Errors are printed to stdout in pdb module

[frostming]

Bug report

Bug description:

As title, most errors in pdb.py are printed to stdout instead of stderr, like

cpython/Lib/pdb.py

Lines 188 to 193 in 8288f36

	if not os.path.exists(self._target):
	print(f'Error: {target} does not exist')
	sys.exit(1)
	if os.path.isdir(self._target):
	print(f'Error: {target} is a directory')
	sys.exit(1)

This is not following the conventions.

CPython versions tested on:

CPython main branch

Operating systems tested on:

No response

[picnixz]

Note that the Pdb class uses stdout for both the errors and the regular messages. It might be a design choice about not having a separate stderr though I'm not sure about this. Changing this may break scripts that rely on this (which is why I categorized it as a feature instead).

cc @gaogaotiantian

[frostming]

@picnixz Good to know. But there are also parser.error() calls that prints to stderr. It just makes it messy.

[picnixz]

FTR, the message() and error() methods were introduced in 0d08962 as part of the fix for #43477. Historically, errors were explicitly printed to self.stdout, because Pdb inherits the Cmd class which only has a stdout and no stderr.

[frostming]

FTR, the message() and error() methods were introduced in 0d08962 as part of the fix for #43477. Historically, errors were explicitly printed to self.stdout, because Pdb inherits the Cmd class which only has a stdout and no stderr.

I have no objections on this. There are also errors printed outside the Pdb class, like

cpython/Lib/pdb.py

Lines 3575 to 3582 in 8288f36

	opts = parser.parse_args()
	if opts.module:
	parser.error("argument -m: not allowed with argument --pid")
	try:
	attach(opts.pid, opts.commands)
	except PermissionError as e:
	exit_with_permission_help_text()
	return

[picnixz]

What I meant is that if you want to change where the errors are sent, you probably need to change the Cmd base class where you could specify a stderr as well, or the Pdb base class to indicate an additional stderr. We can't just change where it will be printed because of backward compatibility. So, yes, it could be "messy", but not that much. PDB errors are consistently sent to wherever they are expected to be (that is, stdout, for now). Whether the CLI send them elsewhere or not is another matter IMO.

[gaogaotiantian]

Hmm, I don't see this as something we should fix. I kind of agree that it could be better if we design it from scratch. However, it's not too bad. I think my argument is that the error you mentioned is from the pdb command line tool, instead of the core debugger. So these are all checks before we actually run the debugger. I don't believe we have a very strict rule about where the explanations should go in cmdline tools in stdlib. I agree with @picnixz that changing this has little benifits (if any), but may break someone's code because they might have some tests relying on it.

[picnixz]

Note: if someone is worried about stdout/stderr distinction, they can always subclass Pdb and override error(). This method was introduced for that specific reason.

[picnixz]

Considering Tian's stance, I'm going to close this one as not planned. Users that are worried about this should subclass Pdb instead. Alternatively, we could add an stderr attribute to cmd which defaults to stderr and make pdb use stdout=stderr by default and advise users to set stderr to whatever they want for errors instead. However I'm not sure there is a real need for that, and thus I would advise opening a thread on https://discuss.python.org/c/ideas/6 first.