REPL exits with traceback when str(sys.ps1) raises

[gvanrossum]

Bug report

Bug description:

Subject says it all. This is on the latest HEAD of main.

~/cpython$ g pull
Already up to date.
~/cpython$ make
Checked 112 modules (34 built-in, 78 shared, 0 n/a on macosx-14.6-x86_64, 0 disabled, 0 missing, 0 failed on import)
./python.exe -E ./Tools/build/generate-build-details.py `cat pybuilddir.txt`/build-details.json
~/cpython$ ./python.exe 
Python 3.14.0a5+ (heads/main:ab11c097052, Feb 28 2025, 08:56:47) [Clang 16.0.0 (clang-1600.0.26.6)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import sys
>>> class C:
... def __str__(self): 1/0
... 
>>> sys.ps1 = C()
Traceback (most recent call last):
 File "<frozen runpy>", line 198, in _run_module_as_main
 File "<frozen runpy>", line 88, in _run_code
 File "/Users/guido/cpython/Lib/_pyrepl/__main__.py", line 6, in <module>
 __pyrepl_interactive_console()
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
 File "/Users/guido/cpython/Lib/_pyrepl/main.py", line 59, in interactive_console
 run_multiline_interactive_console(console)
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^
 File "/Users/guido/cpython/Lib/_pyrepl/simple_interact.py", line 143, in run_multiline_interactive_console
 statement = multiline_input(more_lines, ps1, ps2)
 File "/Users/guido/cpython/Lib/_pyrepl/readline.py", line 389, in multiline_input
 return reader.readline()
 ~~~~~~~~~~~~~~~^^
 File "/Users/guido/cpython/Lib/_pyrepl/reader.py", line 796, in readline
 self.prepare()
 ~~~~~~~~~~~~^^
 File "/Users/guido/cpython/Lib/_pyrepl/historical_reader.py", line 306, in prepare
 super().prepare()
 ~~~~~~~~~~~~~~~^^
 File "/Users/guido/cpython/Lib/_pyrepl/reader.py", line 643, in prepare
 self.calc_screen()
 ~~~~~~~~~~~~~~~~^^
 File "/Users/guido/cpython/Lib/_pyrepl/completing_reader.py", line 261, in calc_screen
 screen = super().calc_screen()
 File "/Users/guido/cpython/Lib/_pyrepl/reader.py", line 371, in calc_screen
 prompt = self.get_prompt(ln, ll >= pos >= 0)
 File "/Users/guido/cpython/Lib/_pyrepl/historical_reader.py", line 326, in get_prompt
 return super().get_prompt(lineno, cursor_on_line)
 ~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^
 File "/Users/guido/cpython/Lib/_pyrepl/reader.py", line 552, in get_prompt
 prompt = f"{ANSIColors.BOLD_MAGENTA}{prompt}{ANSIColors.RESET}"
 ^^^^^^^^
 File "<python-input-1>", line 2, in __str__
 def __str__(self): 1/0
 ~^~
ZeroDivisionError: division by zero
~/cpython$ 

Probably @pablogsal should look into this.

CPython versions tested on:

CPython main branch

Operating systems tested on:

macOS

Linked PRs

• gh-130698: Add safe methods to get prompts for new REPL #131110

[pablogsal]

CC @ambv @lysnikolaou

[sergey-miryanov]

Easy fix is

git diff
diff --git a/Lib/_pyrepl/reader.py b/Lib/_pyrepl/reader.py
index 1252847e02b..6d41802bce0 100644
--- a/Lib/_pyrepl/reader.py
+++ b/Lib/_pyrepl/reader.py
@@ -548,6 +548,11 @@ def get_prompt(self, lineno: int, cursor_on_line: bool) -> str:
 else:
 prompt = self.ps1

+ try:
+ prompt = str(prompt)
+ except:
+ prompt = ">>> "
+
 if self.can_colorize:
 prompt = f"{ANSIColors.BOLD_MAGENTA}{prompt}{ANSIColors.RESET}"
 return prompt

But maybe we should handle unrecoverable exceptions like MemoryError, SystemExit and stop REPL?

[pablogsal]

@sergey-miryanov wanna prepare a PR for this @sergey-miryanov ?

[sergey-miryanov]

@pablogsal yes, will do.

[sergey-miryanov]

@pablogsal I just have added a PR. Please take a look.

[skirpichev]

Note that old repl just clear error indicator and return empty prompt:

cpython/Python/pythonrun.c

Lines 228 to 237 in 8a9c6c4

	const char *ps1 = "";
	if (_PySys_GetOptionalAttr(&_Py_ID(ps1), &attr) < 0) {
	PyErr_Clear();
	}
	else if (attr != NULL) {
	ps1_obj = PyObject_Str(attr);
	Py_DECREF(attr);
	if (ps1_obj == NULL) {
	PyErr_Clear();
	}

But maybe we should handle unrecoverable exceptions like MemoryError, SystemExit and stop REPL?

MemoryError isn't really unrecoverable, unless you have some bug in your C code. The SystemExit is more questionable.

I think it's ok to replicate old-repl behavior, but instead of being just silent - emit a warning.

[sergey-miryanov]

@skirpichev I don't know what is the best way - replicate old behavior, show default prompt (as I did in PR), or emit warning. Maybe @pablogsal can point to the right direction.