BUG: pin_subpackage
does not populate hash correctly in run_constrained
#5572
Labels
type::bug
describes erroneous operation, use severity::* to classify the type
Checklist
What happened?
In the blas-feedstock, as of h-vetinari/blas-feedstock@54921a9, I was running into a weird resolver error:
That is, until I noticed the
h1234567
. The recipe at this point containsand
Obviously these refer to the exact same output in both places (
run:
andrun_constrained
), so the hash divergence is definitely spurious.I presume that this is a bit tricky because the hashes of later packages will only be known once the various dependencies have themselves been resolved (i.e. the hash of
libcblas
depends on knowing the hash oflibblas
in its host environment, making it necessary to populate the metadata forlibblas
first), but that means that either:pin_subpackage
underrun_constrained
gets populated/corrected after all the outputs are knownpin_subpackage
is used underrun_constrained
I obviously lean towards the first point, though I don't know the complexity that would be necessary. However, I note that something along those lines must already be happening. I wanted to compare to the fully(?!) rendered recipe as shown in the log, and I noticed that all package hashes there are still only the
h1234567
placeholders. For example:That's despite the fact that the run-requirements definitely get populated correctly, and this is also evident from the list of outputs that actually get built:
Conda Info
Conda Config
Conda list
The text was updated successfully, but these errors were encountered: