[question] Expected SVF results / Analysis type/settings to use?

[dtzWill]

Hi!

Thanks for all your work on SVF and making PTABen available!

While testing the ongoing SVF -> LLVM 4.0 changes I found these tests very helpful.

Currently it looks like fi_tests are the default target, and using latest SVF (LLVM 3.8) it looks like all those tests pass.

Unfortunately it seems the other test folders do not currently pass (when using the default options, at least).

Is this expected behavior, or is something wrong on my end?

For things like flow-sensitivity and context-sensitivity I can imagine different analyses are appropriate, should these test sets be used with different options to wpa? (-fspta, maybe)?

In short: how do /you/ guys use PTABen for testing SVF? Is it expected that some tests fail (things that have yet to be implemented, etc.)?

[yuleisui]

Hi Will,

For testing flow-sensitive test cases in folder fs_tests, it needs to change the option -ander to be -fspta. When you run the tests cases in fs_tests, all the results analyzed by flow-sensitive analysis should pass (marked in green), and some results of flow-insensitive (Andersen's) analysis will fail (marked in red) for comparison purpose.

Everything should work (results of the targeted analyses should all be in green) before pushing the code to repo.

[dtzWill]

Great, thanks for the quick response!

Am I understanding this correctly?

"Does SVF work":

• fi_tests should all be green with default -ander
• fs_tests should all be green with -fspta (and some fi_tests might fail)

Should the tests in cs_tests and path_tests be used as well, or are those mostly for comparison/evaluation purposes?

[yuleisui]

Yes, correct!

cs_tests and path_tests are for SUPA. One can also use them for comparison purpose.