Symbolic links in llvm/clang tests

[Rafael Espíndola via llvm-commits]

On 29 June 2016 at 10:42, Björn Pettersson A
<bjorn.a.pettersson at ericsson.com> wrote:
> Hello Rafael!
>
>
>
> I found some test cases in llvm/clang that creates symbolic links to
> directories
>
> (which you seem to have been involved according to commit history).
>
>
>
> The problem is that the testcases basically are doing
>
>    ln -sf foo bar
>
> where foo is a directory.
>
>
>
> This will, as expected, create a symbolic link, bar -> foo, the first time
> the test executes.
>
>
>
> But if I run the test a second time (without clean in between), then ln will
> treat the
>
> already present bar as a target directory and create the symbolic link in
> that directory.
>
> So we basically get a link: foo/foo -> ../foo
>
>
>
>
>
> Man page for ln says:
>
>
>
> Usage: ln [OPTION]... [-T] TARGET LINK_NAME   (1st form)
>
>   or:  ln [OPTION]... TARGET                  (2nd form)
>
>   or:  ln [OPTION]... TARGET... DIRECTORY     (3rd form)
>
>   or:  ln [OPTION]... -t DIRECTORY TARGET...  (4th form)
>
>
>
> So basically we end up with “1st form” the first time, and “3rd form” the
> second time.
>
>
>
>
>
> All of this is not really a big problem, but it creates a recursive
> directory structure
>
> (and some poorly written programs that traverses the directory structure
> runs into problems..).
>
>
>
> One solution is to use -T option to always end up treating bar as a
> LINK_NAME and not a DIRECTORY.
>
> Another is to let the test case remove the old symlink (if it exists) before
> creating it again.
>
>
>
> The test cases that I’ve found (so far) that has this problem are:
>
>   llvm/test/MC/ELF/comp-dir.s
>
>   clang/test/Driver/no-canonical-prefixes.c
>
>   clang/test/Tooling/clang-check-pwd.cpp
>
>
>
>
>
> What do you think, should these tests be corrected into using “ln -sf -T”?
>

Looks like that option is missing from ln on OS X. How about just
adding a "rm -f" to delete any existing symbol links?

Cheers,
Rafael