Bug#1064016: fixed in adwaita-icon-theme 46~beta-2

[Simon McVittie]

On Tue, 20 Feb 2024 at 21:05:55 +0100, Amy Kos wrote:
> latest change in 46~beta-4
>  "Use a more minimal set of aliases as requested by upstream"
> is understandable from Gnome perpective.

I did try to get a more comprehensive set of aliases merged, but
the upstream maintainer refused, and I was concerned that if I pushed
harder, he would have rejected the change entirely (leaving e.g. Firefox,
Chromium, SDL in a similarly bad situation).

The policy we ended up with is that where the CSS/freedesktop.org
cursors had an obvious equivalence with one of the XC_ constants in
<X11/cursorfont.h>, we added aliases for those, but we did not add any
aliases for the hashes of custom cursors, or for other non-standardized
names (some of which might have originated in Qt, but the nature of
non-standardized names is that it's hard to tell where they came from
or who might be using them).

> Nevertheless it will take time until other software catch up.
> For instance in LXQt at least these are missing:
>  progress, help, nwse-resize and nesw-resize

Those are some of the standardized (CSS/freedesktop.org) names, so I
assume you mean that LXQt is looking for legacy/non-standardized names
that should be equivalent to those. What names is it looking for?

> Would it be worthwile to provide the missing symlinks
> from 46~beta-3 in Debian for the time being?

I would prefer not to have too many aliases that have been specifically
rejected upstream: if we do that, it will encourage projects whose
upstream maintainers happen to use Debian/Ubuntu to keep using the old
names, and then be surprised when their code doesn't work as intended
on non-Debian-derived distributions.

If these cursors are particularly high-visibility, we could consider
adding back a small number of aliases, but I'd prefer to have a
bug report against the package that needs them (in this case LXQt)
before we do that. That bug report's steps to reproduce will show us
how high-visibility the other package's use of those cursors is, and
watching to see when those bugs are closed will give us a cutoff point
at which we can drop the legacy aliases.

In particular, I don't want to keep the hash-based aliases as a
downstream divergence unless they're absolutely necessary, because
those are *extremely* opaque. (We can at least guess at the intended
semantics of left_ptr_watch, but what are the intended semantics of
1081e37283d90000800003c07f3ef6bf? Nobody knows, unless someone can dig
out the original X11 bitmap that hashes to that value!)

So perhaps you could open a bug report against LXQt as a starting point?
That would consist of a list of some scenarios in which a wrong cursor
is seen, and ideally some suggestions for the CSS/freedesktop.org name
that it should be using in those scenarios. Some useful references:
https://drafts.csswg.org/css-ui/#cursor
https://developer.mozilla.org/en-US/docs/Web/CSS/cursor
https://www.freedesktop.org/wiki/Specifications/cursor-spec/

Or, if you can identify the lower-level code that is actually selecting
these cursors (perhaps Qt?), a bug report against the library that would
need to be updated to switch to CSS/freedesktop.org names would also
be fine.

All of the packages I've looked at so far (Firefox, Chromium, SDL) seemed
to be using either CSS/freedesktop.org names, or <X11/cursorfont.h> names
(which now have aliases set up, where available), or a mixture of the two:
they were only using other legacy names as fallbacks, if at all.

    smcv