Allow customizing showlocals cutoff

[emil-petersen]

The -l flag - showlocals - for PyTest allows a test to include the state of local variables in case of test failure.

However, it seems there is a hard-coded limit of 70 characters in this output. From diving through the code, I believe this happens in the SafeRepr object in py.

This is an issue in my case because the local variables are dictionaries, and trying to understand a test failure is heavily reliant on the exact content of those variables.

I can log them myself, but doing so defeats the purpose of having -l available.

Therefore, I would like to request the ability to configure the cutoff for showing locals. Thank you.

[ApaDoctor]

What is the best way to make possible to configure the unlimited locals output?
In config or as command line argument?

[nicoddemus]

Today -l is a bool flag, I believe we can change it to be an int flag instead, so users can still pass -l and obtain the default 70 characters cuf-off, or -l 200 to set the cut-off to 200.

[emil-petersen]

@nicoddemus Assuming it could easily be set in pytest.ini, that would be a great solution from my perspective.

[The-Compiler]

Today -l is a bool flag, I believe we can change it to be an int flag instead, so users can still pass -l and obtain the default 70 characters cuf-off, or -l 200 to set the cut-off to 200.

I don't think that'd work - presumably -l would then always eat the following argument and show an error if it isn't an int?

[nicoddemus]

@The-Compiler yeah when I proposed I wasn't sure it would work, I would say it is worth trying.

If that doesn't work we will need a separate variable for that, I suggest -l-cutoff=X which sets the cut-off to X and implies -l.

[ApaDoctor]

Cut-off to 200 is problem. Because cutoff implemented in py module.
And _strrepr allows to change only string length, but not other objects.
So to implement possibility to change cut-off limits we need to change 'py' module first and then provide changes to pytest.
I think great solution is to allow to off cutoff.
And after release of py changes we can make possible to change cut-off size in oytest.

[nicoddemus]

Cut-off to 200 is problem. Because cutoff implemented in py module.

I believe this is implemented in pytest._code:

pytest/_pytest/_code/code.py

Lines 541 to 562 in 966391c

	def repr_locals(self, locals):
	if self.showlocals:
	lines = []
	keys = [loc for loc in locals if loc[0] != "@"]
	keys.sort()
	for name in keys:
	value = locals[name]
	if name == '__builtins__':
	lines.append("__builtins__ = <builtins>")
	else:
	# This formatting could all be handled by the
	# _repr() function, which is only reprlib.Repr in
	# disguise, so is very configurable.
	str_repr = self._saferepr(value)
	# if len(str_repr) < 70 or not isinstance(value,
	# (list, tuple, dict)):
	lines.append("%-10s = %s" % (name, str_repr))
	# else:
	# self._line("%-10s =\\" % (name,))
	# # XXX
	# py.std.pprint.pprint(value, stream=self.excinfowriter)
	return ReprLocals(lines)

Although here the 70 limit is commented out. Hmm this warrants more investigation.

[ApaDoctor]

Yes, but at line 554 we call self._saferepr()
Which using py

    def _saferepr(self, obj):
        return py.io.saferepr(obj)

This is i talking about

[nicoddemus]

@ApaDoctor I see thanks. Indeed py.io.saferepr has some hard-coded values.

[ApaDoctor]

@nicoddemus exactly:)
So letls return to my proposal:))

So to implement possibility to change cut-off limits we need to change 'py' module first and then provide changes to pytest.
I think great solution is to allow to off cutoff.
And after release of py changes we can make possible to change cut-off size in pytest

[nicoddemus]

We might consider just copying the saferepr code and modify it as needed; pylib is in maintenance mode and has to be 2.6 compatible, while in pytest we have no such restriction. The py.io.saferepr module has only 70~ lines after all. @RonnyPfannschmidt what do you think about this approach?

[RonnyPfannschmidt]

as we start to copy more and mnore sublibs of pylib into pytest i start to seriously question that approach

[nicoddemus]

@RonnyPfannschmidt in your opinion should we change pylib's py.io.reprlib then?

[blueyed]

Relevant code: https://github.com/pytest-dev/py/blob/5f1f794f5c5aa25802ea61b4430648438c8d4b93/py/_io/saferepr.py#L59-L71 - it uses maxsize=240 by default.
Python's Repr's has some more limits: https://github.com/python/cpython/blob/20d68dfcc07bd389ce2ea7b0773c44d97ebeb68d/Lib/reprlib.py#L36-L49
It might be hard to transform --showlocals-length into that, but it could be applied just on the result then, and maxing out all other limits.
Something to consider test there then is recursion, i.e. without any limit (which does not seem to be supported by Repr itself, but could be hacked by using a never-decreasing int maybe) it might run into a RecursionError.

Apart from that I would say that with -v or -vv it should not apply limits - similar to other places where you need -vv to get the whole diff with assertion errors.

[blueyed]

This should work with -vv via #3681 by now at least.
See #3962 for references / ideas to have a distinct option to have this not be bound to verbosity.