transfer_output_files = . erroneously transfers .machine.ad, so jobs cannot restart
LIGO reports that when users set
transfer_output_files = .
when_to_transfer_output = on_exit_or_evict
when jobs restart from an eviction, they are put on hold because they try to xfer back .machine.ad and .job.ad, but those files already exist, and are owned by condor.
Note that the original documentation was done with tag “HTCondor-267” [sic] and thus did not get picked up in the list of commits. This doc commit was done in SHA e82d1c139052e24b258532b670d17aba760b733c
Needs a version history item. (It may be helpful to test if transfer_output_files = .job.ad broke things before this patch – and verify that it doesn’t after – for writing the version history item, to document the fix as applicable as it actually is.)
The C++ is pretty; I approve. The code passes review, but brings up some follow-on issues I’ve written up as ticket HTCONDOR-268.