[Analyze] Symlinks in library

[letsfindaway]

As mentioned in OpenBoard-org#1167 (comment), symlinks do not work within the library. I just made a quick check and found:

• When building the library palette, the file system is scanned in UBFeaturesController::fileSystemScan().
• This used the function UBFileSystemUtils::allElementsInDirectory():
  

  OpenBoard/src/frameworks/UBFileSystemUtils.cpp

  Lines 247 to 254 in 80fb7dd

  	QFileInfoList UBFileSystemUtils::allElementsInDirectory(const QString& pDirPath)
  	{
  	QDir dir = QDir(pDirPath);
  	dir.setFilter(QDir::Files | QDir::Dirs | QDir::NoDotAndDotDot | QDir::NoSymLinks);
  	dir.setSorting(QDir::DirsFirst);
  	return QFileInfoList(dir.entryInfoList());
  	}

• Here symlinks are explicitly excluded.

But why? That line in the code dates back to 2011 and was taken from OpenSankore and never modified. Was it intentionally for some specific reason? Analyze this and find out whether we could just remove this flag.

[kaamui]

is there any risk of circular paths with symlinks maybe ?

[letsfindaway]

is there any risk of circular paths with symlinks maybe ?

Indeed, there is that risk. However this can be mitigated by using QFileInfo::canonicalPath() to check that you're not traversing upward when following a link.

[letsfindaway]

Cyclic paths and multiple scanning can be avoided as follows:

• We keep a list of roots, i.e. entry points into the file system where a subtree scanning starts. This list contains only canonical paths, i.e. with all symlinks resolved.
• We add out initial starting point to this list.
• We then do a typical path traversal. It does not matter whether we traverse width first or depth first.
• When we encounter a symlink, then we check the link target:
  • If it is above any of the roots in the list, then the symlink is invalid because it causes cyclic or duplicate scanning
  • If it is below any of the roots in the list, then the symlink is also invalid, because it causes duplicate scanning.
  • If it points to a non-existing location, then it is also invalid and skipped.
  • Otherwise it points to a location outside of any currently scanned trees. The symlink is valid and we add the target to the list of roots.

Invalid symlinks are simply skipped and a debug message is written to the log. They are not a fatal error.