squashfs-tools: needs to depend on libgcc-s1

[Andrea Righi]

I'm getting this error running the ADT test for snapd on impish:

021-09-08 07:49:43 Error executing autopkgtest:adt-local:tests/smoke/sandbox (autopkgtest:adt-local) :
-----
++ snap debug confinement
+ '[' strict '!=' strict ']'
+ /home/gopath/src/github.com/snapcore/snapd/tests/lib/tools/snaps-state install-local test-snapd-sandbox
error: cannot pack "/home/gopath/src/github.com/snapcore/snapd/tests/smoke/sandbox/test-snapd-sandbox": mksquashfs call failed:
-----
libgcc_s.so.1 must be installed for pthread_cancel to work
Parallel mksquashfs: Using 1 processor
Creating 4.0 filesystem on /home/gopath/src/github.com/snapcore/snapd/tests/smoke/sandbox/test-snapd-sandbox/test-snapd-sandbox_1.0_all.snap, block size 131072.
-----
error: cannot read snap file: "/var/lib/snapd/snaps/.local-install-443592474"
       is not a snap or snapdir
-----

[Maciej Borzecki (maciek-borzecki)]

AFAIU the extra line in the output (libgcc_s.so.1 must be installed for pthread_cancel to work) appears when mksquashfs is invoked as squashfs-tools is missing a dependency on libgc-s1 (previously libgcc1).

[Alex Murray (alexmurray)]

My understanding was that this was an issue in snapd, not squashfs-tools directly - https://github.com/snapcore/snapd/pull/10757

[Dan Bungert (dbungert)]

For the snapd autopkgtest problem, you are correct and that changes have been committed for the snapd snapcraft file to include libgcc.

The dependency is also correct for impish - I spawned a container to experiment with a system that did not have libgcc-s1. libgcc-s1 is Priority: required and attempting to `apt remove libgcc-s1` offers to remove 61 other essential packages. Manually removing the library and running a trivial mksquashfs case does cause an equivalent failure to the above. Based on this the dependency looks correct but functionally redundant.

I have dropped the patch and unsubscribed sponsors, and forwarded the suggestion to Debian.

[Dan Bungert (dbungert)]

@arighi - to your original question, mksquashfs seems OK when using snapd from the edge channel. I'm building a ppa to help verify that the full autopkgtest is happy.

[Dan Bungert (dbungert)]

Log from local testing.
Forcing snapd to edge passes this test.

[Andrea Righi (arighi)]

Any progress on this? It looks like this issue is currently blocking the promotion of all our impish kernels for the beta.

[Dan Bungert (dbungert)]

> Any progress on this?

If the snapd snap on the edge channel was on stable, this would be a non-issue because libgcc_s has been added to that snap. If you like, you can verify this for yourself by making a modification to the snapd package to force the edge channel snap to be installed and running it thru autopkgtest.

Adding the dependency from squashfs-tools -> libgcc-s1 is ultimately a good thing for package correctness but will do nothing to resolve the current autopkgtest dilemma because it's the mksquashfs inside the snapd snap that wants a matching libgcc_s. See also above comment I made which shows that libgcc_s is de-facto present on any realistic system.

I'm strongly considering just hinting the relevant packages and being done with it, but there is the following problem:
Why are any of the snapd autopkgtests passing right now?

looking at current history of amd64 snapd tests, the last 4 runs were triggered with
* linux-aws (pass)
* linux-gcp (pass)
* linux-oracle (fail)
* linux-meta-kvm (pass)

I looked at the logs and artifacts for oracle versus the others and I'm not seeing interesting reasons on why -oracle would fail when -gcp passes. And when I run them locally, they fail every time, including passing scenarios like the above linux-gcp. And why would -oracle would pass 4 days ago and not now?

Again, maybe we just hint and move on, but something odd is happening.