True, defaults that are generated during an INSERT
or UPDATE operation are post-fetched immediately,
instead of being deferred until later. This mimics
the old 0.3 behavior.
to a collection-based attribute which already had pending changes
would generate incorrect history [ticket:922]
- fixed delete-orphan cascade bug whereby setting the same
object twice to a scalar attribute could log it as an orphan
[ticket:925]
- generative select.order_by(None) / group_by(None) was not managing to
reset order by/group by criterion, fixed [ticket:924]
reconnect after its underlying connection is invalidated, without
needing to connect() again from the engine. This allows an ORM session
bound to a single Connection to not need a reconnect.
Open transactions on the Connection must be rolled back after an invalidation
of the underlying connection else an error is raised. Also fixed
bug where disconnect detect was not being called for cursor(), rollback(),
or commit().
exactly one expression in its columns clause.
- added "helper exception" to select.type access, generic functions raise
the chance of this happening
- a slight behavioral change to attributes is, del'ing an attribute
does *not* cause the lazyloader of that attribute to fire off again;
the "del" makes the effective value of the attribute "None". To
re-trigger the "loader" for an attribute, use
session.expire(instance, [attrname]).
- fix ormtutorial for IS NULL
in this case when mapped against a select statement [ticket:904]
- _hide_froms logic in expression totally localized to Join class, including search through previous clone sources
- removed "stop_on" from main visitors, not used
- "stop_on" in AbstractClauseProcessor part of constructor, ClauseAdapter sets it up based on given clause
- fixes to is_derived_from() to take previous clone sources into account, Alias takes self + cloned sources into account. this is ultimately what the #904 bug was.
the tree can now run the full suite of tests out of the box.
- Migrated most @supported to @fails_on, @fails_on_everything_but, or (last
resort) @unsupported. @fails_on revealed a slew of bogus test skippage,
which was corrected.
- Added @fails_on_everything_but. Yes, the first usage *was*
"fails_on_everything_but('postgres')". How did you guess!
- Migrated @supported in dialect/* to the new test-class attribute __only_on__.
- Test classes can also have __unsupported_on__ and __excluded_on__.
- trying to refine some of the adaptation stuff
- query.from_statement() wont allow further generative criterion
- added a warning to columncollection when selectable is formed with
conflicting columns (only in the col export phase)
- some method rearrangement on schema/columncollection....
- property conflicting relation warning doesnt raise for concrete
the given argument; the previous behavior of constructing a list
of FROM clauses was generally not useful as is required
filter() calls to create join criterion, and new tables introduced
within filter() already add themselves to the FROM clause. The
new behavior allows not just joins from the main table, but select
statements as well. Filter criterion, order bys, eager load
clauses will be "aliased" against the given statement.
process_result_value(), which automatically take advantage of the processing
of the underlying type. Ideal for using with Unicode or Pickletype.
TypeDecorator should now be the primary way to augment the behavior of any
existing type including other TypeDecorator subclasses such as PickleType.
- topological.py cleaned up, presents three public facing functions which
return list/tuple based structures, without exposing any internals. only
the third function returns the "hierarchical" structure. when results
include "cycles" or "child" items, 2- or 3- tuples are used to represent
results.
- unitofwork uses InstanceState almost exclusively now. new and deleted lists
are now dicts which ref the actual object to provide a strong ref for the
duration that they're in those lists. IdentitySet is only used for the public
facing versions of "new" and "deleted".
- unitofwork topological sort no longer uses the "hierarchical" version of the sort
for the base sort, only for the "per-object" secondary sort where it still
helps to group non-dependent operations together and provides expected insert
order. the default sort deals with UOWTasks in a straight list and is greatly
simplified. Tests all pass but need to see if svilen's stuff still works,
one block of code in _sort_cyclical_dependencies() seems to not be needed anywhere
but i definitely put it there for a reason at some point; if not hopefully we
can derive more test coverage from that.
- the UOWEventHandler is only applied to object-storing attributes, not
scalar (i.e. column-based) ones. cuts out a ton of overhead when setting
non-object based attributes.
- InstanceState also used throughout the flush process, i.e. dependency.py,
mapper.save_obj()/delete_obj(), sync.execute() all expect InstanceState objects
in most cases now.
- mapper/property cascade_iterator() takes InstanceState as its argument,
but still returns lists of object instances so that they are not dereferenced.
- a few tricks needed when dealing with InstanceState, i.e. when loading a list
of items that are possibly fresh from the DB, you *have* to get the actual objects
into a strong-referencing datastructure else they fall out of scope immediately.
dependency.py caches lists of dependent objects which it loads now (i.e. history
collections).
- AttributeHistory is gone, replaced by a function that returns a 3-tuple of
added, unchanged, deleted. these collections still reference the object
instances directly for the strong-referencing reasons mentiontioned, but
it uses less IdentitySet logic to generate.
"unique identifier" mechanisms as everything else. This doesn't affect
user code, except any code that might have been hardcoded against the generated
names. Generated bind params now have the form "<paramname>_<num>",
whereas before only the second bind of the same name would have this form.
- bindparam() objects themselves can be used as keys for execute(), i.e.
statement.execute({bind1:'foo', bind2:'bar'})
- changed the various "literal" generation functions to use an anonymous
bind parameter. not much changes here except their labels now look
like ":param_1", ":param_2" instead of ":literal"
- from_obj keyword argument to select() can be a scalar or a list.
When convert_unicode=True, this flag also defaults to `True`, and results in all
unicode conversion operations raising an exception when a non-unicode bytestring
is passed as a bind parameter. It is strongly advised that all unicode-aware
applications make proper use of Python unicode objects (i.e. u'hello' and
not 'hello').
e.g. select([x* 5]) produces "SELECT x * 5 AS anon_1".
This allows the labelname to be present in the cursor.description
which can then be appropriately matched to result-column processing
rules. (we can't reliably use positional tracking for result-column
matches since text() expressions may represent multiple columns).
- operator overloading is now controlled by TypeEngine objects - the
one built-in operator overload so far is String types overloading
'+' to be the string concatenation operator.
User-defined types can also define their own operator overloading
by overriding the adapt_operator(self, op) method.
- untyped bind parameters on the right side of a binary expression
will be assigned the type of the left side of the operation, to better
enable the appropriate bind parameter processing to take effect
[ticket:819]
then re-used. the FROM calculation of a Select normalizes the list of hide_froms against all
previous incarnations of each FROM clause, using a tag attached from cloned clause to
previous.
when its an alias of a Table; added some test coverage for one particular
case from the doctests
- fixed "having" example in doctests, updated eager load example
- locate_all_froms() is expensive; added an attribute-level cache for it
- put a huge warning on all select.append_XXX() methods stating that derived collections like locate_all_froms() may become invalid if
already initialized
selectable expressions. On the SQL side this is represented by the
"corresponding_column()" method. This method is used heavily by the ORM
to "adapt" elements of an expression to similar, aliased expressions,
as well as to target result set columns originally bound to a
table or selectable to an aliased, "corresponding" expression. The new
rewrite features completely consistent and accurate behavior.
- the "orig_set" and "distance" elements as well as all associated
fanfare are gone (hooray !)
- columns now have an optional "proxies" list which is a list of all
columns they are a "proxy" for; only CompoundSelect cols proxy more than one column
(just like before). set operations are used to determine lineage.
- CompoundSelects (i.e. unions) only create one public-facing proxy column per
column name. primary key collections come out with just one column per embedded
PK column.
- made the alias used by eager load limited subquery anonymous.
Jason Kirtland