gh-151763: Restore exception if parser recovery path leaves none set (OOM-0021)

[Vamsi-klu]

Summary

Fix OOM-0021 under umbrella gh-151763: the parser recovery / diagnostic second pass can return NULL with no exception set, which later trips Py_FatalError("a function returned NULL without setting an exception") in _Py_CheckFunctionResult.

---

What the issue is

_PyPegen_run_parser already guards res != NULL && PyErr_Occurred() (OOM-0013). On res == NULL it still runs a heavier second parse pass and _PyPegen_set_syntax_error_metadata, which can PyErr_Clear() internally. Under sustained OOM the recovery path can finish with tstate->current_exception == NULL and return NULL, violating the C-API result/error contract. Callers such as the symtable path then fatal in Objects/call.c.

---

Why I solved it that way

1. Restore the contract at the return site with:

   if (!PyErr_Occurred()) {
       PyErr_NoMemory();
   }

   This matches the idiom already used nearby after OOM-0013 (Parser/pegen.c other arms). PyErr_NoMemory() is safe under OOM (preallocated singleton).

2. Do not remove _Py_CheckFunctionResult's fatal — that enforces the contract for the whole C API; weakening it would hide extension bugs.

3. Do not yet rewrite _PyPegen_set_syntax_error_metadata to save/restore around its PyErr_Clear calls — that is a nicer long-term fix but larger and riskier; the belt at the return site closes the crash now.

4. Separate PR from other OOM findings per umbrella policy (one PR per bug).

---

How I did it

At the end of the res == NULL block in _PyPegen_run_parser, after syntax-error metadata setup and before return NULL:

if (!PyErr_Occurred()) {
    /* Under OOM the recovery pass can clear the exception and fail
       silently; restore the result/error contract. */
    PyErr_NoMemory();
}
return NULL;

NEWS under Core and Builtins for gh-151763 (OOM-0021).

---

Impact

• Converts a hard fatal / abort under parser OOM into a normal MemoryError.
• Success-path parsing unchanged.
• Complements OOM-0013 rather than duplicating it (that fix covered the res != NULL contradiction).

---

Testing plan

• Rebuild debug interpreter.
• Parser-related tests via normal suite smoke; full test_peg_generator needs extra resources (skipped in default run).
• Optional: fusil/OOM-0021 gist repro with set_nomemory sweep — not automated here.
• CI: full parser / test_grammar / test_syntax jobs.

---

Everything else

• Umbrella CPython crashes on memory-allocation failure (OOM): 35 findings (4 dups → 31 distinct) #151763, finding OOM-0021.
• Related merged fix: OOM-0013 (gh-151763: Fix OOM-0013 crash when the parser or compiler fails to allocate #151968).
• Reviewers: parser / pegen owners + OOM umbrella.

---

Requested reviewers (subject-matter experts)

Could the following SMEs take a look when convenient (I cannot formally request reviews from this fork account):

• lysnikolaou / pablogsal — parser / pegen
• devdanzin — OOM umbrella

Thank you!

[python-cla-bot]

All commit authors signed the Contributor License Agreement.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Another round soon, please!

Reviewed commit: f459c4bdf1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Vamsi-klu]

Closing this PR. Thank you for the review attention.