Thursday, June 25, 2009

Up to my eyeballs in tests

As a Change Request Team advocate, I am stringent about asking for test results and always very annoyed when an implementor complains about how complicated the tests are to run.

Now after having spent the last several days finding working test hardware from our pool of test machines, and fighting with test installations and executions... I'm still waiting for my baseline results. I haven't even run the full tests on my own bits yet.

Which is another story.... while my builds were successful and my changes to libelfsign seemed to be kosher, I found that after doing a bfu that my test machines wouldn't even boot.  No, I didn't change libc... so I was very surprised that such behaviour was seen. Yes, I knew things like kerberos and IPsec would not work correctly if libelfsign (a core component of the Cryptographic Framework) wasn't working - but inability to boot? I was shocked.  With some help from pwernau and meem, I finally got one of the systems up in single user mode to discover the linker was doing something... unusual.

Fortunately, a very responsive Rod Evans came and looked at my limping test system and figured out what the linker was doing wrong (and also something one of the libraries in my calling path was doing wrong), and now I've got systems I can play with.

Except when I forget to sync my x86 build workspace with my sparc workspace and I build archives without Rod's fix... and then wedge another test machine.

Hopefully the code will be up for review soon, when I will add another blog entry detailing what it is exactly I'm trying to do and why.