[FIX] composer: fix get/setText of the content editable helper

[rrahir]

The recent use of the property "plaintext-only" came with a change of
behaviour in the composers, specifically in Firefox.

How to reproduce in FF:

• write a single character in a grid composer
• delete the character

-> a new line is inserted

Apparently, when setting the attribute contentEditable to
plaintext-only, deleting the single character does not delete the span
that encapsulates it if the latter has a class attribute. This couples
with a recent refactoring of the content editable helper to handle new
line characters and the empty composer is not detected as such

What happens currently?

Consider the composer content after inputing the letter 'W':

<p>
  <span class="">
    w
  </span>
</p>

The cursor is set just after the letter W.

In a chromium-based browser, hitting the Delete key yeilds the
following result:

<p>
  <span>
    <br>
  </span>
</p>

The letter is replaced by <br> which could be interpreted as a
newline, except that in this situation, we do not want a newline to
appear as we meerly deleted the single character of the composer, so
we do not want new characters.

Also, we introduced some logic to fight this side-effect when we
introduced the support of newlines of pasted content (#6467)

However, it turns out that in firefox, hitting Delete will yield this
result

<p>
  <span class="">
    <br>
  </span>
</p>

The class attribute is not cleared up! And unfortunately, the logic that
counteracts the presence of the unwelcome <br> would test against the
exact form seen in a chromium-based browser, so it did not work on
Firefox.

This commit extends the check to work on both Firefox-based and
Chromium-based browsers.

Note that other leads were investigated and it seems that if we stopped
using paragraphs to represent new lines, this issue with the <br>
doesn't seem to occur. An improvement (and good cleanup) could be
achieved by dropping that paragraph strategy and rely on spans and br
characters.

Task: 5082601

Task: 5082601

Description:

description of this task, what is implemented and why it is implemented that way.

Task: 5082601

review checklist

• feature is organized in plugin, or UI components
• support of duplicate sheet (deep copy)
• in model/core: ranges are Range object, and can be adapted (adaptRanges)
• in model/UI: ranges are strings (to show the user)
• undo-able commands (uses this.history.update)
• multiuser-able commands (has inverse commands and transformations where needed)
• new/updated/removed commands are documented
• exportable in excel
• translations (_t("qmsdf %s", abc))
• unit tested
• clean commented code
• track breaking changes
• doc is rebuild (npm run doc)
• status is correct in Odoo

[robodoo]

[LucasLefevre]

It doesn't behave as expected if I type class="sdf" in a cell (plain text).

Also, it doesn't seem be problematic in this specific case, but as always, playing with innerHTML is dangerous. It doesn't trigger the security check in odoo, but it should if it was stricter

Any easy other way to achieve the same thing ?

[rrahir]

the problem arises with our own way of injecting stuff in the composer html,
theoretically, the other part of the fix (not adding empty classes) should fix the issue at hand but it quite fragile and someone else that works on the content helper (the upcoming dark theme PR might be a good candidate as it injects style AFAIK) might break it again. I will see if we can have a more solid comparison without crushing the performances

[rrahir]

the good solution would be to use innerText all the way but requires some polyfill for the tests, I'm not sure we will reach a "good" point where we don't end up testing the polyfill in the existing tests though

[rrahir]

@LucasLefevre I added a wip commit with an alternative implementation using actual node comparison. I did not notice any performance issue with it and the initial problem seems fixed as well :)

[LucasLefevre]

robodoo r+