fix virtual relation detection: erschema is no more a relation type but a relation def in _synchronize_permissions

necessary to fix some security test, unclear why did I set perms below at some point...

import cleanup, check data file exists

should turn password into a Binary properly

should not simplify previous to preprocess

use ._cw instead of .req (reintroduced by merge of stable)

oops

backport stable

more extra_args cleanup

only give extra_args if there is something to be fed (breaks some db connectors otherwise)

merge

SqlServer: support single sign on / Windows credential authentication This is done through a new db-extra-arguments section in instance_dir/sources which is passed as a string to the extra_args named argument of lgc.db.get_connection. If this argument is Trusted_Connection, windows authentication is used instead of login/password. This change requires at least revision 5475ec3f6412 of logilab.common which will be included in logilab-common 0.46. debian/control updated to mention this new dependency.

SQL Server port: convert Password to Binary This conversion is required because pyodbc uses buffers for Binary fields and won't be able to cast a string/unicode value to the db VARBINARY column type otherwise.

[javascript] cleanup, remove tabs

[javascript] remove ajaxHtmlHead from dom response once it has been processed. This fixes #549138: treeview: folding is broken The new, systematic, pageid management (introduced by rev 37b376bb4088) made the treeview return : <ul class="treeview">...</ul> <div class="ajaxHtmlHead">...</div> which was then processed by getDomFromResponse and wrapped into a single div node : <div> <ul class="treeview">...</ul> <div class="ajaxHtmlHead">...</div> </div> In the treeview case, the node inserted into the tree was the wrapping <div> instead of the <ul>, causing the folding bug mentioned in the ticket.

msg isn't defined there, but we've to give traceback information to record error

all web tests OK

make new editcontroller works, based on a _cw_edited_fields hidden input

crecord introduced junk...

strip string by default

nicer StringField __init__

fix render (formerly form_render) prototype

process_posted yield field instead of field's name

proper field's role handling: may be 'subject' / 'object' *in case of entity's relation*, but None in other case. So you should correctly specify it to field_by_name

form_add_hidden -> add_hidden form_needs_multipart -> needs_multipart

substitute _render_attrs by values_and_attributes method, keeping bw compat

replace form_field_[encoding|vocabulary] methods on form by encoding|vocabylary(form) methods on fields

refactor vocabulary handling to avoid having to define methods on form objects, use .choices instead even for non final relations. Details: * deprecated support of <role>_<relation>_vocabulary methods on forms * new utility functions for relations vocabulary * most logic now on the RelationField class

cleanup

refactor form field value handling, to get a nicer api and an easier algorithm to get field's value Details: * new .typed_value / .display_value on fields * droped form_field_value on form * .value attribute of field instead of .initial * nicer field's __init__, allowing to give a lambda as value's value