When loading the dependency table from cache we did only adjust its data types if we detected that the index was not stored as data type object as we switched from string to object when introducing the new pyarrow data types for the dependency table in version 1.7.0 of audb.
But we forgot, that we had used object as data type of the index before, and switched to string only in version 1.6.4 of audb:
Which means, if a user has a dependency table in cache stored with audb <1.6.4 when trying to publish a new version of a database, they will run into the issue reported in #436.
To fix it, I updated the code to always apply the data type conversion now. I also checked if it had any implications on execution time (as this was the main motivation for the if clause before), but I was not able to find any change in execution time.
Closes #436
When loading the dependency table from cache we did only adjust its data types if we detected that the index was not stored as data type
objectas we switched fromstringtoobjectwhen introducing the newpyarrowdata types for the dependency table in version 1.7.0 ofaudb.But we forgot, that we had used
objectas data type of the index before, and switched tostringonly in version 1.6.4 ofaudb:Which means, if a user has a dependency table in cache stored with
audb <1.6.4when trying to publish a new version of a database, they will run into the issue reported in #436.To fix it, I updated the code to always apply the data type conversion now. I also checked if it had any implications on execution time (as this was the main motivation for the
ifclause before), but I was not able to find any change in execution time.