4.107 - compiler generating wrong addresses?

loupan · Joined: 22 Oct 2007 Posts: 21

In a previous post, I discussed a case where it appeared to me that the compiler was generating incorrect addresses for some of the function calls:
http://www.ccsinfo.com/forum/viewtopic.php?t=41396&highlight=

CCS tells me I am the only one reporting this problem, but another user appears to have a similar problem:
http://www.ccsinfo.com/forum/viewtopic.php?t=41702&highlight=

The processor is a PIC18F8722, 128K, about 95% ROM usage.

I am using 4.017 and still see the same problem. Below is my analysis.
If someone sees something wrong with the below analysis, please let me know.

Compiler Version 4.107 – Summary of 4 calls to function get_hi_vis_options()

Function Definition for get_hi_vis_options. Function is located at address 04722

sjb · Joined: 13 Apr 2010 Posts: 34 Location: UK

I assume this is a large program. Have you tried cutting it down to a simple example?

If you can post a small, but complete example I'd be happy to see if it does the same for me , which I assume it will, but at least it will confirm there's nothing 'funny' about you setup.

I've just started to use CCS for a new project and have been finding a few bugs myself.

Steven

FvM · Joined: 27 Aug 2008 Posts: 2337 Location: Germany

It's obviously a compiler bug, assumíng the generated binary code is corresponding to the listing text. You may
want to find out, what are the exact conditions for generating (totally wrong) calls, just a distance above a certain
value, or arbitrarily, so the code may be correct if you move it by a small amount? In any case, it has to be fixed by CCS.

loupan · Joined: 22 Oct 2007 Posts: 21

Thanks for your responses. If I move the code around, or even change the order of the declartions, the problem will move to another function (or in a few cases, not occur).

CCS is not interested in pusuing this unless I send them the source code, which I am prohibited from doing. That's unfortunate.

Thanks again.

PCM programmer · Joined: 06 Sep 2003 Posts: 21708

Right, but you could fake up a program that uses generic names for
the functions, and maybe changes the internals of the functions to
be more generic code. You could probably duplicate the problem
and satisfy your non-disclosure agreement, and get CCS to work on
the problem, by doing it this way.

Douglas Kennedy · Joined: 07 Sep 2003 Posts: 755 Location: Florida

If you can't trust CCS with your code then you can't trust them for the compiler either. The best security is to write your own code and your own compiler and maybe even your own PC operating system. The point is you have to trust someone. If the source is the major concern then disguise it in someway as suggested. If 4.107 is the issue does an earlier version say 4.106 work?

loupan · Joined: 22 Oct 2007 Posts: 21

Mr. Kennedy,

I don't know how you determine who you trust, but based on my life experience, trust has to be earned.

For those of you that offerered non-sarcastic responses, thanks again for your suggestions.