[klee-dev] Packaging klee on Nix - gtest broken?

Morgan me at numin.it
Sat Jan 8 22:19:24 GMT 2022 

>Is this also the issue you ran into? If yes, maybe you want try the
>patches from the PR I linked above. If not and are you having a
>different problem, maybe you could try to provide some more details?
>Then I will try and see if can help resolve them.

This is the problem. No tests discovered. That patch fixed it, thanks a ton!
Hydra will be running the full KLEE system and unit test suite from now on.


Morgan

On Wed, Jan 5, 2022 at 2:48 AM Cristian Cadar <c.cadar at imperial.ac.uk> wrote:
>
> Hi all,
>
> Indeed, it would be great to update
> https://klee.github.io/getting-started/ (via a PR at
> https://github.com/klee/klee.github.io) to mention the Fedora and Nix
> packages.  And thanks to everyone who is maintaining KLEE packages!
>
> Best,
> Cristian
>
> On 05/01/2022 10:01, Julian Büning wrote:
> > Hi Lukas,
> >
> > nice and thanks for letting me know!
> >
> > I was briefly considering to go the same route, but didn't encounter
> > your fix. But as it turns out, not using gtest_main (which I understand
> > is more or less offered for convenience) has certain other advantages
> > for KLEE (e.g. stack traces; reducing the number of combinations between
> > vanilla Google Test, LLVM's Google Test, llvm-lit, and their respective
> > versions). Still, it's certainly a nice addition for llvm-lit, hopefully
> > somebody with commit access will pick it up soon!
> >
> > Thanks to your email I also found out that there is actually a Fedora
> > package for KLEE in the main repository. Awesome! I'm not sure how I
> > missed that. You should definitely get it mentioned on klee.github.io!
> >
> > Best,
> > Julian
> >
> > On 1/5/22 10:15, Lukas Zaoral wrote:
> >> Hi Julian,
> >> I've encountered the same problem with lit and latest gtest when
> >> I was packaging KLEE for Fedora as I had to use gtest from repos
> >> due to Fedora's packaging guidelines.
> >>
> >> I sent a patch to LLVM to fix this incompatibility at the beginning
> >> of last April and it was finally accepted last month [1].  It still needs
> >> to be committed, though.
> >>
> >> Sincerely,
> >> Lukas
> >>
> >> [1] https://reviews.llvm.org/D100043
> >>
> >> On Wed, Jan 5, 2022 at 9:44 AM Julian Büning
> >> <julian.buening at rwth-aachen.de> wrote:
> >>>
> >>> Hi Morgan,
> >>>
> >>> nice to see your packaging efforts for KLEE!
> >>>
> >>> I recently ran into some issues with more recent versions of Google Test
> >>> when building KLEE (and running unit tests). I just opened a PR that
> >>> addresses these: https://github.com/klee/klee/pull/1458
> >>>
> >>> Among these issues is one that I image you may also have run into (as I
> >>> assume your package will not be built against Google Test 1.7.0), but it
> >>> differs quite a bit from the issue that you linked. Thus, I will go
> >>> ahead and describe what I experienced (hoping you can tell me if that
> >>> matches what you saw).
> >>>
> >>> When building KLEE with Google Test 1.7.0 and running the unit tests, I
> >>> get 36 successfully passed tests. When instead using a newer Google Test
> >>> version, like 1.11.0, I get the same number of passed tests, but the
> >>> following 10 unresolved tests in addition:
> >>>
> >>> Unresolved Tests (10):
> >>>     KLEE Unit tests :: ./AssignmentTest/Running main() from
> >>> /some/absolute/path/to/gtest_main.cc
> >>>     KLEE Unit tests :: ./DiscretePDFTest/Running main() from
> >>> /some/absolute/path/to/gtest_main.cc
> >>>     KLEE Unit tests :: ./ExprTest/Running main() from
> >>> /some/absolute/path/to/gtest_main.cc
> >>>     KLEE Unit tests :: ./RNGTest/Running main() from
> >>> /some/absolute/path/to/gtest_main.cc
> >>>     KLEE Unit tests :: ./RefTest/Running main() from
> >>> /some/absolute/path/to/gtest_main.cc
> >>>     KLEE Unit tests :: ./SearcherTest/Running main() from
> >>> /some/absolute/path/to/gtest_main.cc
> >>>     KLEE Unit tests :: ./SolverTest/Running main() from
> >>> /some/absolute/path/to/gtest_main.cc
> >>>     KLEE Unit tests :: ./TimeTest/Running main() from
> >>> /some/absolute/path/to/gtest_main.cc
> >>>     KLEE Unit tests :: ./TreeStreamTest/Running main() from
> >>> /some/absolute/path/to/gtest_main.cc
> >>>     KLEE Unit tests :: ./Z3SolverTest/Running main() from
> >>> /some/absolute/path/to/gtest_main.cc
> >>>
> >>> For each of these "tests" I see some earlier output like this:
> >>>
> >>> UNRESOLVED: KLEE Unit tests :: ./AssignmentTest/Running main() from
> >>> /some/absolute/path/to/gtest_main.cc (1 of 46)
> >>> ******************** TEST 'KLEE Unit tests :: ./AssignmentTest/Running
> >>> main() from /some/absolute/path/to/gtest_main.cc' FAILED
> >>> ********************
> >>> Unable to find '[  PASSED  ] 1 test.' in gtest output:
> >>>
> >>> Running main() from /some/absolute/path/to/gtest_main.cc
> >>> Note: Google Test filter = Running main() from
> >>> /some/absolute/path/to/gtest_main.cc
> >>> [==========] Running 0 tests from 0 test cases.
> >>> [==========] 0 tests from 0 test cases ran. (0 ms total)
> >>> [  PASSED  ] 0 tests.
> >>>
> >>> ********************
> >>>
> >>> The last 3 lines look similar to the output in the issue you linked. But
> >>> this is simply the output of Google Test when there are no `TEST()`s
> >>> next to `main()` in an executable. The rest stems from a different
> >>> problem (detailed below).
> >>>
> >>> Is this also the issue you ran into? If yes, maybe you want try the
> >>> patches from the PR I linked above. If not and are you having a
> >>> different problem, maybe you could try to provide some more details?
> >>> Then I will try and see if can help resolve them.
> >>>
> >>> --- BEGIN: More details ---
> >>>
> >>> The issue we see above actually stems from llvm-lit, not from Google
> >>> Test itself. Starting from 1.8.1, Google Test's gtest_main.cc uses
> >>> `__FILE__` [1] instead of a fixed string [2] to output a line like this:
> >>>   > Running main() from /some/absolute/path/to/gtest_main.cc
> >>>
> >>> To determine which tests exist, llvm-lit will call each executable with
> >>> the `--gtest_list_tests` argument. However, the (usually) first line
> >>> will be the above "Running main()" output. To skip this, each line is
> >>> compared to "Running main() from gtest_main.cc" [3], which is a fixed
> >>> string assuming the behavior of 1.8.0 and before.
> >>>
> >>> Hence, the line with path will be recorded as a test, and result in a
> >>> corresponding call to the test executable with `--gtest_filter` set
> >>> accordingly. As there is no test that matches the given pattern, we see
> >>> the output shown above. As it does not include the expected "[  PASSED
> >>> ] 1 test." line, it is counted as unresolved.
> >>>
> >>> [1]
> >>> https://github.com/google/googletest/blob/release-1.8.1/googletest/src/gtest_main.cc
> >>>
> >>>
> >>> [2]
> >>> https://github.com/google/googletest/blob/release-1.8.0/googletest/src/gtest_main.cc
> >>>
> >>>
> >>> [3]
> >>> https://github.com/llvm/llvm-project/blob/llvmorg-13.0.0/llvm/utils/lit/lit/formats/googletest.py#L60-L64
> >>>
> >>>
> >>> --- END:   More details ---
> >>>
> >>> Looking forward to your answer!
> >>>
> >>> Best,
> >>> Julian
> >>>
> >>>
> >>>
> >>> On 1/1/22 01:28, Morgan wrote:
> >>>> Hey there,
> >>>>
> >>>> I like Klee and have been trying to package it in nixpkgs so more
> >>>> people can reproducibly use it without resorting to things like setup
> >>>> scripts or Docker. Here are the cmake flags I'm using:
> >>>>
> >>>> https://github.com/NixOS/nixpkgs/pull/153014/files#diff-cb8d40a4e82c0c50ce6ec4031c12e06a4dac4bded86b9f01afcb2b4f22532dbbR46
> >>>>
> >>>>
> >>>> Everything works including the system tests, which is a very good
> >>>> sign. However, I'm having trouble with the unit tests that resembles
> >>>> this problem:
> >>>>
> >>>> https://github.com/google/googletest/issues/2157
> >>>>
> >>>> Has anyone else run into this?
> >>>>
> >>>> Thanks!
> >>>> Morgan
> >>>>
> >>>> _______________________________________________
> >>>> klee-dev mailing list
> >>>> klee-dev at imperial.ac.uk
> >>>> https://mailman.ic.ac.uk/mailman/listinfo/klee-dev
> >>>>
> >>>
> >>>
> >>> _______________________________________________
> >>> klee-dev mailing list
> >>> klee-dev at imperial.ac.uk
> >>> https://mailman.ic.ac.uk/mailman/listinfo/klee-dev
> >>>
> >>
> >
> >
> > _______________________________________________
> > klee-dev mailing list
> > klee-dev at imperial.ac.uk
> > https://mailman.ic.ac.uk/mailman/listinfo/klee-dev
>
> _______________________________________________
> klee-dev mailing list
> klee-dev at imperial.ac.uk
> https://mailman.ic.ac.uk/mailman/listinfo/klee-dev

More information about the klee-dev mailing list