Memory leak with oneOf in pypy

[funous]

When jsonschema is used with pypy and the schema used contains a oneOf rule, the memory consumption grows indefinitely over time. This only seems to happen if there is a matching subschema, but it's not the last subschema in the oneOf list. This only happens in pypy:

pypy3 --version
Python 3.5.3 (fdd60ed87e94, Apr 24 2018, 06:10:04)
[PyPy 6.0.0 with GCC 6.2.0 20160901]

See the code and json schema below (schema.json):

import json
import jsonschema

with open("schema.json", encoding='utf-8') as schema_file:
    schema = jsonschema.Draft4Validator(json.loads(schema_file.read()))
    obj = {'field': 3}  # matches first subschema - causes a leak
    # obj = {'field': True}  # matches last subschema - does not cause a leak
    for j in range(0, 1000000):
        schema.is_valid(obj)

{
  "$schema": "http://json-schema.org/draft-04/schema#",
  "type": "object",
  "oneOf": [
    {
      "properties": {
        "field": { "type": "number" }
      }
    },
    {
      "properties": {
        "field": { "type": "boolean" }
      }
    }
  ]
}

Running this in pypy quickly shows the memory consumption going up to a few gigabytes.

This might be a pypy bug, but I was not able to re-create this leak without the use of jsonschema.

If it helps, as a workaround I use modified jsonschema with this is_valid function in jsonschema/validators.py:

        def is_valid(self, instance, _schema=None):
            # Fully iterate through the generator to avoid a pypy memory leak
            errors = list(self.iter_errors(instance, _schema))
            return not errors

[Julian]

This seems almost certainly to be a bug in pypy3 (or less likely in pypyX 6, but I can't test pypy2 6 yet because it isn't yet in homebrew...).

But pypy2 5.10.0 settles just fine on 84MB of memory usage here, while pypy3 grows to about 25GB after 2 or 3 minutes, yeah.

[funous]

I've managed to make an example independent on jsonschema and opened up an issue on pypy bug tracker: https://bitbucket.org/pypy/pypy/issues/2833/memory-leak-with-non-exhausted-nested

Although the pypy documentation recommends closing iterators in finally statements manually (which would be here: https://github.com/Julian/jsonschema/blob/4f2eb1f533070300dff6853b176db490b3e14b88/jsonschema/validators.py#L283), it doesn't seem to actually work in this case. Let's see what the feedback on the pypy issue is.

[Julian]

Great! Closing this out but will keep an eye on the upstream ticket.