See merge request nexedi/erp5!1615

Rationale:

Converting * to data frame / numpy array efficiently is required in all
wendelin projects, without this functionality wendelin is useless.
Currently all projects allow this functionality in an insecure way.
This commit aims to improve the situation by supporting a secure way of
this functionality.

(See nexedi/wendelin!99 (comment 158474))

Because pandas (in restricted Python) can also be useful in 'pure' ERP5
(without Wendelin) the functionality is added to ERP5 source code.

---

Security:

Security is guaranteed by patching selected read_* functions and
allowing the patched versions. The patch prohibits anything but
string input which directly contains the data (e.g. no urls, file
paths). New unit tests ensure the restrictions of the patches
are actually effective.

---

Notes on implementation decisions:

Instead of offering new ERP5 extension methods (e.g. Base_readJson)
this commit adds patched pandas read functions in restricted Python.
In this way the change of the known API is as minimal as possible.

Instead of globally monkey-patching pandas read_* functions, only the
functions inside restricted python are patched.
In this way the fully-functional, original functions are still available
in Zope products or ERP5 extension code.

Minor changes in the way how pandas is allowed in restricted python
have been applied. Please consult the following discussions in the Merge
request for details:

nexedi/erp5!1615 (comment 159203)
nexedi/erp5!1615 (comment 159341)

See merge request nexedi/erp5!1630

...for  getPortalDataConfigurationTypeList.

See nexedi/erp5!1630 (comment 159889).

Translated property accessor accessor accidentally changed to return
on python2 unicode for translated messages and str for non translated
messages in a17bb910 

To make it worse, CMFCategory's Renderer was swallowing exceptions.

This restores translated accessors behavior and changes Renderer to
let exceptions propagate

See merge request nexedi/erp5!1629

partially revert a17bb910 (py2/py3: Make Products code compatible
with both python2 and python3., 2022-04-13) so that
TranslatedPropertyGetter keeps returning utf-8 encoded str (bytes) on
python2 and returns text only on python3.

This regression caused displaying categories to raise unicode error or
to silently not display the labels for.

This just hides issues.

In commit a17bb910 ("py2/py3:
Make Products code compatible with both python2 and python3"),
2to3 changed `_.has_key(...)` to `... in _` whereas _ is not a dict.

Traceback (innermost last):
 ...
 Module OFS.PropertyManager, line 309, in manage_editProperties
   manage_tabs_message=message)
 Module Shared.DC.Scripts.Bindings, line 322, in __call__
   return self._bindAndExec(args, kw, None)
 Module Shared.DC.Scripts.Bindings, line 359, in _bindAndExec
   return self._exec(bound_data, args, kw)
 Module App.special_dtml, line 185, in _exec
   try: result = render_blocks(self._v_blocks, ns)
 Module DocumentTemplate.DT_In, line 707, in renderwob
   try: append(render(section, md))
 Module DocumentTemplate.DT_Let, line 76, in render
   else: d[name]=expr(md)
 Module DocumentTemplate.DT_Util, line 210, in eval
  - __traceback_info__: _
   return eval(code, d)
 Module <string>, line 1, in <module>

TypeError: argument of type 'TemplateDict' is not iterable

See merge request nexedi/erp5!1628

Delete the content of the cache storage (used by erp5js), unregister the service worker and force the browser to reload.

It is now replaced by erp5_web_renderjs_ui

This is for *StandaloneDeploymentScriptTest* test suites, which
download a large disk image during slapos node instance step.

For example, on a Wendelin.StandaloneDeploymentScriptTest-Debian.Stretch
test run [1], we can see that the slapos node instance started at
2022-05-13 07:50:30,788 and stopped trying at 2022-05-13 07:51:35,885.

In the last run, slapos node instance failed, especially this promise:

    2022-05-13 07:51:35,701 INFO     slapgrid_cp: 2022-05-13 07:51:35 slapos[3548917] ERROR ERROR '/srv/slapgrid/slappart22/t/cjl/i/1/srv/monitor/private/virtual-hard-drive-url/virtual-hard-drive-url-processed-config.state' not empty, content available at https://[???]:8026/private/virtual-hard-drive-url/virtual-hard-drive-url-processed-config.state
    ...

and if we access this URL, we can see the error:

    [Errno 2] No such file or directory: '/srv/slapgrid/slappart22/t/cjl/i/1/var/virtual-hard-drive-url/update-image-processed.md5sum'

which means that the image is still not downloaded.

testnode still ignore slapos node instance state return code and run
tests anyway, so the test ran and failed later.

This image is 1.8Go, when I tried to download it locally it took
56.41 seconds, so it must be generally almost OK.

By retrying up to 60 times, we give more time for the slapos node
instance step to download the image.

[1]: https://softinst161166.host.vifib.net/cjl-oSgtg7v20D/suite.log

This reverts commit b404b724.

Redirects does not make any sense with timerserver responses, let's
keep this an error.

To workaround a bug with our old pylint / astroid versions

Revert "py3: sortKey() must be a string as comparison between str and None/int/tuple now raises TypeError."

This reverts commit 192c2000.

This commit introduced a test failure with testCommitOrder
testInvalidationBug.TestInvalidationBug [1]

    Traceback (most recent call last):
      File "Products/ERP5/tests/testInvalidationBug.py", line 93, in testCommitOrder
        self.assertEqual(result_list[1], [0,0])  # activity buffer first
    AssertionError: Lists differ: [1, 0] != [0, 0]

    First differing element 0:
    1
    0

    - [1, 0]
    ?  ^

    + [0, 0]
    ?  ^

That commit was apparently in the good direction, because transaction
API specifies that sortKey must return string

https://github.com/zopefoundation/transaction/blob/6cf10102948a8f4e36092a9b402e3859d6db4ccd/CHANGES.rst#141-2013-02-20
https://github.com/zopefoundation/transaction/blob/33448417ab526327757f05e23b1884786c805629/src/transaction/interfaces.py#L524-L534

using a css selector also matches:

    <div class="transition_message transition_message_success">Deleted.</div>

unlike the xpath selector which matches only element where class
attribute is exactly "transition_message"

Define min and max to integer html field

when defining start and end property in ERP5 integer field

"template_keep_path_list" definition was wrong, the tool is uninstalled
when updating in wrong order

This allows executing a script doing REQUEST.RESPONSE.redirect() without
error.

Before this patch it's an AttributeError similar to:

    ------
    2022-04-12 03:52:49,083 WARNING ActivityTool Could not call method ...
    Traceback (most recent call last):
      ...
      File "Script (Python)", line 34, in Base_redirect
        return request.RESPONSE.redirect(redirect_url, status=status_code)
    AttributeError: TimerResponse instance has no attribute 'redirect'

Instead of temporarily changing the global timeout, which can impact
other parts of the system and which is not free from race conditions (
another thread might have changed the timeout when socket.getdefaulttimeout()
is called and then we restore a wrong default timeout).

See commit 0c2dd107.

See merge request nexedi/erp5!1616

   For facebook and openid buttons.

  Like in WebSite_login, came_from is an URI Template so we must expand the URI to handle the {&n.me}
  that can be present on the URL.

With Python2, properties were all str(). with Python3, we need to distinguish
binary-like properties from text-like properties.

This should ideally be implemented by checking PropertySheet elementary_type
(such as `data` for bytes() and `string` or `text` for str()) for each property.
For now (bootstrap/addERP5Site) though, let's consider all properties in .xml to
be UTF-8 str() and statically define which is which for non-.xml files.

And _mysql/mysqldb API (_mysql.connection.query()) converts the query string to
bytes() (additionally, cursor.execute(QUERY, ARGS) calls query() after
converting everything to bytes() too).