Fix out of memory bug #1270

[antimora]

Pull Request Template

Checklist

• Confirmed that run-checks all script has been executed.
• Made sure the book is up to date with changes in this PR.

Related Issues/PRs

#1270

Changes

1. Default value sequence generation uses a bound. This caused OOM in cases when a field was not present in .pt file.
2. Panic if a struct field is missing from pt file.
3. Add a new regression test.

Testing

Existing unit tests are passing and new test was added.

[codecov]

Codecov Report

Attention: 214 lines in your changes are missing coverage. Please review.

Comparison is base (9df2071) 84.41% compared to head (3e9c6cf) 84.48%.
Report is 25 commits behind head on main.

Files	Patch %	Lines
backend-comparison/src/burnbenchapp/base.rs	0.00%	58 Missing ⚠️
burn-dataset/src/vision/image_folder.rs	79.91%	49 Missing ⚠️
burn-wgpu/src/codegen/dialect/gpu/vectorization.rs	77.66%	23 Missing ⚠️
backend-comparison/src/burnbenchapp/term/base.rs	0.00%	17 Missing ⚠️
burn-wgpu/src/codegen/dialect/wgsl/base.rs	83.50%	16 Missing ⚠️
burn-wgpu/src/fusion/elemwise/builder.rs	93.83%	14 Missing ⚠️
burn-tch/src/ops/int_tensor.rs	0.00%	9 Missing ⚠️
burn-wgpu/src/codegen/dialect/wgsl/operations.rs	86.66%	4 Missing ⚠️
burn-wgpu/src/codegen/dialect/wgsl/shader.rs	33.33%	4 Missing ⚠️
backend-comparison/src/bin/burnbench.rs	0.00%	3 Missing ⚠️
... and 9 more

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1286      +/-   ##
==========================================
+ Coverage   84.41%   84.48%   +0.06%     
==========================================
  Files         549      561      +12     
  Lines       61952    62774     +822     
==========================================
+ Hits        52295    53032     +737     
- Misses       9657     9742      +85     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[laggui]

LGTM! Just some minor comments related to typos 👍

[nathanielsimard]

I suggest we consider renaming that struct at some point. Returning the default value isn't accurate for all deserialization types. The current name primarily defines 'how' it works, but it doesn't precisely convey 'what' task is performed by deserializing default values. My understanding is that the objective is to extract an empty data structure tree, creating any 'container' node with default values, while the leaf node should not undergo deserialization (deserialize_struct). Am I correct?

[antimora]

Yes, it has become hybrid now. Initially, I was thinking it would be ok to fill-in with default values for all unmatched "tree" branches. Although it works, it leads to odd outcomes (module with default values) if the field name mismatched. So here instead I opted to throw an error (field name + type).

I am open for a better name. We have some time to think since this is a private deserializer struct.