Reduce number of linux release binaries #11083
Conversation
|
|
hasufell
force-pushed
the
stable-haskell/feature/release-ci-pr2
branch
2 times, most recently
from
d3e473e to
2b167fa
Compare
hasufell
force-pushed
the
stable-haskell/feature/release-ci-pr2
branch
from
2b167fa to
35aabea
Compare
- glibc (dynamic) - musl (fully static) 'gmp' and 'zlib' are always statically linked.
hasufell
force-pushed
the
stable-haskell/feature/release-ci-pr2
branch
from
35aabea to
fa10655
Compare
|
I think this is an interesting direction to investigate but I think that we should let the migration from gitlab to github release CI settle down first before making other major changes to the packaging and distribution story. It could be worth discussing this issue in the cabal developers meeting or on a cabal proposal if there are differences of opinion about what's best to do. I'm normally just quite conservative when it comes to changing things to do with distribution. |
| run: | | ||
| cd out | ||
| tar xf *.${TARBALL_EXT} | ||
| ldd cabal | grep --quiet gmp && exit 1 |
There was a problem hiding this comment.
This check could also check that there aren't other libraries dynamically linked just-in-case others slip into the build plan accidentally.
| }, | ||
| { image: "fedora:36" | ||
| { image: "fedora:37" |
There was a problem hiding this comment.
These test platform changes seem unrelated.
| curl -O -L ${{ env.GMP_URL }}/gmp-${{ env.GMP_VERSION }}.tar.xz | ||
| tar xf gmp-${{ env.GMP_VERSION }}.tar.xz | ||
| cd gmp-${{ env.GMP_VERSION }} | ||
| CFLAGS=-fPIC ./configure --prefix=$HOME/.local/ --disable-shared |
There was a problem hiding this comment.
AFAIU, static linking of the gmp library is suitable for the distribution of cabal-install since it is possible for an end user to rebuild the relevant cabal-install release (since the source code is distributed) and link against a different version of gmp.
As I think this point is commonly misunderstood, at least linking to a description of this issue would be a good idea I think in case someone else wonders the same thing.
There was a problem hiding this comment.
I was wondering about a more practical issue: cabal doesn't use gmp itself, the dependency comes from the RTS. How safe is linking against a different version than GHC was built against?
I don't think I will have capacity to write a cabal proposal or come back to this in a couple of months, so feel free to close this PR as rejected in that case. |
|
I just copied the Alpine update into |
|
We have discussed this on today’s Cabal-dev call. It is an interesting PR, we're interested in exploring this direction in future. Thanks for your work in this area. We prefer to postpone any such changes after the recent rewrite of CI has stabilised. This means at least after the next major release (3.18). I've removed the |
Whether something is a Draft or not is up to the PR author to decide, unless you can demonstrate what the PR specifically is lacking (which you couldn't). I'm not going to work on this in the future. |
As described here: #11078
Depends-on: #11072