Sylvain Thénault <sylvain.thenault@logilab.fr> [Wed, 18 Jan 2017 15:04:27 +0100] rev 11904
[migration] Fix addition of entity type including boundary constraints on its own attributes
This was failing because of the sequence of sql executing when adding an
attribute:
1. add entity type <commit>
2. add relation type for attribute 1 <commit>
3. add attribute 1 and associated constraints <commit>
etc.
In the case of e.g. start/end constraint, we were trying to add the constraint
before addition of the constrained attribute (e.g. add constraint on 'start'
referencing 'end', but only 'start' has been added yet, not 'end').
This patch fix this by:
* adding the relation type to the schema without having to commit, but keeping
the operation to revert the addition if necessary - this allows to a single
commit for all attributes of the entity type ;
* using a LateOperation on constraint operation, so we ensure attributes are
actually added before any constraint is added.
Sylvain Thénault <sylvain.thenault@logilab.fr> [Thu, 19 Jan 2017 09:53:31 +0100] rev 11903
[schema sync] Refactor AfterAddCWRTypeHook and AfterAddCWComputedRTypeHook so the latter inherit from the former
and enhance their docstring.
Sylvain Thénault <sylvain.thenault@logilab.fr> [Wed, 18 Jan 2017 12:36:02 +0100] rev 11902
Drop embed-allowed option, gone away for a long time
Sylvain Thénault <sylvain.thenault@logilab.fr> [Wed, 04 Jan 2017 10:02:05 +0100] rev 11901
Never record undo information for session or data import
for session, it may causes unlimited database size expansion while it makes no sense
to undo this kind of changes. For data import, they are created programatically hence
we may not want to undo this either.
Philippe Pepiot <philippe.pepiot@logilab.fr> [Thu, 19 Jan 2017 14:53:05 +0100] rev 11900
[cwvreg] load registry using modules names instead of directories
Introspect cubicweb, cubes and apphome using pkgutil to generate the full list
of modules names for loading registries.
Avoiding using bogus logilab.common.modutils.modpath_from_file().
Philippe Pepiot <philippe.pepiot@logilab.fr> [Thu, 19 Jan 2017 15:27:39 +0100] rev 11899
[schema] load schema from modules names instead of directories
Introspect cubicweb, cubes and apphome using pkgutil to generate the full list
of modules names for loading the schema.
Keep historical behavior and check if source .py file exists if a module is
found using python bytecode file (.pyc and .pyo)
Loading schema from apphome require apphome to be present in sys.path and that
"schema" module resolve to a file located in apphome.
Update migraction tests to explicitely update sys.path when loading schema from
different apps, use a contextmanager for this so it's more readable.
Require updated logilab-common and yams
Denis Laxalde <denis.laxalde@logilab.fr> [Wed, 18 Jan 2017 17:16:00 +0100] rev 11898
[devtools/test] Skip qunit tests in case of timeout
Follow-up on 87443f279b0f where other similar tests got
adjusted not to fail (in particular in our CI environment).
Sylvain Thénault <sylvain.thenault@logilab.fr> [Thu, 18 Feb 2016 14:22:07 +0100] rev 11897
[autoform] Avoid two calls to field.process_form for the same field in some cases
when some entity is being created and data include non-inlined relations, the
values for this relation are now stored for later usage, avoiding two calls
to field's process_form method, which may be unexpected for custom fields.
This has been discovered in saem_ref#f5444b1f9770.
Reorganize imports in the test along the way.
Sylvain Thénault <sylvain.thenault@logilab.fr> [Mon, 19 Dec 2016 17:22:48 +0100] rev 11896
[pyramid] Don't use unsafe_cnx_context_manager for write queries
we may have clumsy error that hide previous problems on attempting to commit the transaction.
Closes #16753531
Sylvain Thénault <sylvain.thenault@logilab.fr> [Wed, 14 Dec 2016 08:37:11 +0100] rev 11895
Repair database wrt indexes / unique constraints
* recreate indexes dropped by 3.23 migration (but it's still unclear why)
* attempt drop remaining extra indexes (there may be a bunch of these on old instances)
* warn about missing expected indexes
Closes #16666137
Sylvain Thénault <sylvain.thenault@logilab.fr> [Wed, 14 Dec 2016 17:06:24 +0100] rev 11894
Enhance postgres index filter with values from real life
This has been found on fairly old instances of ours (cwo/elo). Dunno if we
really want this in, it may lead to false negatives.
Sylvain Thénault <sylvain.thenault@logilab.fr> [Wed, 14 Dec 2016 17:04:29 +0100] rev 11893
Extract functions to list expected/found database indexes from check_indexes
so they may be used from other contexts.
Use consistent naming along the way: always use indexes, not indices.
Sylvain Thénault <sylvain.thenault@logilab.fr> [Thu, 24 Nov 2016 15:36:26 +0100] rev 11892
Simplify and fix _cw.drop_entity_cache
* it's never called with an eid as argument, beside in a useless case in test
(removed)
* the only place where it's called from outside the tests is in full-text
reindexation in server.checkintegrity: we could removed the request
implementation and move it in unittest_rset, byt I decided to keep it for
consistency with all other entity cache handling methods
* get back a fix from Julien Cristau for the connection's implementation,
quoting is commit message:
When removing an entity from the transaction's cache, clear the
entity's own cache
May avoid issues where an entity object is still accessible somewhere else
(e.g. an operation) after dropping it from the transaction's cache, with a
stale attribute or relation cache.
Sylvain Thénault <sylvain.thenault@logilab.fr> [Fri, 18 Nov 2016 17:50:56 +0100] rev 11891
[entity/optimization] Cache rset when entity.related is called with entities=False
If fail to see why we couldn't cache in this case, while this is important for
optimization reason: when doing a single HTTP request, some queries are done two
or three times because of predicates / uicfg or alike.
Also always store tuple and not list in the cache, because:
* else we get some regression
* and inconsistent result type (tuple or list)
* and it simply feels better to cache an unmutable object.