Create the SQLMethod class based on ZSQLMethods.SQL
class and XMLObject. Also, move attributes to
property in 'SQL Method' property sheets.

Also, we don't need to look for `transactionless` connection-id here, as we don't use them in
catalog methods.

And, update the files where we can use the new function.

/cc @aurel 

/reviewed-on nexedi/erp5!411

To generate (and execute) SQL, use catalog tool.

/reviewed-on nexedi/erp5!408

This reverts commit 206fa603 (which was
itself a revert commit), re-applying the change now that surrounding
code is ready for it.

Ignored columns are produced when aliasing a column. For example,
aliasing "catalog.reference" as "reference".
Before this change, this would cause conditions on "reference" to be
rendered non-mapped, which can cause SQL execution issues when there is
more than one "reference" column available (catalog.reference and its
alias counting as only one), which is the case when
catalog-category-catalog joins happen.

Instead, render all columns which could be mapped, independently from
their "ignored" status.

Also, use a different local variable for table aliases than for column
aliases.
Also, use more "return" statements, and simplify conditional structure.

As per Jérome, who implemented the test, it was written to test the
current state rather than testing the desired outcome. And it makes
little sense to have (and test for) 100 being present in both debit and
credit columns ("normal" lines), and 0 to be present in the stat line.

Update test to check for a more consistent outcome.
Acked-by: Jérome Perrin <jerome@nexedi.com>

Explicitely state which values represent empty values. Coercing to boolean is not sufficient.

-  Remove field_json.value because that one is never send by ERP5 backend
-  Set comprehensive initial state and avoid sneaking state variables afterwards
-  Handle better NaNs which represent empty numerical value
-  Refactor for shorter and simpler code
-  Rename "percents" -> "percentage" according to coding style guidelines

otherwise TALES in input_style does not work and changes in the original proxy field will not be reflected.

This change is not the correct one, and not at the correct time. Will be
re-applied when select_dict later looses the ability to strip table in
favour of stricter argument schema.

Fix b6574626

This reverts only one hunk of the original commit, as this part causes a
regression not trivial to fix.

/reviewed-on nexedi/erp5!403

caused by no search criterions

graph_editor was using `rsvp.Monitor` in order to dynamically add new promises to the chain of promise in its `declareService` promise. New DOM elements were added during gadget's lifetime and and event handler on these new elements was added to the chain of promise by using [monitor method](
 https://lab.nexedi.com/nexedi/erp5/blob/76ecef89d0b4f6aa6bc5/bt5/erp5_graph_editor/SkinTemplateItem/portal_skins/erp5_graph_editor/dream_graph_editor/jsplumb/jsplumb.js.js#L607) of a monitor instance which was [returned](https://lab.nexedi.com/nexedi/erp5/blob/76ecef89d0b4f6aa6bc5/bt5/erp5_graph_editor/SkinTemplateItem/portal_skins/erp5_graph_editor/dream_graph_editor/jsplumb/jsplumb.js.js#L835) in `declareService`.

`rsvp.js` included in `erp5_xhml_style` exported `Monitor`, but this was removed in  af9c57db . If I understand correctly, this is now included in renderjs, but it's only internal.

So this old way of doing is not longer possible. We realized that instead of dynamically setting `dblclick` event handlers to each graph node elements, we could  simply rely on event bubbling and use a event handler on the parent DOM element. Also, we used renderjs builtin `onEvent` that makes event callback function executed in the promise chain . 

At this stage we did not try to switch all event handling to this approach of using a "global" event handler on the parent DOM, because the goal here was just repairing the graph editor and making sure we have tests running. Also jsplumb uses its own event system.

To enable tests for this:
 - Running the existing qunit test through Zelenium. As far as I know we cannot run qunit test as part of ERP5 test suite.
 - Install the business template in testXHTML so that it is tested by `jsl`, which by the way produce different messages that the jshint integrated in ERP5's code mirror and jslint from `WebScript_checkSyntax`. For now, this passes jshint and jsl, but jslint complains about some indentation and space problems.

/cc  @romain @vincentB @xiaowu.zhang  @seb @gabriel 

/reviewed-on nexedi/erp5!321