UNIX for Dummies Questions & Answers

Every time we build an executable the date and time are put into the file, I need to run checksum on just the working lines.(IE, no header files)

Is this even possible, if so how would I go about it?

I am using a HP-UX server any help you can give me will be greatly appreciated.

Thanks

I don't think that crazykelso expressed the problem as clearly as possible. Since I also work on HP-UX, I thought that I would explain the problem more fully.

If you have a c program called hello.c and you run:
cc hello.c -o hello1
cc hello.c -o hello2

those two files hello1 and hello2 will be different. If you run cksum or other another checksum program you will get two different results. The difference arises from timestamps that HP-UX puts into the executable. It would be nice to extract the non-timestamp stuff out of an executable and just cksum it.

I don't know of a solution.

You scare me when you say that!

Hmmmm...

Is the time-date stamp always the same offset from the beginning of the file and a fixed length? If so, you could copy the entire executable minus the timestamp (leave blank or fill with some character) into a temp file, run CRC32 (MD5, whatever) against it and then do the same with the second file and then compare CRCs to determine if the files are different or not.

I would think that you would have to generate the temp file by fseek'ing byte after byte into the temp binary file leaving out the timestamp. Of course, the temp file probably couldn't be executed but it oughta be good for comparison sake since both files would have had the common timestamp effectively removed.

I tried your example under FreeBSD and md5 reported that both executables were the same so I didn't try any further.

There is one timestamp per compiled module. And the last time that I looked at this, HP didn't document enough of the a.out format to do this. I just learned to live without checksumming executables.

And yes, HP-UX is the only version of unix that I know of with this problem/feature.

If I were to peruse this further, I would compile hello1 and hello2 and take a gander at where the files were different and eliminate those portions from a compare with the method from the previous post.

The timestamp may not be documented, but crazykelso ought to be able to find the offset and length of the timestamp. But that's with some simple assumptions, like the offset never changes, etc. It would be a neat excercise and if it worked.

I appreciate your help.

I did a file comparison using a compare tool called: TSBinComp
which is freeware I downloaded from

http://www.screaminet.com/~tswirsky/bincomp.htm

this tool told me that the only differences in the file occur in bytes 223 and 224. If this isn't a reliable tool could someone recomend another one for me?

My next question is, How would I go about removing these two bytes so that I could cksum the two files.

I also did a bdiff and founs that the problem occurs on line fourteen, if it isn't possible to remove those two bytes how would I be able to remove this line or a certain part of this line?

thanks again for your help
crazykelso

I would write a routine in C (or C++) that would open the source files and then fseek offsets from 0 to (n-1) (where n is the length of the file) and just skip over or substitute a common character (say, captial X) for bytes 223 and 224 and write it out to a temporary binary file and run checksum against it and then perform the same with the second file and then compare checksums.

You could probably also write a similiar routine in perl as well but C seems to appropriate.