Re: clearcase and inodes

[Eric Blake]

> I'm using diff 2.7.1 (but newer ones probably work the same way).
> 
> On clearcase trees, the fact that two directory have the same inode
> number doesn't mean the
> children under them are the same (this I found by inspection).

ClearCase violates POSIX, which requires that st_ino be unique
per file with the same st_dev.  As much as I think the concept of a
version control system that behaves as a mountable file system is
cool, I think the ClearCase implementation of this concept is
flawed.  Another flaw in their implementation is that they do not
increment st_nlink; when you use VOB hard links and can see the
same element via more than one path name, st_nlink still wrongly
reports 1.

As a result of these violations of basic POSIX requirements, programs
that make assumptions that should otherwise be portable, such
as the example you posted with diff, don't stand a chance.  I
don't know if raising a bug report with IBM would get them to
fix ClearCase, but it is worth a shot.

> 
> A brute force approach would be to have an option to turn off the inode
> compares.

It would be more than just diff that would need this, and I'm not
sure how likely you are to be successful at bloating GNU software
due to bugs in a proprietary version control system.

> (After using clearcase for nearly a year, I'm NOT a fan of it).

I also have to use it at my work, and agree with your sentiments.  My
other gripe with ClearCase is that because it is proprietary, it costs
money to buy licenses, and I have been sometimes been locked out from
doing useful work because my company did not buy enough licenses.
Open source version control systems, such as CVS, svn, arch, (or fill in your
favorite here), do not have this problem - the version control system is
always available to any number of clients, for no additional money.

-- 
Eric Blake