Preprocessing files with zero events

[alexander-held]

Describe the bug
When running preprocessing over files containing zero events, num_entries will be 0 and a division in the step size determination will fail with ZeroDivisionError here: https://github.com/CoffeaTeam/coffea/blob/e06c4b84d0a641ab569ae7c16fecc39fe74c9743/src/coffea/dataset_tools/preprocess.py#L127

I assume the solution would be to bypass all of https://github.com/CoffeaTeam/coffea/blob/e06c4b84d0a641ab569ae7c16fecc39fe74c9743/src/coffea/dataset_tools/preprocess.py#L79-L156 in this case to instead run https://github.com/CoffeaTeam/coffea/blob/e06c4b84d0a641ab569ae7c16fecc39fe74c9743/src/coffea/dataset_tools/preprocess.py#L158-L173
and return.

To Reproduce

import uproot
import awkward as ak
from coffea import dataset_tools

arr = ak.Array([])
with uproot.recreate("f.root") as f:
    f["tree"] = {"arr": arr}

fileset = {"dummy": {"files": {"f.root": "tree"}}}
dataset_tools.preprocess(fileset)

The actual case where I saw this happen contained a tree without any arrays inside. I'm not sure if uproot supports writing that for a reproducer (I don't think so?) but in either case I think only the .num_entries property matters so the tree content should be irrelevant here.

Expected behavior
handle empty files without crash

Output

  File "[...]/lib/python3.11/site-packages/coffea/dataset_tools/preprocess.py", line 127, in get_steps
    n_steps_target = max(round(num_entries / target_step_size), 1)
                               ~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~
ZeroDivisionError: division by zero

Desktop (please complete the following information):
n/a

Additional context
cc @sebastien-rettie

[alexander-held]

I'm happy to PR if the solution makes sense to you.

[sebastien-rettie]

Hi @alexander-held, thanks a lot! This makes sense to me.

[alexander-held]

I had a brief look and believe the solution could be this

diff --git a/src/coffea/dataset_tools/preprocess.py b/src/coffea/dataset_tools/preprocess.py
index cef72ad5..69ffe3fa 100644
--- a/src/coffea/dataset_tools/preprocess.py
+++ b/src/coffea/dataset_tools/preprocess.py
@@ -76,6 +76,9 @@ def get_steps(

         num_entries = tree.num_entries

+        if num_entries == 0:
+            continue
+
         form_json = None
         form_hash = None
         if save_form:

[lgray]

Please make a PR with this change. We should make this configurable by the user (the continue behavior should be on by default).

[lgray]

No one ever made a PR for this.