Test propagation (issue #266)

[wolfv]

I added a test for the eager propagation issue.

Not sure how to resolve this properly yet!

[baszalmstra]

Do you mind if I close this one @wolfv ?

[baszalmstra]

bump @wolfv

[wolfv]

Should we try to add this test to resolvo now? I think it's still a valid concern.

Basically the problem is that the algorithm:

• first selects libgfortran 5 BN 32
• then selects libgfortran5 12.3.0, but the constraint does not work with the selected build number of libgfortran
• backtracks, and selects older libgfortran 5 BN 0 and then
• selects latest libgfortran5 12.3.0

While it (should?) have done:

• first select libgfortran 5 BN 32
• figure out that constraint on libgfortran5 12.3.0 does not work
• instead select libgfortran5 12.2.0 where constraint works

[wolfv]

I think this is in fact the same behavior that is the problem in prefix-dev/rip#191. We might want to double-check why libsolv backtracks differently here.

[wolfv]

Hmm, actually this might also be the "inverse" problem where the solver backtracks too early :D

libgfortran = 5.0.0 32
    libgfortran5>= 11.3
libgfortran = 5.0.0 0
    libgfortran5>=11.3

libgfortran5 = 12.3.0
    -> libgfortran = 5.0.0 0
libgfortran5 = 12.2.0
    -> libgfortran = 5.0.0 32

We first select

libgfortran = 5.0.0 32, then find out that the latest libgfortran5 = 12.3.0 is not compatible, backtrack, select an older libgfortran (instead of trying an older libgfortran5).

[baszalmstra]

@wolfv This issue has become quite stale, do you mind if we close it?

[wolfv]

I think it's something I might like to come back to. But we can also close it and not delete the branch

[baszalmstra]

We will take a look at this later. Closing for now.