GitLab

schema_change: new implementation of Semaphores as separate objects that can cross pipeline database boundaries

Improvement of the ability to disconnect, which is necessary for pgsql to drop test databases. Does not always help - may need further analysis

bugfix: varchar(N) fields tend to shrink whereas char(N) do not, and the padding spaces left turn our numbers into strings

When trapping a die, if the job is incomplete, log as a WORKER_ERROR rather than WORKER_CAUTION (ENSCOMPARASW-975)

* version/2.4:
  bugfix: the interface of _get_table_list changed, it now requires a dbc

This can't be easily patched in sqlite

bugfix: discard_jobs_for_analysis_id() should only change the jobs that are _not_ involved in semaphores

It seems we may have to push a migration to Perl 5.24 in the next weeks ! Trying this on Travis, first

500 commits is more than the maximum gap between two schema changes ever

patchfix: sqlite has a single global space for index names, so we need to drop the old ones to avoid collisions

* version/2.4:
  patchfix: the patch was not patching the views
  patchfix: in PostreSQL, an enum value is not equal to its text counterpart, so the default value has to be changed
  patchfix: in PostreSQL, each column property must be changed with a separate ALTER COLUMN
  patchfix: this role_id column is supposed to be NOT NULL
  patchfix: the patch was not patching the views
  patchfix: the patch was not patching the views

This means that patches that add columns cannot produce the same new schema, so we can't use the ORDINAL position to compare the schemas