On 25/07/2026 22:25, Samuel Thibault wrote:
Michael Kelly, le sam. 25 juil. 2026 22:22:47 +0100, a ecrit:
On 24/07/2026 00:39, Samuel Thibault wrote:
A data point: I had a look at which files are corrupted, in the
haskell-vector case:
- ok: ./dist-ghc/build/Data/Vector/Generic/New.dyn_o
- ok: ./dist-ghc/build/libHSvector-0.13.2.0-E0JkXUvrp1JFKkC2UQ3cX8-ghc9.10.3.so
- ko:
./debian/tmp/usr/lib/haskell-packages/ghc/lib/x86_64-hurd-ghc-9.10.3-inplace/libHSvector-0.13.2.0-E0JkXUvrp1JFKkC2UQ3cX8-ghc9.10.3.so
so it's at cabal-installation time that things go wrong.
Is this the case for all other haskell package build corruptions?
I haven't checked other packages, but in this case there are like 3-4
corruptions, and all only in the installed file, not in the linked file.
I've spent quite a lot of time over the last few weeks trying all sorts
of combinations of test case but without any success at generating
corruption. I think that I've recreated a very close match to the
boralus virtual machine so perhaps I need a closer match to the
underlying hardware to make progress this way. I'm likely going to have
to generate an alternative plan.
The only new data that might be relevant is that the corruption is not
limited to .so files. I have also found some .dyn_hi files that have a
very similar corrupted state:
usr/lib/haskell-packages/ghc/lib/x86_64-hurd-ghc-9.10.3-inplace/skylighting-0.14.7-Hh4gkgkOhRoB6RoMnCyy1f/Skylighting/Syntax/Octave.dyn_hi
usr/lib/haskell-packages/ghc/lib/x86_64-hurd-ghc-9.10.3-inplace/vector-0.13.2.0-E0JkXUvrp1JFKkC2UQ3cX8/Data/Vector/Primitive.dyn_hi
usr/lib/haskell-packages/ghc/lib/x86_64-hurd-ghc-9.10.3-inplace/vector-0.13.2.0-E0JkXUvrp1JFKkC2UQ3cX8/Data/Vector.dyn_hi
I'm trying to decipher the Haskell Cabal build system code to understand
precisely what happens during the installation from 'dist-ghc/build' to
'debian/tmp'. There doesn't seem to be anything especially unusual but
we'll see.
Mike.
