Because of a flaw in the logic, lastnotifyfailure was not set unless
there was a failure. This means that after 1 hour of no failures, the
function would execute every cron cycle, which has a pretty high cost
on large sites.
Together with this, refactoring and cleanup of search_references() and
similar methods was done. SQL was improved to use the INNER JOIN instead
of the LEFT JOIN as we are really interested in records that have a
reference. Also, joining the {repository_instances} table assures that
only references with valid repository_instance are returned (the check
against is_external_file() can't be applied to the methods that return
the count of references so I dropped it completely).
The combination of repositoryid and reference field is supposed to be a
unique identification of an external file. Because the reference is a
TEXT field, we can't use it to compose the index. So we use the
referencehash instead and the file API is responsible to keep it
up-to-date.
This patch unifies the way how records in {files_reference} get created.
Previously, each reference file (i.e. a file with referencefileid set)
created its own record in {files_reference}. This patch makes sure that
existing record is reused if possible.
Bye bye 1:1 relationships, you suck!
This replaces the previously defined external_file_exception that was
not used anywhere in the code. The new exception will be thrown in
situations where there is some unexpected problem detected with the
records in the {files_reference} table (e.g. unable to find a record
referenced by some file from {file}, invalid reference value etc.)
- changed function repository::copy_to_area so it accepts more file attributes, pass changed attributes to it
- changed file_info::copy_to_storage and file_info_stored::copy_to_storage to accept more file attributes
- simplified file size check for moodle files in repository/repository_ajax.php
- removed unused code in repository/repository_ajax.php
This commit moves from the batch SCOPE_IDENTITY() way to return
the inserted ids to the OUTPUT alternative (supported and
recommended since 2005). Also the handling of possible return
values is improved, with all the alternatives leading to expected results.
