fix(describe): guard pr_description config reads against missing keys (#2238)

[naorpeled]

What

Fixes the 'DynaBox' object has no attribute 'enable_large_pr_handling' crash on /describe.

Closes #2238

Why it still happens after #2234

#2234 restored enable_large_pr_handling / max_ai_calls / async_ai_calls as defaults in configuration.toml, but the reads were left as direct attribute access. The catch is custom_merge_loader replaces a whole section instead of merging fields (it documents this in its own docstring). So any user whose custom .pr_agent.toml has a [pr_description] section that omits these keys loses the defaults at runtime, and the unguarded read throws. retry_with_fallback_models swallows it as a failed prediction and keeps trying models until the run fails - which is exactly what a v0.34 user with an older config reported.

The fix

Read the three keys via .get(..., default) at the call sites, so the documented defaults apply even when a custom config replaces the section:

• pr_agent/tools/pr_description.py:211 - enable_large_pr_handling (default True)
• pr_agent/tools/pr_description.py:247 - async_ai_calls (default True)
• pr_agent/algo/pr_processing.py:262 - max_ai_calls (default 4)

Defaults match configuration.toml.

Test

Added test_pr_description_reads_fall_back_when_keys_missing, which reproduces the overridden section with a DynaBox missing the keys, asserts bare attribute access raises the original error, and verifies the guarded reads fall back to the defaults.

[qodo-free-for-open-source-projects]

PR Summary by Qodo

Fix /describe crash when pr_description config keys are missing
🐞 Bug fix 🧪 Tests 🕐 10-20 Minutes

Description

• Prevent /describe from crashing when custom .pr_agent.toml omits pr_description defaults.
• Use guarded Dynaconf reads with explicit defaults for large-PR handling and AI call limits.
• Add regression test covering missing-key behavior on DynaBox config sections.

Diagram

graph TD
  user(("User")) --> describe["/describe"] --> cfg[("Dynaconf settings")] --> prdesc["pr_description.py"] --> prproc["pr_processing.py"] --> llm{{"LLM models"}}
  toml[["Custom .pr_agent.toml"]] --> cfg

  subgraph Legend
    direction LR
    _actor(("Actor")) ~~~ _api["API/Command"] ~~~ _cfg[("Config")] ~~~ _mod["Module"] ~~~ _ext{{"External"}} ~~~ _file[["Config file"]]
  end

Loading

High-Level Assessment

The following are alternative approaches to this PR:

1. Deep-merge config sections in custom_merge_loader

• ➕ Fixes this class of missing-default bugs across all sections, not just pr_description
• ➕ Keeps call sites simple (attribute access remains safe)
• ➖ Higher blast radius: changes config semantics for existing users
• ➖ Requires careful precedence/merge rules and additional tests across settings layout

2. Centralize defaults via schema/validation step on settings load

• ➕ Single place to enforce required keys and defaults
• ➕ Can emit clearer diagnostics when keys are missing/mistyped
• ➖ More implementation overhead than a targeted fix
• ➖ Still requires decisions about how to handle partially overridden sections

Recommendation: The PR’s targeted .get(..., default) reads are the best immediate fix: low risk, aligns with documented defaults, and directly prevents /describe crashes for partially overridden [pr_description] sections. Consider a follow-up to address section replacement behavior more broadly (deep-merge or validation) if similar issues appear elsewhere.

Files changed (3) +23 / -3

Bug fix (2) +3 / -3

pr_processing.pyDefault-safe read for pr_description.max_ai_calls +1/-1

Default-safe read for pr_description.max_ai_calls

• Replaces direct attribute access to 'pr_description.max_ai_calls' with 'get("max_ai_calls", 4)' when computing the number of allowed iterations for large-PR diff generation. Prevents AttributeError when a user overrides the [pr_description] section without this key.

pr_agent/algo/pr_processing.py

pr_description.pyGuard missing pr_description keys for large-PR + async behavior +2/-2

Guard missing pr_description keys for large-PR + async behavior

• Switches 'enable_large_pr_handling' and 'async_ai_calls' reads from attribute access to 'get(..., default)' with defaults matching configuration.toml. This keeps /describe functional when a custom config replaces the section and drops these keys.

pr_agent/tools/pr_description.py

Tests (1) +20 / -0

test_pr_processing_core.pyRegression test for missing-key fallback behavior in DynaBox +20/-0

Regression test for missing-key fallback behavior in DynaBox

• Adds a unit test that reproduces a [pr_description] section missing large-PR keys using 'DynaBox', asserts old-style attribute access raises AttributeError, and verifies '.get(..., default)' returns the expected documented defaults.

tests/unittest/test_pr_processing_core.py

[qodo-free-for-open-source-projects]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0)

Great, no issues found!

Qodo reviewed your code and found no material issues that require review

• Author self-review: I have reviewed the code review findings, and addressed the relevant ones.