Programming

Hi,
I'm looking for a quite "interesting" bug at the moment - any idea could help! (o;
Language is fortran 90, compiler gfortran (sry, dont find which version it is). As I suppose you dont want to read like 10k lines of code here, I won't post the entire thing ^^
The problem is that I get segmentation faults when deallocating four of my arrays while running the program. The output is depending on which of the arrays i deallocate first.. I always get one of those:

*** glibc detected *** double free or corruption (out)
*** glibc detected *** free(): invalid next size (normal)
*** glibc detected *** double free or corruption (!prev)
segmentation fault

This happens only when a variable - which has NOTHING to do with the arrays i'm trying to deallocate - has a certain value.
My problem is that the cause isn't one of the normal stuff - the arrays are totally well defined, i'm not trying to write values outside the boundaries, ect.
Also, I can deallocate the arrays directly after changing them the last time.

I do suppose that at some other point of the program something is going terribly wrong and sort of overwriting my arrays (But nooo, the values as well as size(), ubound() and lbound() are what they should be, and allocated() is true.)

Does anybody have an idea what is happening here? Have you seen something similar yet? What could i still try?

Greetings & thank you,
Giogio

Hi, Giogio.

Welcome to the forum.

1) Did your program ever work correctly? If so, what was the most recent change you made? If not, have you compiled and run with -fbounds-check ?

2) Are you using any c / c++ code?

The short code at Double free or corruption error produces the "glibc ... out' message.

cheers, drl

Hi drl,
Thank you very much for your reply (-:
The -fbounds-check totally did the trick.
The problem was a line of the form

Code:

array1(x,:)=Loong formula of array2(:,y)

where the dimensions of the arrays didnt match - please dont ask me why that one worked all the other times ^^.
Thanks and bye,
Giogio

Hi, Giogio.

Glad to hear it pointed you in the right direction.

There is always a balance between desired speed and risk. I think Niklaus Wirth recommended bounds checking on arrays particularly for a production code. The idea is that one's results are worthless, even dangerous, if they are produced quickly but incorrectly.

At a place I once worked specializing in big iron: Big iron - Wikipedia, the free encyclopedia , we often said that if you don't care about the reliability of the answers, we can make the computation arbitrarily fast

In any event, -fbounds-check is a useful debugging feature.

Best wishes ... cheers, drl