Since BusinessProcess.py was moved to portal_components, the cache in mixin/composition.py,
which was kept at the level of the zope process, was causing issue when portal_components
was reset : the classes kept in __class_cache would not get resetted, and later calls
to asComposedDocument would reuse them. In consequence, these classes would kept pointers
to objects in the old class BusinessProcess, causing bugs as these objects would have
been cleaned by the reset.

The first intuition was to remove this cache, unfortunately the reason of this introduction
(fixing a memory leak) is still valid nowadays, so I've decided to move this cache within
BusinessProcess, as it has the advantage of emptying it whenever portal_components is resetted.

For more information of the bug mentionned above, and the trials about removing the cache can
be found in the discussion of this MR :
nexedi/erp5!1032

Since commit d2b20acd004441d1cb8d36f33ed7ca112dff0d17 on slapos.git we
are using gcc 8.2 which seem to generate slightly faster code, so we
adjust the min time because tests were failing because we are now too
fast.

erp5_corporate_identity: improve slideshow rendering including displaying legacy presentation as slideshow

ERP5: Remove *Capacity and ImmobilisationMovement Documents never used since their introduction many years ago.

Receiving name & password parameters does not prevent us from also
receiving the authentication cookie. It must not be kept in the request.
And likewise for requests also containing an HTTP authentication header.
And likewise for requests for which CookieCrumbler disables itself on.

The reason why root document is being used as serialization_tag is to cover
document which get recursively reindexed: such indexation activity would
not be able to be validated against all the the serializatio_tag of
documents it will process. But here, fulltext indexations are all spawned
per-document anyway, so we can use such per-document values.

Spawn fulltext indexation activity with the same priority as the
top-priority activity part of current activity group.
And expose the priority of current activity (top-priority for grouped
activities).
Ideally, the priority of each spawned activity should be the top-priority
of all activities member of this activity group for corresponding document.
But there is no obvious way to achieve that through indexation API without
increasing complexity significantly.

plus a few python3 fixes for problem that appeared now that we have more test coverage

See merge request !1128

and a few cleanups in test suite

See merge request !1126

This removes some logs happening a lot during tests, when installing business templates or indexing documents.

See merge request !1105

Updater was using this command to find the latest revision modifying a
path:

  git log -1 -- $PATH

but because of git log's history simplificiation [1] when the head of
the branch is a merge commit, the considered revision was not the merge
commit but its parent.

Use --full-history so that git does not simplify history and return the
merge commit if the latest commit is a merge.

[1]: https://git-scm.com/docs/git-log#_history_simplification

See also https://stackoverflow.com/a/50724301/7607763

instead of randomly truncating the existing file, which sometimes does
not cause an invalid index, create an invalid index from scratch.

I used this blog post
https://mincong.io/2018/04/28/git-index/#2-inside-gitindex for a
description of the index format.

See merge request !1123

These tests are calling runUnitTest twice, first to save then to run
test. On the second run the directory is already created.

Causing the browser time to be used instead of the calculated time of the server, which was incoherent with previous behavior (fix 5c4a03f6)

Display the timezone (if the field is configured like that) in both editable and non-editable mode

See merge request !1113

Using the test class name was causing conflicts on some test classes.
CodingStyleTest was a case, all test use CodingStyleTest class with an
environment variable to control which business template is tested. In
this case, we want one log directory for each business template.

Introduce a getLogDirectoryPath method that's supposed to create and
return the path of the directory to use for log and override this method
for the special case of coding style test.

This test will guarantee that the issue of PyPdf2 overwrite warnings.showwarning method. By default, overwriteWarnings is True in PyPdf2 and we don't want change in many place in ERP5.

Then, this issue should be fixed in PyPdf2 package.

As header h* and link a use the same color, keep the underline to differenciate links

See merge request !1120