Log math function

thomasj50 · Joined: 23 Aug 2006 Posts: 19

Hi, I have a problem with version 4.023 that doesn't exist in version 3.245. Basically the log function gives the wrong answer. In an otherwise unchanged program these are some results for the two versions:

3.245: log(50) gives 3.92, log(5) gives 1.509, both correct
4.023: log(50) gives 17.41, log(5) gives 0!

It took a loong time to find these as compiler bugs aren't the first I would supect. Would anybody please test this on their machines ? The processor is a PIC18LF258.

Also, has anyone found any problem with automatic casting from ints to floats ? I have an assignment like x = i, where x is a float and i is an unsigned int 16, and it does not work without a cast (float)i. I will try to make a small example demonstrating what I mean.

Best Regards, Thomas Johansson

PCM programmer · Joined: 06 Sep 2003 Posts: 21708

I think the problem is that when CCS did the change described in this
thread, that they shot themselves in the foot:
http://www.ccsinfo.com/forum/viewtopic.php?t=29735

Load the math.h file and scroll down to the log function. It starts with this:

thomasj50 · Joined: 23 Aug 2006 Posts: 19

Hi, yes it works after the proposed changes, thanks.

Haven't CCS heard of regression testing - if you change ANYTHING you have to test EVERYTHING ? In this particular case it is even simpler - have they tested the log function at all ? Perhaps the compiler is too cheap :-)

Thanks for the help,
Thomas Johansson

PCM programmer · Joined: 06 Sep 2003 Posts: 21708

The lack of regression testing has been a complaint on here before.
How to make them do it ? I don't know.

thomasj50 · Joined: 23 Aug 2006 Posts: 19

I received a new math.h file from CCS that should correct the problem. Only one thing: they added two casts to int8 but it should have been int8*. Didn't they test their own corrections ? I had the two changes, one from this forum and one from CCS, choose CCS and got stuck with the same problem again. It took me several days to consider the change from CCS as the cause of the problem ...
thomasj