| Frameworks | |
| Baloo |
Assume user has ~/stuff folder, that is symlinked to i.e. /storage/stuff.
User don't want it to be indexed, so he adds ~/stuff in the KCM. However, this entry gets silently ignored when indexer
runs over /storage (trivially because it does not match ~/stuff), resulting to excluded folder being indexed.
Instead I propose to resolve symlinks right when loading exclude folder list from config.
It compiles, it seems to be working
| No Linters Available |
| No Unit Test Coverage |
| Buildable 4935 | |
| Build 4953: arc lint + arc unit |
IMHO we should just disallow specifying symlinks in both include/excludeFolders. The user can just use exludeFolders = /storage/stuff if he wants to exclude it.
Why not? This kind of (partial) symlink support is quite easy to implement. And the more restrictions users have, the worse their experience is.
(I've just stumbled upon this issue randomly, when added some folder - not even thinking at that time that it is a symlink - and, after some time, found out that files there were indexed and my exclusion rule was silently ignored)
Because it can never be consistent.
What happens when I create two symlinks to the same folder, and put one link into includeFolders, the other one in excludeFolders?
What really should happen, the indexer should never follow symlinks, but only add files by their canonical path. This avoids a bunch of problems, symlink loops, nondeterministic pathes for files when these are added to the index, ...
I believe can do something better here.
I think if we stick to canonical paths everywhere, and resolve symlinks ASAP (but still follow them), that might solve all the problems.
Here the behavior would be the same as if user just add the same folder to both lists. Not quite sure how it works now (I guess, one of two rules will just pop first).
But here the undefined behavior would be acceptable, because it seems like user tried to shoot his leg intentionally :)
What really should happen, the indexer should never follow symlinks, but only add files by their canonical path. This avoids a bunch of problems, symlink loops, nondeterministic pathes for files when these are added to the index, ...
As for symlink loops, QDirIterator with FollowSymlinks seems to be able to handle this nicely (at least according to Qt docs).
As for nondeterministic pathes - just add canonical path, and that's it.
The only problem I see here is that if we try baloosearch -d folder/ and inside there will be symlink folder/subfolder -> /somewhere/else, we won't find any results inside subfolder.
This is less trivial, to work with it we need to store the whole FS graph (which is not tree anymore) with symlinks. But for now we can live without it, I suppose.