[TARGET] each option of target str should only contain one '='

[windclarion]

src/target/target_id.cc ParseAttrsFromRawString L222:
if ((pos = FindUniqueSubstr(s, "=")) != -1)
require option contains only one '='

The CHECK code in FindUniqueSubstr will cause build fail.

Signed-off-by: windclarion windclarion@gmail.com

[junrushao]

Ah I see, I am a bit surprised the typo is not found before. BTW, please rebase to master, and change “-target” to “-mtriple”. Thanks!

[windclarion]

done! old code works ok before "[Target] Migrate data structure of TargetNode (#5960)", the commit add a requirement in src/target/target_id.cc. FindUniqueSubstr require options string has only one '=', we didn't force the limit before that commit.

[junrushao]

Thanks!

[junrushao]

@windclarion Looks like a flaky test. Could you file an empty commit using git commit --allow-empty -m "retrigger" to retrigger the CI?

[ZihengJiang]

Could you also add unit tests generating code for those target, which will also re-trigger the CI?

[windclarion]

@ZihengJiang I try to add test, but if I feed the option with a string which contains two '=', target_id.cc FindUniqueSubstr CHECK will abort, so I don't know how can I test the abort operation, it's not a exception which I can catch, it's just abort. Can you give me some advice or tell me similar test file which can test the CHECK?

[junrushao]

@windclarion use pytest.raises: https://stackoverflow.com/questions/23337471/how-to-properly-assert-that-an-exception-gets-raised-in-pytest

[windclarion]

@junrushao1994 What I have said maybe not very clear, I mean I can't catch a exception in python code, because C++ CHECK just abort, the abort function didn't throw a exception.

[windclarion]

@junrushao1994 I add a small test which just test target whether can be created, and I just make sure the target shouldn't be None.

[windclarion]

today's test is flaky, I always failed at the unexpected position, and I don't know how can I cause the fail

[junrushao]

Yeah seems like another flaky test...Could you re-trigger? Thanks!

[junrushao]

I see. There is a flaky test in the recent week: #6009. Don't worry, let's wait until it is fixed.

[ZihengJiang]

@windclarion Could you try to build a small program likeA+B with every target for testing the code generation? ( you can jump some target like cuda if it requires extra efforts)

[junrushao]

The flaky test is resolved in #6014 :-)

[windclarion]

@ZihengJiang many other tests has tested tvm.build, ex. tests\python\integration\test_reduce.py test_reduce_prims Line64 has checked metal, vulkan, cuda, opencl, rocm