Re: Idea for new Display Manager

[Andrew Johnson <anj@aps.anl.gov>]

Hi Whit,

> That it is not a nightmare is one feature. MEDM is pretty great 
> considering how old it is, and it easy to compile/install.

However it is starting to reach the stage where the underlying
technologies are breaking or being obsoleted. Recent maintenance has
been done by Robert Soliday here at APS who has had to make several
changes to get it to work properly with the version of the Gnome desktop
that comes with RHEL 7. Newer window managers just don't respond in the
same way the older ones did, but with the original MEDM authors retured
and few developers left with any experience of Motif and X11 we may not
be able to make these kinds of fixes for too much longer.

OpenMotif is tied to X11, and since Wayland is replacing X as the future
fundamental API for GUI interfacing on Linux the number of layers
between the application code and the display will be going up with the
commensurate chance of introducing more problems.

> One thing that would be nice is if the "adl" format was something 
> standard like json. I have not looked at the MEDM code yet.

The structure of an adl file is pretty flat. The only relationship
between say a text input widget, the label describing it and a box
surrounding them both is the spatial one, there's nothing in the file to
indicated that they are connected at all. Some of the newer display
managers provide hierarchy, so both the input widget and the label would
be children of the box, and when you move the box the others would come
along with it (MEDM provides grouping, but that's not quite the same
thing). You can usually flatten hierarchical things much more easily
than creating structure out of a flat list of objects.

However it has taken us several generations of display managers to get
this kind of insight. Anyone who starts to develop YADM from scratch may
be at risk of not realizing these kinds of issues. Pete Jemian's list of
criteria has lots of good points in it.

> I have been looking for an excuse to play around with Dear ImGui. It 
> turns out that a display manager is an ideal use for it. ImGui is 
> written in c++ (actually it is written like c code). I don't care what 
> language is used but it is probably best to do everything in c++ for 
> speed purposes. There are a lot of wrappers for Dear ImGui. 

How long do you think Dear ImGui will be actively maintained for? When
we're developing instrumentation for an experiment that may have a
25-year or longer lifetime, we have to be careful to avoid depending on
software that has a low bus factor. If you don't get too much interest
from the established labs, that's partly why, we can't afford to jump on
all the latest cool technologies even if they are FLOSS.

Good luck though!

- Andrew

-- 
Arguing for surveillance because you have nothing to hide is no
different than making the claim, "I don't care about freedom of
speech because I have nothing to say." -- Edward Snowdon