399681 - Failures to upgrade legacy DB to shared DB go unreported in results.html

Status: RESOLVED FIXED

(NSS :: Test, defect, P1)

Description

[Julien Pierre]

Bob, Slavo,

While porting NSS to 64-bit windows and running all.sh, I experienced crashes in certutil when doing the database upgrade every single time ("legacy to shared library update" tests).
I went into the debugger to investigate. I ended up killing the debugger.
All these tests in results.html were green (passed).
This is a scripting problem that needs to be fixed. Right now, these tests are worse than useless since they are always reported as passing even when they fail in the worst way, by crashing.

Comment 1

[Nelson Bolyard (seldom reads bugmail)]

Again, IFF this is a Win64-only problem, it's not P1 for NSS 3.12, because
Win64 support is not yet a tier 1 platform for NSS 3.12

Comment 2

[Robert Relyea]

Julien, does the rest of the shared DB work?

Failure to update the shared DB could be because the databases in question were never created. The expectation was that shared DB issues would be caught when the rest of the tests were run.


bob

Comment 3

[Julien Pierre]

Nelson,

Re: comment 1, this particular failure of the db tests last friday were seen on win64 only. However, I have seen them fail too on officially supported platforms while doing other development, such as the util patch, and the tests crashed too and results.html was always green. This bug is about the fact that results.html is always green for the DB tests, regardless of their actual pass status. You could probably make softoken dereference a NULL pointer for the upgrade case, and still get those upgrade tests to be green in results.html .

Bob,

Re: comment 2, I haven't had time to investigate yet, my guess is that there are some problems right now. But I will deal with that in a separate bug. This bug is about the problem of unreported errors only, not about the errors themselves.

Comment 4

[Slavomir Katuscak]

I see the reason:

Bob uses constructions like:
certutil -G -g 512 -d sql:$i -f ${PWFILE} -z ${NOISE_FILE} 2>&1 | tee -a ${LOGFILE} 
html_msg $? 0 "Upgrading $i"

Then html_msg is not checking the return value of certutil, but actually the return value of tee. I'm going to fix this in within bug 398379.

Comment 5

[Julien Pierre]

Slavo,

I am very glad you found the reason for this. In general, you should fix separate bugs in separate patches, even if they affect the same file as other bugs being worked on.

Since in this case all the broken code is moving, we can make an exception here and leave the other bug as a dependency.

Comment 6

[Slavomir Katuscak]

Fixed within bug 398379.