commit 3eecbe1f5ce468769d5d884172eeca677e2f1b30 Author: Damian Johnson atagar@torproject.org Date: Wed Jan 25 10:58:50 2017 -0800

Defer defaulting to speed descriptor parsing

Defaulting descriptor attributes when they're accessed rather than doing this right away. This speeds reading descriptors with validation by ~25%...

Before descriptor.microdescriptor... success (15.38s) descriptor.networkstatus... success (23.25s)

After descriptor.microdescriptor... success (11.67s) descriptor.networkstatus... success (19.80s) --- docs/change_log.rst | 2 +- stem/descriptor/__init__.py | 48 ++++++++++++++++++++++----------------------- 2 files changed, 25 insertions(+), 25 deletions(-)

diff --git a/docs/change_log.rst b/docs/change_log.rst index 477db2e..1111afd 100644 --- a/docs/change_log.rst +++ b/docs/change_log.rst @@ -45,7 +45,7 @@ The following are only available within Stem's `git repository

* **Descriptors**

- * Sped descriptor reading by ~5% by not cloning immutable fields + * Sped descriptor reading by ~25% by deferring defaulting when validating * Support for protocol descriptor fields (:spec:`eb4fb3c`) * Shared randomness properties weren't being read in votes (:trac:`21102`)

diff --git a/stem/descriptor/__init__.py b/stem/descriptor/__init__.py index 67ea374..1262f6a 100644 --- a/stem/descriptor/__init__.py +++ b/stem/descriptor/__init__.py @@ -437,6 +437,15 @@ def _parse_key_block(keyword, attribute, expected_block_type, value_attribute = return _parse

+def _copy(default): + if default is None or isinstance(default, (bool, stem.exit_policy.ExitPolicy)): + return default # immutable + elif default in EMPTY_COLLECTION: + return type(default)() # collection construction tad faster than copy + else: + return copy.copy(default) + + class Descriptor(object): """ Common parent for all types of descriptors. @@ -516,21 +525,6 @@ class Descriptor(object): if parser_for_line is None: parser_for_line = self.PARSER_FOR_LINE

- # set defaults - - for attr in self.ATTRIBUTES: - if not hasattr(self, attr): - value = self.ATTRIBUTES[attr][0] - - if value is None or isinstance(value, (bool, stem.exit_policy.ExitPolicy)): - pass # immutable - elif value in EMPTY_COLLECTION: - value = type(value)() # collection construction tad faster than copy - else: - value = copy.copy(value) - - setattr(self, attr, value) - for keyword, values in list(entries.items()): try: if keyword in parser_for_line: @@ -654,19 +648,25 @@ class Descriptor(object): if name in ('_lazy_loading', 'ATTRIBUTES'): return super(Descriptor, self).__getattribute__(name)

- # If attribute isn't already present we might be lazy loading it... + # If an attribute we should have isn't present it means either... + # + # a. we still need to lazy load this + # b. we read the whole descriptor but it wasn't present, so needs the default

- if self._lazy_loading and name in self.ATTRIBUTES: + if name in self.ATTRIBUTES and not hasattr(self, name): default, parsing_function = self.ATTRIBUTES[name]

- try: - parsing_function(self, self._entries) - except (ValueError, KeyError): + if self._lazy_loading: try: - # despite having a validation failure check to see if we set something - return super(Descriptor, self).__getattribute__(name) - except AttributeError: - setattr(self, name, copy.copy(default)) + parsing_function(self, self._entries) + except (ValueError, KeyError): + try: + # despite having a validation failure check to see if we set something + return super(Descriptor, self).__getattribute__(name) + except AttributeError: + setattr(self, name, _copy(default)) + else: + setattr(self, name, _copy(default))

return super(Descriptor, self).__getattribute__(name)