realpath() usage incompatible with POSIX 2004 (IEEE Std 1003.1-2004) or earlier

[pudquick]

I like your tool and am working on seeing what it would take to port it to BSD (I primarily use Mac OS X).

I'll include a minor fix in here for that: <unistd.h> needs to be included in helpers.h to get 'off_t' to resolve.

But the current hurdle I'm working on is the difference with Linux vs. BSD implementation of realpath.

BSD implementation doesn't accept NULL for the second argument and this will trigger a Bus Error / Segmentation Fault as a result.

When I come up with a patch fix I'll post it, but thought I'd mention it - should you want to try to make this wonderful tool more portable.

[pudquick]

Update:

Looks like the issue has more to do with older versions of OS X. I just tried compiling it on 10.6 and it seems to have worked flawlessly.

The original box in question that I was attempting it on was a 10.4 machine, still looking at the compatibility differences.

Edit:

It looks like 10.6 introduced the following difference:

"If the resolved_name argument is non-NULL, the resulting absolute pathname is copied there (it must refer to a buffer capable of storing at least PATH_MAX characters)."

In short, they added support for NULL as the second argument. This makes me happy, since it would appear that if I can patch this, it'll work nicely on 10.4.

[pudquick]

Found the difference. Came down to a change in the POSIX (IEEE Std 1003.1) specification, specifically IEEE Std 1003.1-2008 version. With the 2008 POSIX specification, they added support for the now standard NULL for the second argument. So the only systems that this would be necessary for would be POSIX 2004 (IEEE Std 1003.1-2004) or earlier.

Made my commit to my own fork (though I don't have conditionals in place to compile with/without this compatibility, so I won't do a pull request or anything quite yet).

Details here:
pudquick@fabd172