Re: [Wrapper-user] Mac OS X - Segmentation Fault

[Leif M. <le...@ta...>]

Mike,
Thanks lots of great info here. I had not heard of ElectricFence before.
Once
I got it setup I was able to locate the problem in about 10 minutes. :-D
The 0 length malloc was being caused when there are no output filters
configured.
I now check for that so the Wrapper now passes ElectricFence without any
warnings.

Andy, could you please try building the MacOSX version again with the latest
code from CVS?

>But that turned up another issue.  Efence spits out some text on
>initialization, which means the first thing wrapper sees is '\n'.
>Wrapper was not amused.  I had to do the following patch:
>
I implemented this fix a little bit differently so there is not any
duplicate code.
But that was a definite bug. I was able to reproduce it by having a Java
program output a single line feed before any other output. That has been in
the code for quite a while. But still I am glad it is fixed.

>First:  in _C_, one should _NEVER_ cast the return value of malloc().  If
>you get warnings without it, your ether your code or compiler is broken.
>
Ok, I went through and removed all of the casts. I am not sure exactly where
I got in the habit of casting those. In my very early C days I remember
having
had some problems on a Solaris machine. I knew that it was not required with
the compilers I was using, but wanted to make the code as portable as
possible
so I kept using them.

>Second, if strings[i] contains an %, bad things can happen in the sprintf.
>
I had gone through a while back and modified the code to always use
sprintf(buf, "%s", str) rather than sprintf(buf, str) in places that could
contain % characters. strcpy is clearer so I made the change as you
suggested.

Thanks for the help with this.
Leif