Github user HeartSaVioR commented on the issue:
https://github.com/apache/spark/pull/22952
@zsxwing
Yeah, it would be ideal we can enforce `archivePath` to which don't have
any possibility to match against source path (glob), so my approach was to find
directory which is the base directory without having glob in ancestor, and
`archive path + base directory of source path` doesn't belong to sub-directory
of found directory.
For example, suppose source path is `/a/b/c/*/ef?/*/g/h/*/i`, then base
directory of source path would be `/a/b/c`, and `archive path + base directory
of source path` should not belong to sub-directory of `/a/b/c`.
(My code has a bug for finding the directory so need to fix it.)
This is not an elegant approach and the approach has false-positive, ending
up restricting the archive path which actually doesn't make overlap (too
restrict), but it would guarantee two paths never overlap. (So no need to
re-check when renaming file.)
I guess the approach might be reasonable because in practice end users
would avoid themselves have to think about complicated case on overlaps, and
just isolate two paths.
What do you think about this approach?
cc. @gaborgsomogyi Could you also help validating my approach?
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
