[GHC] #7971: doesDirectoryExist description differs from its behavior

[GHC]

#7971: doesDirectoryExist description differs from its behavior
----------------------------------------+-----------------------------------
Reporter:  dsf                          |          Owner:                     
    Type:  bug                          |         Status:  new                
Priority:  normal                       |      Component:  libraries/directory
 Version:  7.6.3                        |       Keywords:                     
      Os:  Unknown/Multiple             |   Architecture:  Unknown/Multiple   
 Failure:  Incorrect result at runtime  |      Blockedby:                     
Blocking:                               |        Related:                     
----------------------------------------+-----------------------------------

Comment(by Fuuzetsu):

 I have created testDir: it was one level up. It seems that the formatting
 swallowed up ‘../’ as is also apparent in your own snippet (so my link was
 pointing to ‘../testDir’ which existed (I guess ‘/tmp/testDir’ should be
 used in the future to avoid swallowing up in the WikiFormatting blocks)).
 Having said that, I did some further investigation and here's what I
 uncovered:

 If the directory linked to exists in the same directory or in any
 subdirectories, doesDirectoryExists will return True. If the directory
 linked to exists higher up in the file hierarchy, it will return False.
 relative/absolute paths seem to not matter.

 So a link ‘/tmp/test/linkToSomething’ to ‘/tmp/something’ will return
 False even if ‘/tmp/something’ is present on the filesystem. A link in
 ‘/tmp/test/linkToOther’ to ‘/tmp/test/foo/bar/baz/someDir’ will return
 True given that the someDir exists.

 This also means that it might in fact be following symlinks: it just
 doesn't matter because the only links to return True for directories are
 in the same/subdirectory that is being removed so they would be gone
 anyway.

 I don't think this strange behaviour is by design and should be fixed, or
 carefully documented if it is by design.

-- 
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/7971#comment:9>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler