[cmake-developers] Help with diagnosing dashboard failure

[Stephen Kelly]

Brad King wrote:

> On 09/22/2015 05:03 PM, Stephen Kelly wrote:
>> A few days ago I merged a commit which moves the construction of
>> cmLocalGenerator objects.
>> 
>>  http://cmake.org/gitweb?p=cmake.git;a=commitdiff;h=a3aa333d
>> 
>> It fails on the RunCMake.try_compile test on several dashboards in the
>> CMP0056 test at the end where the policy is set to NEW.
> [snip]
>> I am unable to reproduce the problem.
> 
> I'm able to reproduce it on 64-bit Linux with the Unix Makefiles
> generator.  However, I cannot do it with just a3aa333d.

Interesting, thanks for trying it. I had a report that it failed on the 
first commit of the fix-max-path-initialization branch which is currently in 
stage.

> The
> test failures reported on the dashboard a few days ago were
> actually using this version:
> 
>  Merge topic 'fix-max-path-initialization' into next
>  http://cmake.org/gitweb?p=cmake.git;a=commitdiff;h=2cdc6d09
> 
> That topic includes 26cae90b, a revision of a3aa333d, 

Yes, however the revision was not correct. The current fix-max-path-
initialization branch does not use that revision.

> followed
> by several more changes.  The merge base of that topic does not
> fail but the head of the topic does.  Within the topic the problem
> bisects to:
> 
>  cmGlobalGenerator: Create local generator after configuring the makefile.
>  http://cmake.org/gitweb?p=cmake.git;a=commitdiff;h=d3f6ff43
> 
> Can you reproduce it with that version?

That commit fails because it is based on 26cae90b. I would expect every test 
to fail with that commit, not just the RunCMake.try_compile test.

If I checkout fix-max-path-initialization and run

 git rebase -i --exec 'cd /home/stephen/dev/src/cmake-master-build2 && ninja 
install && ctest -R ke.try_comp' origin/master

then every commit passes RunCMake.try_compile for me.

Thanks,

Steve.