Fixing jit.trace tracing a constant number of nodes with add_self_loops

[Vuenc]

Fix of issue #7226.

The problem I described came from the maybe_num_nodes function which computed int(edge_index.max()) + 1. The int() call made the computed number of nodes a constant for the torch.jit.trace function.

I fixed it with an if branch that is only executed while tracing. (I could not get the same code for all cases to work: the workaround used only for tracing now would also work for normal operation, but then would break when using torch.jit.script).

[codecov]

Codecov Report

Merging #7330 (e1cbbf8) into master (23c2836) will increase coverage by 0.00%.
The diff coverage is 100.00%.

❗ Current head e1cbbf8 differs from pull request most recent head ed36692. Consider uploading reports for the commit ed36692 to get more accurate results

@@           Coverage Diff           @@
##           master    #7330   +/-   ##
=======================================
  Coverage   91.53%   91.53%           
=======================================
  Files         438      438           
  Lines       24361    24364    +3     
=======================================
+ Hits        22298    22301    +3     
  Misses       2063     2063           

Impacted Files	Coverage Δ
torch_geometric/utils/num_nodes.py	97.05% <100.00%> (+0.28%)	⬆️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[rusty1s]

Thank you! I somehow missed the issue, so thanks for fixing it :) Cool solution.