when loading mapped entities. The function overhead of applying
a per-object deferred callable to an instance at load time was
significantly higher than that of just loading the data from the row
(note that ``defer()`` is meant to reduce DB/network overhead, not
necessarily function call count); the function call overhead is now
less than that of loading data from the column in all cases. There
is also a reduction in the number of "lazy callable" objects created
per load from N (total deferred values in the result) to 1 (total
number of deferred cols).
[ticket:2778]
become an externally usable package but still remains within the main sqlalchemy parent package.
in this system, we use kind of an ugly hack to get the noseplugin imported outside of the
"sqlalchemy" package, while still making it available within sqlalchemy for usage by
third party libraries.
all platforms. let's instead store callcounts for *all* observed platforms in a datafile.
Will try to get enough platforms in the file for jenkins to have meaningful results.
for platforms not in the file, it's just skiptest.
parts of the pstats that are idiosyncratic to different platforms.
the goal is no per-version assertions on tests, version differences
in theory would go into the list of profiling exceptions.
- resolved some serious speed hits I missed, we need to ensure
only deannotated columns are used in the local/remote collections and soforth
so that hash lookups against mapped columns don't dig into __eq__()
- fix some other parity mismatches regarding stuff from [ticket:2453],
including finding another case where _deep_annotate() was doing the wrong thing,
new tests.
- [feature] Major rewrite of relationship()
internals now allow join conditions which
include columns pointing to themselves
within composite foreign keys. A new
API for very specialized primaryjoin conditions
is added, allowing conditions based on
SQL functions, CAST, etc. to be handled
by placing the annotation functions
remote() and foreign() inline within the
expression when necessary. Previous recipes
using the semi-private _local_remote_pairs
approach can be upgraded to this new
approach. [ticket:1401]
against ORM-level column could be treated as
a distinct entity when producing certain
kinds of joined-inh joins. [ticket:2316]
- [bug] related to [ticket:2316], made some
adjustments to the change from [ticket:2261]
regarding the "from" list on a select(). The
_froms collection is no longer memoized, as this
simplifies various use cases and removes the
need for a "warning" if a column is attached
to a table after it was already used in an
expression - the select() construct will now
always produce the correct expression.
There's probably no real-world
performance hit here; select() objects are
almost always made ad-hoc, and systems that
wish to optimize the re-use of a select()
would be using the "compiled_cache" feature.
A hit which would occur when calling select.bind
has been reduced, but the vast majority
of users shouldn't be using "bound metadata"
anyway :).
classes. Note the repr here can't fully support
the "repr is the constructor" idea since schema
items can be very deeply nested/cyclical, have
late initialization of some things, etc.
[ticket:2223]
collections present in given merge field. This shows that the change proposed
in [ticket:2221] adds over 10% method call overhead including significantly more
SQL statements.
- remove extra _load_fixtures() method. using the classbound version as the fixtures
seem to have everything classlevel at the moment (wondering if we want to change that
for non-classlevel setup/teardown)
access to the cls/self.tables/classes registries
- express orm/_base.py ORMTest in terms of engine/_base.py TablesTest,
factor out common steps into TablesTest, remove AltEngineTest as a
separate class. will further consolidate these base classes
these are back to being part of LoaderStrategy
- simplify attribute.get()
- inline the dict get inside of attribute.__get__()
- revert some memoized attrs from InstanceState which are called in almost
all cases regardless
- inlining
just call type._cached_bind_processor(dialect), cache the impl *and* the processor function.
same for result sets.
- use plain dict + update for defaultexecutioncontext.execution_options
just the identity map lookup + expired check, the other
which does the load unconditionally. All the refresh/deferred load
calls use the unconditional load method, query.get() and
LoadLazyAttribute call the identity check by itself first. m2o
lazyloads for object already in the identity map callcounts are
now cut in half, since no Query object is created.
Eric Streeper