Add empty jinja block

[berceanu]

Address #457.

[codecov]

Codecov Report

Merging #458 (91cdca8) into master (e684554) will decrease coverage by 0.29%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #458      +/-   ##
==========================================
- Coverage   71.63%   71.34%   -0.30%     
==========================================
  Files          30       30              
  Lines        2958     2949       -9     
  Branches      547      543       -4     
==========================================
- Hits         2119     2104      -15     
- Misses        696      702       +6     
  Partials      143      143              

Impacted Files	Coverage Δ
flow/project.py	75.37% <0.00%> (-0.57%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update e684554...5f55973. Read the comment docs.

[vyasr]

@berceanu can you also modify stampede2.sh? You'll notice that it completely overwrites the body block, and the new body block has two separate branches based on a variable use_launcher. The branch corresponding to use_launcher == False should look basically like the body block from base_script.sh, and you can copy-paste this solution in there as well.

[csadorf]

This might be the academic within me, but can we maybe name the block pre_operation?

[bdice]

I'm going to attempt to test this PR (generating a sample script for Stampede2 and checking to see if it makes sense -- probably not going to run it).

[vyasr]

@bdice thanks for testing on Stampede2, that was going to be my main outstanding request. Once you are convinced that tests look ok (perhaps @b-butler can also take a look since he has looked at Stampede2 scripts before) @berceanu will need to regenerate the template testing scripts. See here for how to do that, but please hold off committing new versions of the testing output until @bdice confirms that it looks correct.

[bdice]

I tried a few different things here but did not succeed and don't have time to do any more at the moment. In the variations I tried, either the offsets were wrong or the template failed to render entirely. We might need to just revert the changes to the Stampede2 template so that it retains the previous behavior. I really don't like that machine, it presents so many problems for signac-flow...

[vyasr]

@bdice I think I've resolved the problem, aside from the known issue of pretend outputs not incrementing the counter correctly (see here for a refresher on that). Can you test on stampede again when you get a chance, and see if you're still getting output that you don't think makes sense? There's no way to validate the offset itself without actually submitting, but if you run the template tests and you see that the offset is always 0 in both cases (what I'd expect) then the output seems consistent.

If that seems right to you, then the reference output needs to be regenerated and this PR should be good to go.

[bdice]

@vyasr your changes appear to work for me. Thanks! I did not test on the actual Stampede2 machine but I verified that the generated templates from our tests all say export _FLOW_STAMPEDE_OFFSET_=0.

Changes resolved.

[bdice]

Ugh, I committed with a custom setting in my $HOME/.signac.rc. Let me fix that.