`MemoryError` during parse can escalate to `SystemError` and abort on small input

[StanFromIreland]

Crash report

What happened?

string = (
    b'(*h(((h(h(h(' +
    b'[' * 35 +
    b'*h(' +
    b'[' * 38 +
    b'**h(' +
    b'[' * 31 +
    b'(' * 4 +
    b'~' * 20 +
    b'F{' +
    b'[' * 27 +
    b'**h(' +
    b'[' * 31 +
    b'(' * 4 +
    b'~' * 23 +
    b'F{F"{F"{F"{F"{F"{F"{F"{FF""{{F"{""{F"{F{{F"{F"{F"\\}'
)

compile(string, "<input>", "single")

Gives me:

Fatal Python error: _Py_CheckFunctionResult: a function returned a result with an exception set
Python runtime state: initialized
MemoryError: Parser stack overflowed - Python source too complex to parse

The above exception was the direct cause of the following exception:

SystemError: <class 'SyntaxWarning'> returned a result with an exception set

Current thread 0x00007efd1b8e8780 [python] (most recent call first):
  File "/tmp/repro.py", line 19 in <module>
Aborted                    (core dumped) ./python /tmp/repro.py

Found by OSS-Fuzz.

CPython versions tested on:

CPython main branch

Operating systems tested on:

No response

Output from running 'python -VV' on the command line:

No response

[swap357]

I'm starting to look into this. I can confirm this reproduces on main,

cpython main  ❯ ./python.exe
Python 3.15.0a8+ (heads/main:299cf3c497, May 18 2026, 11:18:24) [Clang 17.0.0 (clang-1700.6.3.2)] on darwin
...
>>> compile(string, "<input>", "single")
Fatal Python error: _Py_CheckFunctionResult: a function returned a result with an exception set
Python runtime state: initialized
MemoryError: Parser stack overflowed - Python source too complex to parse

The above exception was the direct cause of the following exception:

SystemError: <class 'SyntaxWarning'> returned a result with an exception set
...

I couldn't reproduce this with Python 3.14.4 which gives expected SyntaxError.

[meghprkh]

It seems that there is a recursion limit of 6000 which this code-string seems to be exceeding. The recursion happens with function calls / expressions (~) and maybe other parens ([). This string is only 900 characters so there is the case that there is some pessimization of stack recursion in one of the operators being parsed.

[swap357]

I ran git bisect with the reproducer above. The first bad commit appears to be ccbe41e as the first bad commit.
(where exit code 1 SyntaxError is treated good, and exit code 134 with SystemError /Abort is bad)

[encukou]

It seems that we're missing some error handling around here:

diff --git a/Parser/parser.c b/Parser/parser.c
index c55c081dfc3..e50bad56da7 100644
--- a/Parser/parser.c
+++ b/Parser/parser.c
@@ -33420,6 +33420,10 @@ _loop0_79_rule(Parser *p)
         p->mark = _mark;
         D(fprintf(stderr, "%*c%s _loop0_79[%d-%d]: %s failed!\n", p->level, ' ',
                   p->error_indicator ? "ERROR!" : "-", _mark, p->mark, "fstring_middle"));
+        if (p->error_indicator) {
+            p->level--;
+            return NULL;
+        }
     }
     asdl_seq *_seq = (asdl_seq*)_Py_asdl_generic_seq_new(_n, p->arena);
     if (!_seq) {

The challenge now is to figure out where this is generated from :)