mirrorring relative symlinks with ../.. in them which are wrong...

MetPX / sarracenia

https://MetPX.github.io/sarracenia

GNU General Public License v2.0

46 stars 22 forks source link

mirrorring relative symlinks with ../.. in them which are wrong... #1270

Closed petersilva closed 1 month ago

petersilva commented 1 month ago

so.. when mirroring with libsr3shim:

shim posting:

realpathFilter on
realpathPost on
realpathAdjust -1

subscriber:

baseDir /source
directory /dest

now there are paths which are /home/user/excellent (where excellent is a link into /source/excellent) so the path of operations would be:

cd /home/user/excellent
mv a b
mkdir oneDirDeep
mkdir oneDirDeep/twoDirDeep
echo "hoho" >>oneDirDeep/twoDirDeep/2dd_file1
ln -s 2dd_file1 oneDirDeep/twoDirDeep/link_to_2dd_file1
mv  oneDirDeep/twoDirDeep/link_to_2dd_file1  oneDirDeep/twoDirDeep/renamed_link_to_2dd_file1

you end up with links with (wrong) extra ../../ in them. also rename... maybe same issue maybe different.

petersilva commented 1 month ago

user found this...

petersilva commented 1 month ago

OK, what I think is going on:

when you are processing file renames, removes, and symlinks. the cwd is the directory. So the rename's relative path needs to be adjusted to make it relative to directory ...
I applied that relative path adjustment logic to all fileOps ('hlink', 'link', 'rename') but... 'link' is a symbolic link. the argument supplied is just written into the link, so no adjustment is needed.
so 'hlink' (hard links) and 'rename' do need adjustment, but symbolic links do not.

petersilva commented 1 month ago

I added some test cases to sr_insects to cover this case, and they are fixed by the proposed patch.

petersilva commented 1 month ago

Question... would it not be simpler to just chdir into the directory, and do relative ops from there... then no adjustment would be needed. hmm.

I look at the logic in do_download... and looks like it is doing a chdir... before the fileOp... so I don't understand why the relative path adjustment is needed at all. but if I remove if for renames, my test cases break.

petersilva commented 1 month ago

After the problem with sarrac was corrected, now tests pass when all weird processing of relative rename and link passes all tests.