You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[macOS] csync errors do not appear in new error view and block the sync process #5888
I was testing testing out the before/after of applying #5875 on the client under macOS and noticed a couple of odd behaviors there:
Expected behavior
The [ warning sync.csync ]: Invalid characters in file/directory name, please rename: "💃" (/Users/<user>/ownCloud) propagates to the new "List of Issues" (Add a more functional error view #5516 #5861) under the "Not Synced" tab
The overlay icon of the erroneous file switches to ❌
Create a local folder with an emoji as filename "💃"
See the account tab
I know the specific issue about the file name encoding will be solved after #5875 and I still need to check if other csync warnings display the same behavior and if it also happens on different platforms.
@SamuAlfageme You're quick; I think I've already addressed the issue with the errors tab in a local patch and thought I could fix it without anyone noticing. :)
The overlay or sync-aborting properties are independent of that though.
Even then, emoji in filenames should not have been blocking the sync for a long time. What makes you think the sync is blocked? The red just say there was an error during the sync, but the sync should have completed
@guruz I'm not sure which patch you're referring to. Your log output suggests that this is indeed an error that will abort the whole sync run after the discovery phase? This part of this issue doesn't seem fixed. (but I'd need osx to verify)
@guruz. Right, it was aborting the sync.
But this should not happen anymore with all the refactoring, normally. Does it?
Or what is this bug about really?
Yes, this seems to have happened when csync_vio_local_readdir encounters a dirent->d_name that c_utf8_from_locale can't convert. However, I don't see how that function would return an empty string on OSX in the current version of the code unless the input is null.
Not reproducible anymore - tested out against a huge list of unicode random strings accepted by HFS+/APFS and neither errors were raised in the client side, nor the sync was locked at any point.
Activity
ckamm commentedon Jul 11, 2017
@SamuAlfageme You're quick; I think I've already addressed the issue with the errors tab in a local patch and thought I could fix it without anyone noticing. :)
The overlay or sync-aborting properties are independent of that though.
EDIT: PR that has the fix: #5890
ckamm commentedon Jul 21, 2017
The "csync errors don't appear in new error view" is addressed, the "block the sync process" isn't addressed.
guruz commentedon Jul 21, 2017
FYI That error is created early by the csync discovery already, not by an individual upload job.
Also note that with 2.4 emojis won't block the sync, but invalid utf8 still will (i think)
ogoffart commentedon Jul 21, 2017
Even then, emoji in filenames should not have been blocking the sync for a long time. What makes you think the sync is blocked? The red just say there was an error during the sync, but the sync should have completed
guruz commentedon Jul 21, 2017
(This is 2.3 branch before @ogoffart 's patch)
ckamm commentedon Oct 11, 2017
@guruz I'm not sure which patch you're referring to. Your log output suggests that this is indeed an error that will abort the whole sync run after the discovery phase? This part of this issue doesn't seem fixed. (but I'd need osx to verify)
ogoffart commentedon Oct 17, 2017
@guruz. Right, it was aborting the sync.
But this should not happen anymore with all the refactoring, normally. Does it?
Or what is this bug about really?
ckamm commentedon Oct 24, 2017
Yes, this seems to have happened when
csync_vio_local_readdirencounters adirent->d_namethatc_utf8_from_localecan't convert. However, I don't see how that function would return an empty string on OSX in the current version of the code unless the input is null.ckamm commentedon Oct 26, 2017
@SamuAlfageme Is this reproducible with 2.4?
ogoffart commentedon Nov 28, 2017
Right, 2.3 might have had the issue (i tought it was fixed, but there was a
goto errorthat broke it)Anyway, 2.4 should now really be fixed.
SamuAlfageme commentedon Dec 12, 2017
Not reproducible anymore - tested out against a huge list of unicode random strings accepted by HFS+/APFS and neither errors were raised in the client side, nor the sync was locked at any point.
🤙 Close here