Promote generic utils/constants from gigl.src.common to gigl.common

[kmontemayor2-sc]

Promote the reusable, application-agnostic helpers out of the pipeline source tree into the top-level gigl.common package:

• gigl/common/utils/{timeout,file_loader,model,time,bq}.py
• datetime format constants merged into gigl/common/constants.py

Each old gigl.src.common path is left as a re-export shim that logs a deprecation warning, so external imports keep working. All internal call sites (gigl, examples, scripts, tests) are repointed to the new gigl.common locations; the dedicated bq tests move into the common test tree.

Application/task logic (pb_wrappers, task types, launchers, pipeline path constants, translators, graph_builder, modeling_task_specs) stays in gigl.src by design.

Scope of work done

Where is the documentation for this feature?: N/A

Did you add automated tests or write a test plan?

Updated Changelog.md? NO

Ready for code review?: NO

[kmontemayor2-sc]

/all_test

[github-actions]

GiGL Automation

@ 18:09:22UTC : 🔄 Integration Test started.

[github-actions]

GiGL Automation

@ 18:09:22UTC : 🔄 Scala Unit Test started.

@ 18:18:35UTC : ✅ Workflow completed successfully.

[github-actions]

GiGL Automation

@ 18:09:23UTC : 🔄 Python Unit Test started.

@ 19:16:31UTC : ✅ Workflow completed successfully.

[github-actions]

GiGL Automation

@ 18:09:24UTC : 🔄 C++ Unit Test started.

@ 18:11:18UTC : ✅ Workflow completed successfully.

[github-actions]

GiGL Automation

@ 18:09:25UTC : 🔄 Lint Test started.

@ 18:17:47UTC : ✅ Workflow completed successfully.

[github-actions]

GiGL Automation

@ 18:09:32UTC : 🔄 E2E Test started.

@ 19:58:59UTC : ✅ Workflow completed successfully.

[semgrep-code-snapchat]

Semgrep identified an issue in your code:

Using torch.load() to deserialize model files enables arbitrary code execution via pickle. Attackers can embed malicious Python code in model files that executes when loaded.

More details about this

The torch.load() call on tmp_file.name deserializes a PyTorch model file using pickle, which can execute arbitrary Python code embedded in the serialized data.

Exploit scenario:

1. An attacker crafts a malicious PyTorch model file containing embedded Python code (e.g., code to exfiltrate data, install backdoors, or launch ransomware).
2. This file is uploaded or placed where your application loads it via load_from_uri.
3. When torch.load(tmp_file.name, map_location=device) executes, pickle deserializes the malicious payload and automatically executes the attacker's code with the full privileges of your application.
4. The attacker now has arbitrary code execution on your system.

This is particularly dangerous if load_from_uri accepts untrusted sources (e.g., user uploads, external URLs, or any input not under your strict control).

To resolve this comment:

✨ Commit fix suggestion

1. Change the torch.load call to load weights only: replace torch.load(tmp_file.name, map_location=device) with torch.load(tmp_file.name, map_location=device, weights_only=True).
2. Keep this function limited to files created by save_state_dict(...), because that path writes model.state_dict() instead of a full pickled model object. This prevents PyTorch from deserializing arbitrary Python objects from the file.
3. Add a type check after loading and reject anything that is not a state dict, for example by verifying the result is a dict or OrderedDict and that the values are tensors.
4. Leave load_scripted_model_from_uri(...) separate from this path, and do not reuse load_state_dict_from_uri(...) to load full models or other serialized objects.

Alternatively, if weights_only=True is not available in your PyTorch version, upgrade PyTorch and use weights_only=True rather than loading the file with plain torch.load(...).

💬 Ignore this finding

Reply with Semgrep commands to ignore this finding.

• /fp <comment> for false positive
• /ar <comment> for acceptable risk
• /other <comment> for all other reasons

Alternatively, triage in Semgrep AppSec Platform to ignore the finding created by pickles-in-pytorch.

_{You can view more details about this finding in the Semgrep AppSec Platform.}