PeasyAI: default to collecting-mode compilation in subprocess.run

[ankushdesai]

Summary

PeasyAI's compile path shells out to p compile via three subprocess.run sites — none of which previously propagated the P_COMPILER_COLLECT_ERRORS env var. With P compiler 3.0+'s new multi-error mode (#957/#963/#965), enabling the env var lets PeasyAI's peasy-ai-fix-all loop converge in O(N/k) LLM round trips instead of O(N) where k is the average errors-per-iteration. This is the motivating use case for the entire multi-error initiative.

Change

At each of three subprocess.run sites:

env = os.environ.copy()
env.setdefault("P_COMPILER_COLLECT_ERRORS", "1")
result = subprocess.run([...], ..., env=env)

setdefault (not assignment) preserves the user's explicit override:

• Unset → child sees "1" (collecting mode, new default)
• Set to "0" / "false" → child sees the user's value (strict mode)
• Set to anything else → child sees the user's value

Sites changed

• Src/PeasyAI/src/core/services/compilation.py — canonical compile entry point used by peasy-ai-compile MCP tool + fix_iteratively loop
• Src/PeasyAI/src/utils/compile_utils.py — legacy helper for evaluation pipelines
• Src/PeasyAI/src/core/services/generation.py:1445 — candidate-scoring compile during code generation

Not changed

• Src/PeasyAI/src/utils/checker_utils.py — invokes p check, not p compile; env var has no effect there
• Src/PeasyAI/src/core/compilation/environment.py — toolchain validation calls (dotnet --list-sdks etc.), not P compilation
• The error parser already supports multiple errors via compilation.get_all_errors and error_parser.parse (both iterate all matches with re.finditer). No parser changes needed.

Follow-up (separate PR)

The multi-agent audit recommended updating fix_iteratively to call get_all_errors instead of parse_error and send all errors to the LLM in one batch. That's a behavioral change to the fix loop. This PR is the prerequisite (env var must propagate first); the batching work is a separate PR.

Test plan

• CI green (Python tests under Src/PeasyAI/tests/)
• Manual: run peasy-ai-fix-all on a P file with 3 independent errors; observe that the LLM sees all 3 in one round trip

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Propagates P_COMPILER_COLLECT_ERRORS=1 (via os.environ.copy() + setdefault) to all three subprocess.run sites in PeasyAI that invoke p compile, so the P 3.0+ collecting-mode default lets the fix loop see all type errors at once instead of one per round trip. The default is overridable by the user's environment.

Changes:

• Default P_COMPILER_COLLECT_ERRORS=1 in compilation.py's compile entry point.
• Same default propagated in legacy compile_utils.try_compile.
• Same default propagated in generation._compile_check_candidates.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
Src/PeasyAI/src/core/services/compilation.py	Adds os import and passes an env with collecting-mode default to the canonical p compile subprocess call.
Src/PeasyAI/src/utils/compile_utils.py	Builds an env with collecting-mode default and forwards it to the evaluation-pipeline p compile invocation.
Src/PeasyAI/src/core/services/generation.py	Builds a subprocess_env with collecting-mode default for the candidate-scoring p compile call.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[ankushdesai]

Superseded by #970, which makes multi-error mode the compiler default. PeasyAI no longer needs to propagate the env var because the compiler defaults to collecting mode for all callers. fix-all now converges in O(N/k) iterations automatically without any PeasyAI-side changes.