Originally asked on pixi #5385
Problem
When adding a new constraint or dependency, it can be very difficult to work out the root cause of unsatisfiability from the current logging.
I was using pixi (0.63.2).
I think that from a user-perspective, highlighting the minimal top-level dependencies that cause the contraditon would be highly valuable.
Example (minimal) conflict (using pixi.toml):
[workspace]
name = "pytorch-fail"
channels = ["https://prefix.dev/conda-forge"]
platforms = ["linux-64"]
[system-requirements]
linux = "4.14"
libc = { family = "glibc", version = "2.28" }
cuda = "12"
[dependencies]
python = ">=3.13"
pytorch-gpu = ">=2.5"
# These two are mutually incompatible (but work individually):
pyarrow = ">=20"
gcc = ">=15.2"
Commenting out either pyarrow or gcc deps gives a solution, but having both gives over 200 lines of output, with no clear indicator as to what the problem is.
There's lots of logging like:
│ │ │ └─ libabseil >=20240722.0,<20240723.0a0, which cannot be installed because there are no viable options:
│ │ │ ├─ libabseil 20240722.0, which conflicts with the versions reported above.
│ │ │ ├─ libabseil 20240722.0, which conflicts with the versions reported above.
but there is no abseil version 'reported above' (so that's rather confusing).
With hindsight of knowing the clashing packages, this is the only hint (but there are many other conflicts reported):
│ │ │ │ │ └─ gcc_impl_linux-64 >=6,<13, which conflicts with any installable versions previously
Bear in mind that my original problem involved the pixi dependencies of a large mono-repo with maybe 200 dependencies in the .toml file - it took quite a while of trial and error to refine this down to a minimal example. Before that, I really had no clue which of my constraints were incompatible.
In whittling down to a minimal contradiction for this request, I realised that knowing the minimal dependencies that result in conflict is hugely valuable for a user. I'm wondering if resolvo has enough information in the solve to be able to indicate which top-level dependencies result in direct contradiction (especially when that contradiction is transitive)?
Once you know a minimal clashing top-level dependency set, you can start to look into possible relaxation of constraints (or decide what you need to rebuild yourself).
BTW - many thanks for bringing us out of the dark days of 30-minute sat solve times...!
Originally asked on pixi #5385
Problem
When adding a new constraint or dependency, it can be very difficult to work out the root cause of unsatisfiability from the current logging.
I was using
pixi(0.63.2).I think that from a user-perspective, highlighting the minimal top-level dependencies that cause the contraditon would be highly valuable.
Example (minimal) conflict (using pixi.toml):
Commenting out either pyarrow or gcc deps gives a solution, but having both gives over 200 lines of output, with no clear indicator as to what the problem is.
There's lots of logging like:
but there is no abseil version 'reported above' (so that's rather confusing).
With hindsight of knowing the clashing packages, this is the only hint (but there are many other conflicts reported):
Bear in mind that my original problem involved the pixi dependencies of a large mono-repo with maybe 200 dependencies in the .toml file - it took quite a while of trial and error to refine this down to a minimal example. Before that, I really had no clue which of my constraints were incompatible.
In whittling down to a minimal contradiction for this request, I realised that knowing the minimal dependencies that result in conflict is hugely valuable for a user. I'm wondering if resolvo has enough information in the solve to be able to indicate which top-level dependencies result in direct contradiction (especially when that contradiction is transitive)?
Once you know a minimal clashing top-level dependency set, you can start to look into possible relaxation of constraints (or decide what you need to rebuild yourself).
BTW - many thanks for bringing us out of the dark days of 30-minute sat solve times...!