RF+ENH: rerun: Process inputs before unlocking/removing any files

When content isn't present for a file that was added/modified in a
revision, it'd be better to remove the file rather than trying to
unlock it, which fails.  But if we do that, we should make sure we
process inputs first since that step may retrieve an added/modified
file (perhaps via --input=.).  To do both these things, piggyback on
run's outputs argument.

datalad/interface/rerun.py

@@ -14,6 +14,7 @@
 import logging
 from itertools import dropwhile
 import json
+import os
 import re
 import sys
 
@@ -316,14 +317,15 @@ def __call__(
                 # bring back the entire state of the tree with #1424, but
                 # we limit ourself to file addition/not-in-place-modification
                 # for now
-                for r in ds.unlock(new_or_modified(ds, hexsha),
-                                   return_type='generator', result_xfm=None):
-                    yield r
+                auto_outputs = (os.path.relpath(ap["path"], ds.path)
+                                for ap in new_or_modified(ds, hexsha))
+                outputs = run_info.get("outputs", [])
+                auto_outputs = [p for p in auto_outputs if p not in outputs]
 
                 for r in run_command(run_info['cmd'],
                                      dataset=ds,
                                      inputs=run_info.get("inputs", []),
-                                     outputs=run_info.get("outputs", []),
+                                     outputs=outputs + auto_outputs,
                                      message=message or rev["run_message"],
                                      rerun_info=run_info):
                     yield r

datalad/interface/run.py

@@ -283,7 +283,9 @@ def run_command(cmd, dataset=None, inputs=None, outputs=None,
         'exit': cmd_exitcode if cmd_exitcode is not None else 0,
         'chain': rerun_info["chain"] if rerun_info else [],
         'inputs': inputs,
-        'outputs': outputs
+        # Get outputs from the rerun_info because rerun adds new/modified files
+        # to the outputs argument.
+        'outputs': rerun_info["outputs"] if rerun_info else outputs
     }
     if rel_pwd is not None:
         # only when inside the dataset to not leak information

datalad/interface/tests/test_run.py

@@ -157,6 +157,10 @@ def test_rerun(path, nodspath):
     # Or --since= to run all reachable commits.
     ds.rerun(since="")
     eq_('xxxxxxxxxx\n', open(probe_path).read())
+    # If a file is dropped, we remove it instead of unlocking it.
+    ds.drop(probe_path, check=False)
+    ds.rerun()
+    eq_('x\n', open(probe_path).read())
     # If the history to rerun has a merge commit, we abort.
     ds.repo.checkout("HEAD~3", options=["-b", "topic"])
     with open(opj(path, "topic-file"), "w") as f: