On 20.10.2021 14:25, Boris Kolpackov wrote:
Thorsten Berger <thorsten.berger@xxxxxx> writes:
New UI extensions are made to xconfig with panes and buttons to allow users
to express new desired target options, calculate fixes, and apply any of
found solutions.
[...]
You can see a YouTube video demonstrating this work [2].
While the demo looks impressive, I wonder if you ran into many cases
where the number of solution and/or the number of fixes in a solution
is large (and therefore would be hard for a human to make a decision
about)?
My closest experience with something like this is aptitude and the
few times I tried to use it to solve package dependency issues were
futile because of that (i.e., large number of alternative solutions
and large number of changes in each solution).
Thanks! BTW, I've opened a new thread, since the emails with the main
patches didn't go through before (were too large) -- perhaps you can
switch to that one.
For the user interface, we limit the number of fixes shown to three,
which in our experience is a good compromise. A fix can require changing
multiple symbols. However, it's not too many usually. Take a look at
Fig. 5 in [0], which shows results from our large-scale experiments. For
the probably most common case of asking for changing 1 symbol (conflict
size of 1 on the x-axis) that the user wants to enable/disable, the
median fix size is relatively small. So it should be easy for a user to
understand most of the fixes. However, in many cases it might not even
be necessary to understand the changes, or at least see which symbols
would be changed, then the user can apply the fix.
Note that in Fig. 5 in [0], these statistics are even before Patrick
Franz has done some rewriting of some internals, which lead to better
numbers.
Best, Thorsten
[0] http://www.cse.chalmers.se/~bergert/paper/2021-icseseip-configfix.pdf
