erp5_site_global_id was from the tiolive time, when we were using ERP5
with mariadb and memcached instances shared by many zope instances, this
was used as a way to implement namespaces in memcached, to prevent
conflicts such as two different zope using the same cache keys.

Nowadays, we no longer share memcached, each ERP5 instance has its own
memcached and this prefixing is no longer needed, but there is still one
exception, when we run test using runTestSuite with --node_quantity
higher than 1, we have multiple running test instances sharing the same
mariadb and the same memcached. In that case, each test instance
uses a different mariadb database, but the risk of conflict remain for
memcached.

To solve this, we use derive a prefix from mariadb connection string
and use it in memcached.

This was more or less what the current implementation was trying to do,
but we were setting erp5_site_global_id on the wrong object, so it was
not working since 4889d523 (Define property's default value at class
level., 2012-12-31). Also, this was only done at the level of SharedDict,
but some APIs such as CacheTool or SessionTool use MemcachedDict
directly.

Also, there was a problem that we did not notice yet, because the
previous implementation was doing nothing: memcached keys are limited in
size, to overcome this, we don't use the full connection string in
base64 (which is around 70 characters), but a short hash of the
connection string.

String value raised with `ValueError` had an extra `%s` at the end, hence
raising another error when formatting:

> TypeError: not enough arguments for format string

This commit simply removes the extra formatting character, since I could not
guess what should have been printed here.

The ZSQLMethod was using a wrong table `movement` instead of the `stock` table,
hence throwing errors. This commit makes it use the right table, by adding a
filter on `node_uid`, taking into account the structure of the new table.

/reviewed-by @jerome
/reviewed-on nexedi/erp5!2030

See merge request nexedi/erp5!2029

See merge request nexedi/erp5!2028

 - follow Zope4 ZMI style guide
 - remove the confusing "Upgrade" button and the associated method
 `manage_refresh` doing nothing
 - use `form-control` and `code` CSS classes for better looking fields

When refresh_token is still valid google's endpoint does not include the
current refresh token in the response when refreshing the token, we need
to keep the current one.
This fixes user logout every one hour.

So that version view works as expected, see b415abbb (dms: add version
view on notification messages, 2024-06-19)

See merge request nexedi/erp5!2011

Fix bugs:
- Fix an acquisition context bug: the user found here would be wrapped in
  the acquisition context of self, and as a result SecurityManager.validate
  may consider the user to be outside of the acquisition path of the
  document being checked (ex: when accessing a module while publishing a
  web section).
- While unusual, there may be multiple users matching a given request,
  which is handled by ZPublisher but was skipped here.

Also:
Document:
- Why this method is needed.
- assumptions made to get simpler code.
Improve performance:
- portal_membership._huntUser looks the user up twice, which is expensive.
  Stop using this method.
- When the request is a fake request (from restrictedTraverse) nothing can
  nor should be done, so bypass the entire logic that case.
- Assorted tiny improvements: do not retrieve security manager twice, avoid
  extraneous local assignments, ...
Improve coding style:
- Stop accessing portal_membership's underware.
- Stop accessing PluggableAuthenticationService's underware.
- Simplify disabled cache support: this is exceedingly rare, optimise for
  when it is enabled.
- Do not hardcode log level, also increase the severity: this really is a
  warning.
- Do not try to decode Basic-auth, this is the job of the user folder.
  This removes duplicated code.

Set table:number-columns-spanned to each cell outside listbox/matrixbox.

  Like it is done for option, remove default from schema after append the value into decription.

See merge request nexedi/erp5!2004

Limit printed decimals in flight log file for readability.

See merge request !2017

By pre-fetching some catalog entries to retrieve objects UIDs that can directly
be used in the inventory query, to avoid needless (costly) joins and hinting
the catalog to use the resource_section_node_uid index of the stock table.

See merge request !2012

See merge request !2016

The test from 6316d9bb (Formulator: test form serialization with non
ascii elements, 2024-10-24) revealed that form with an encoding other
than UTF-8 are not supported.

DB.query accepts either bytes or str on python3. We need to decode the
query here to display it in traceback and we must not fail when we
have bytes with non UTF-8 characters, which can happen with binary
columns

It currently not used and it's better not to have such public method

ref: 906761ed (erp5_web: change one-day-max policy max-age to 14400s, 2021-09-01)