Hi Clemens,

> As per the comment above the affected code, the client can specify the face
> when prefix and suffix are provided. The prefix is already checked earlier
> and what remained was checking the suffix not the prefix.

Shouldn't then this code with font-lock-prepend-text-property
be removed completely?  Since both prefix and suffix are non-nil,
this makes code no-op.

> There is another thing I would like to bring up in this context: When the
> annotations returned by annotation/affixation functions already specify
> a face I think it would be nicer if the completion-annotations face
> wouldn't be applied generally. In Selectrum we use something like:
>
> (if (text-property-not-all 0 (length str) 'face nil str)
>     str
>   (propertize str 'face 'completions-annotations))

So you propose to search for the face text-property in the provided string
to decide whether to add the default face in completion--insert-strings?

> This gives the client full control over the visual appearance if that is
> preferred. Maybe this approach could also make sense to be included in
> Emacs?

Do you see any possible backward-compatibility issues with changing this in
Emacs?  For example, when a package like Selectrum puts another face
on the completion string, then it will be displayed instead of the default
completion-annotations face.

> Currently for the annotation function the completions-annotations
> face is always applied and for the affixation function it also still gets
> applied when the affixation function returns a two candidate list (like
> read-extended-command--affixation on current master).  The case of also
> allowing a two candidate list to be returned by affixation functions is
> also currently undocumented.

Thanks for noticing the documentation problem.  Do you think
this fix is sufficient: