In the new REPL, multiline input execution should fail fast

[bswck]

Bug report

Bug description:

To reproduce, simply paste this to the new REPL:

raise Exception
print("But I'm powerful enough to bypass it")

Make sure what you are seeing is:

>>> raise Exception
... print("But I'm powerful enough to bypass it")

Once hit enter, the new REPL executes both statements raise Exception and print("But I'm powerful enough to bypass it") from the same, multi-line input block, despite the failure of the first one:

>>> raise Exception
... print("But I'm powerful enough to bypass it")
... 
Traceback (most recent call last):
  File "<python-input-0>", line 1, in <module>
    raise Exception
Exception
But I'm powerful enough to bypass it

This can get especially confusing if a statement defines a variable. Like

>>> n = 1 / 0
... print(k-n)

causing

Traceback (most recent call last):
  File "<python-input-0>", line 1, in <module>
    n = 1 / 0
        ~~^~~
ZeroDivisionError: division by zero
Traceback (most recent call last):
  File "<python-input-0>", line 2, in <module>
    print(k-n)
          ^
NameError: name 'k' is not defined

I'd expect different behavior: since those two statements belong to the same input, it is more logical for them to behave like a suite, where the consecutive execution of a statement depends on a successful completion of the one before it.

CPython versions tested on:

3.13, 3.14

Operating systems tested on:

Linux, macOS, Windows

Linked PRs

• gh-128231: Use runcode() return value for failing early #129488
• [3.13] gh-128231: Use runcode() return value for failing early (GH-129488) #130513
• [3.13] Revert "gh-128231: Use runcode() return value for failing early (GH-129488) (#130513)" #131834
• gh-128231: Use runcode() return value for failing early (fix test) #131835

[bswck]

Making a PR for the fix.

[Eclips4]

Hi! Thanks for the report.
From my perspective as a someone who using REPL everyday, I'm suspect that it would stop executing after first exception.
Let's imagine the next input:

x = foo(arg)  # for some reason, foo raises an exception
y = x * 10  # Would fail with NameError

I'm think that usually lines of code in code blocks are related and depends on each other, so I'm in favor of changing current behavior to "fail on first exception".