[lmi] Running unit tests in less time

[Greg Chicares]

See commit a905d1969eac34f599c7b6ef4f3856a606d33be1 .

Timings with all test programs already built...before that commit:

/opt/lmi/src/lmi[0]$time make $coefficiency unit_tests >../log-old 2>&1
make $coefficiency unit_tests > ../log-old 2>&1  25.23s user 3.32s system 85% 
cpu 33.445 total

...and after that commit:

/opt/lmi/src/lmi[0]$time make $coefficiency unit_tests >../log 2>&1
make $coefficiency unit_tests > ../log 2>&1  25.45s user 4.68s system 304% cpu 
9.906 total

33.445 / 9.906 = 3.4 times as fast

As always with parallel tests (given excess CPUs), the total time
is the time it takes to run the slowest test:

/opt/lmi/src/lmi[0]$time make $coefficiency unit_tests 
unit_test_targets=calendar_date_test.exe >../log 2>&1
make $coefficiency unit_tests unit_test_targets=calendar_date_test.exe >  2>&  
9.52s user 0.14s system 98% cpu 9.812 total

[How strange that 'time' scrambled the redirection
- >../log 2>&1
+  >  2>&
when echoing it back.]

In one trial, I saw:

Running premium_tax_test:

** uncaught exception: std::runtime_error: Unable to parse xml file 
'Z:/opt/lmi/src/build/lmi/Linux/gcc/ship/sample.policy': Document is empty
[file /opt/lmi/src/lmi/xml_lmi.cpp, line 69]

which would seem to be the problem discussed here:

http://lists.nongnu.org/archive/html/lmi/2014-10/msg00018.html
http://lists.nongnu.org/archive/html/lmi/2014-10/msg00019.html

I hadn't seen it again until today. I speculate that it might occur
only the first time the unit tests are run after 'make clean', which
would mean that it's not a high enough priority to fix right now,
unless it keeps arising.