At 12:18 PM 10/8/2004, Stefan Slapeta wrote:

Hi all,

there are two failures reported for the intel-win32-8.1 toolset (both in operations_test): http://tinyurl.com/5dbla

After some investigation I've found out now that the tests fail only because they are run on a FAT32 disk! It was just coincidence that the failures popped up in the Intel 8.1 tests (...and it's quite amusing to say that we would need another toolset dimension for the filesystem :) )

These same two failures have also shown up on some other platforms from time-to-time. There is apparently a new debug mode assert in VC++ 8.0 which is firing due to the second error.

Anyway, here some analysis (the second failure seems to be a general design problem!):

The second error is actually a really stupid typo in my original code that just happens to result in valid C++ that compiles (but does the wrong thing). I must have looked at that code five times in the last year, but didn't spot the typo until I read your message. See below.

First failure:

operations_test.cpp(287): test time_diff >= 0.0 && time_diff < 60.0 failed

--> the actual difference is -2!

I'll change the test to plus or minus 60 seconds.

Second failure:

std::exception: boost::filesystem::last_write_time ("wrong parameter")

last_write_time contains this code:

::utimbuf buf; buf.actime = std::time_t(); ^^^^^^^^^^^^^

That was meant read: buf.actime = std::time(0) (The same typo appears later in the code, too.) Now, it may be a poor design to update actime to the current time. It might be better to do a stat(), and get st_atime, and then use that. Opinions? Thanks, --Beman