warnings we may turn on the automatic email for them so people are quickly warned about new issues they have introduced.

Final note: One of the best reasons to make PETSc code issue free with the clang static analyzer is that it will give us the right to criticize all other software projects that do not :-)


  Barry