[Discovery] Fix possible issue with retrieval of permissions from the server

[guruz]

Detected by @nickvergessen in owncloud/smashbox#134 (comment)

[ckamm]

I have verified the code that reads remotePerm from the database. The csync_file_stat_t is nulled correctly and a NULL db value leads to an empty remotePerm, as expected.

[ckamm]

Relevant lines from the logs at https://gist.github.com/nickvergessen/8f5b8ea746a32a7a934b:

step03:
SyncJournalFileRecord("owner_worker/dir") 
"INSERT OR REPLACE INTO metadata (remotePerm) VALUES ("")

step05:
csync_walker:  directory: owner_worker/dir
_csync_detect_update:  Database entry found, compare:   <-> RDNVW

[ckamm]

@nickvergessen This is still confusing. It looks like step03 writes NULL, but step05 reads "RDNVW". Could you verify that this is what's going on by running the test only after step03 and then checking the database?

Use something like sqlitebrowser on .csync_journal.db in the sync directory and then look for the "dir" row in the metadata table. What's the value of the remotePerm column?

[nickvergessen]

The permissions are saved as RDNVW in the DB after step 3.
Running oc_sync a second time would fix the faulty permissions.

[ckamm]

@nickvergessen Thanks!

[ckamm]

I think the problem is this code in SqlQuery::bindValue

                // unbound value create a null entry.
                res = SQLITE_OK;

i.e. we don't call sqlite3_bind_null() when we want to produce a null table entry. Since bindings are not cleared by sqlite3_reset() the previous file's bindings are inadvertantly written to the database.

I'll try to reproduce the problem and will then fix it.

[nickvergessen]

I just built a new master client, so it includes 7bd4f95

Fixed the issue and my locking test finally passes in all cases 👍

[dragotin]

great finding @ckamm and @nickvergessen thanks!