[Boost-users] [test] Possible race condition? (minor)

From: Paul Giaccone (paulg_at_[hidden])
Date: 2006-04-05 07:33:48

Boost unit test framework seems to have a race condition in its output.

When I run my test program from a shell, I get this output:

paulg_at_klingon 120% ./CubicTest
Running 8 test cases...
 
*** No errors detected
paulg_at_klingon 121% ./CubicTest>/dev/null
 
*** No errors detected

Note that the second call above, redirecting output to /dev/null, 
suggests to me that "Running 8 test cases..." goes to standard out and 
"*** No errors detected" goes to standard error.

Because standard error is unbuffered, this can mean that the messages 
are output in the reverse order. This is the case when I run my program 
from Eclipse (under Linux), and, in general, will happen if the tests 
run faster than the time it takes for the buffer to flush to the output.

This suggests one of the following is required:

* Flush standard out before running the tests
* Send all output to the same stream (but this is not recommended, as 
this would break the defined behaviour of test suites)
* Synchronise the two streams by some other means

It's possible that Eclipse displays the output to standard error before 
the output from standard out, but such behaviour isn't seen when 
executing a makefile, so it seems likely there there is a race condition 
in Boost unit test framework.

Paul