4.3. Retrace and reenact

“Thanks a lot for sharing your dataset with me! This is super helpful. I’m sure I’ll catch up in no time!”, your room mate says confidently. “How far did you get with the DataLad commands yet?” he asks at last.

“Mhh, I think the last big one was datalad run. Actually, let me quickly show you what this command does. There is something that I’ve been wanting to try anyway.” you say.

The dataset you shared contained a number of datalad run commands. For example, you created the simple podcasts.tsv file that listed all titles and speaker names of the longnow podcasts.

Given that you learned to create “proper” datalad run commands, complete with --input and --output specification, anyone should be able to datalad rerun these commits easily. This is what you want to try now.

You begin to think about which datalad run commit would be the most useful one to take a look at. The creation of podcasts.tsv was a bit dull – at this point in time, you didn’t yet know about --input and --output arguments, and the resulting output is present anyway because text files like this .tsv file are stored in Git. However, one of the attempts to resize a picture could be useful. The input, the podcast logos, is not yet retrieved, nor is the resulting, resized image. “Let’s go for this!”, you say, and drag your confused room mate to the computer screen.

First of all, find the commit shasum of the command you want to run by taking a look into the history of the dataset (in the shared dataset):

# navigate into the shared copy
$ cd ../mock_user/DataLad-101
# lets view the history
$ git log --oneline -n 10
993e896 add note on clean datasets
603752c [DATALAD RUNCMD] Resize logo for slides
ba5f032 [DATALAD RUNCMD] Resize logo for slides
d9085ab add additional notes on run options
41203ec [DATALAD RUNCMD] convert -resize 450x450 recordings/longn...
02f06d3 resized picture by hand
c5884c5 [DATALAD RUNCMD] convert -resize 400x400 recordings/longn...
4694d46 add note on basic datalad run and datalad rerun
f8578cc add note datalad and git diff
5e1c5a3 [DATALAD RUNCMD] create a list of podcast titles

Ah, there it is, the second most recent commit. Just as already done in section DataLad, Re-Run!, take this shasum and plug it into a datalad rerun command:

$ datalad rerun 603752cd9bce646518bbfb3a77b4127d649f82b9
[INFO] run commit 603752c; (Resize logo for s...)
[INFO] Making sure inputs are available (this may take some time) 
get(ok): recordings/longnow/.datalad/feed_metadata/logo_salt.jpg (file) [from web...]
run.remove(ok): recordings/salt_logo_small.jpg (file) [Removed file]
[INFO] == Command start (output follows) ===== 
[INFO] == Command exit (modification check follows) ===== 
run(ok): /home/me/dl-101/mock_user/DataLad-101 (dataset) [convert -resize 400x400 recordings/longn...]
add(ok): recordings/salt_logo_small.jpg (file)
action summary:
  add (ok: 1)
  get (notneeded: 1, ok: 1)
  run (ok: 1)
  run.remove (ok: 1)
  save (notneeded: 2)

“This was so easy!” you exclaim. DataLad retrieved the missing file content from the subdataset and it tried to unlock the output prior to the command execution. Note that because you did not retrieve the output, recordings/salt_logo_small.jpg, yet, the missing content could not be unlocked. DataLad warns you about this, but proceeds successfully.

Your room mate now not only knows how exactly the resized file came into existence, but he can also reproduce your exact steps to create it. “This is as reproducible as it can be!” you think in awe.