Description of problem: The rawhide (f39) build of whatip failed despite the f38 build succeeding: https://kojipkgs.fedoraproject.org//work/tasks/4470/97484470/build.log /usr/bin/blueprint-compiler batch-compile data/. ../data ../data/ui/shortcutsWindow.blp ../data/ui/headerbar.blp ../data/ui/lan_listbox_row.blp ../data/ui/listbox_with_empty_state.blp ../data/ui/net_iface_listbox_row.blp ../data/ui/net_port_listbox_row.blp Traceback (most recent call last): File "/usr/lib/python3.11/site-packages/blueprintcompiler/main.py", line 58, in main opts.func(opts) File "/usr/lib/python3.11/site-packages/blueprintcompiler/main.py", line 106, in cmd_batch_compile xml, warnings = self._compile(data) ^^^^^^^^^^^^^^^^^^^ File "/usr/lib/python3.11/site-packages/blueprintcompiler/main.py", line 141, in _compile if len(ast.errors): ^^^^^^^^^^ File "/usr/lib64/python3.11/functools.py", line 1001, in __get__ val = self.func(instance) ^^^^^^^^^^^^^^^^^^^ File "/usr/lib/python3.11/site-packages/blueprintcompiler/ast_utils.py", line 78, in errors return list(self._get_errors()) ^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/lib/python3.11/site-packages/blueprintcompiler/ast_utils.py", line 90, in _get_errors yield from child._get_errors() File "/usr/lib/python3.11/site-packages/blueprintcompiler/ast_utils.py", line 90, in _get_errors yield from child._get_errors() File "/usr/lib/python3.11/site-packages/blueprintcompiler/ast_utils.py", line 90, in _get_errors yield from child._get_errors() File "/usr/lib/python3.11/site-packages/blueprintcompiler/ast_utils.py", line 83, in _get_errors validator(self) File "/usr/lib/python3.11/site-packages/blueprintcompiler/ast_utils.py", line 149, in inner func(self) File "/usr/lib/python3.11/site-packages/blueprintcompiler/language/gtkbuilder_child.py", line 53, in parent_can_have_child if gir_class.assignable_to(parent_type): ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/lib/python3.11/site-packages/blueprintcompiler/gir.py", line 398, in assignable_to elif self.parent and self.parent.assignable_to(other): ^^^^^^^^^^^ File "/usr/lib64/python3.11/functools.py", line 1001, in __get__ val = self.func(instance) ^^^^^^^^^^^^^^^^^^^ File "/usr/lib/python3.11/site-packages/blueprintcompiler/gir.py", line 374, in parent return self.get_containing(Repository)._resolve_dir_entry(entry) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/lib/python3.11/site-packages/blueprintcompiler/gir.py", line 603, in _resolve_dir_entry return self.lookup_namespace(ns).get_type(dir_entry.DIR_ENTRY_NAME) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ AttributeError: 'NoneType' object has no attribute 'get_type' Arguments: ['/usr/bin/blueprint-compiler', 'batch-compile', 'data/.', '../data', '../data/ui/shortcutsWindow.blp', '../data/ui/headerbar.blp', '../data/ui/lan_listbox_row.blp', '../data/ui/listbox_with_empty_state.blp', '../data/ui/net_iface_listbox_row.blp', '../data/ui/net_port_listbox_row.blp'] ***** COMPILER BUG ***** The blueprint-compiler program has crashed. Please report the above stacktrace, along with the input file(s) if possible, on GitLab: https://gitlab.gnome.org/jwestman/blueprint-compiler/-/issues/new?issue ninja: build stopped: subcommand failed. Version-Release number of selected component (if applicable): blueprint-compiler-0.6.0-2.fc38 (in f39-build) How reproducible: Multiple times in rawhide, but not in stable branches
Hello ! It's an annoying bug that only happens on s390x builders. 0.6.0 wasn't even meant to be pushed, but I let the commit sit in rawhide by error, as I thought the bug was going to be fixed by upstream quickly, but they didn't. I ended up forgetting about it until today, and it was pushed with the mass rebuild. I am aware this breaks the guideline about SCM needing to be ready to be functional at any time. So, I'll revert this with an epoch, unless you know of a way to instruct noarch packages to be built on any builder except s390x? (I asked about it on matrix fedora-devel, but my question wasn't answered...) Since blueprint-compiler would still produce usable ui files for s390x, I'd prefer to keep this update, but just make sure Fedora wouldn't build it and its dependent packages on s390x.
(Just to be clear, if I end up having to revert the update, I'll still keep this issue open and try to look into the bug myself)
Ok, I think I got an idea where the bug could be (I think it's in this function, if you want to look further: https://gitlab.gnome.org/jwestman/blueprint-compiler/-/blob/main/blueprintcompiler/typelib.py#L229). Although, I am unfortunately not able to look at it right now, and it's s390x, an arch I know nothing about. So, I'll try to look at it tomorrow!
The odd thing is that 0.6.0 includes a supposed big-endian fix at that spot: https://gitlab.gnome.org/jwestman/blueprint-compiler/-/merge_requests/76 Let's try to get some help here, blocking the s390x tracker.
blueprint-compiler 0.4.0 also fails on s390x, I somehow never bothered to check, and it was never built on s390x at the time by some kind of luck: https://copr.fedorainfracloud.org/coprs/lyessaadi/blueprint-compiler-0.6.0/build/5561735/ And since it was noarch, it didn't get tested for every arch at review/import... So, downgrading by increasing the Epoch is out of the question :/. Anyway, while the bug is solved, there are two paths forward since it doesn't seem that we can have the infra build on everything except a specific arch for noarch ? Options: - Wait for a fix, blocking multiple apps and introducing surprise failures from time to time in updates. - Turn blueprint-compiler to an arch-specific package, but would be kind of playing with guidelines? I think if this path forward is chosen, asking fesco for a guideline exception might be preferable. Anyway, would like to hear about your opinion to move forward, as I am quite conflicted between both options, as we do not know how long this bug might last...
Dan this looks like an endianness issue, any chance you could help?
Hi Yaakov, how urgent is it? I could probably find some spare cycles in next weeks. As I know nothing about blueprimnt-compiler, is there a minimal test case that exposes the failure on s390x?
Yes, there is. blueprint-compiler's builds on s390x also fail in the test section. You can also find more about the bug here : https://gitlab.gnome.org/jwestman/blueprint-compiler/-/issues/96
This bug appears to have been reported against 'rawhide' during the Fedora Linux 39 development cycle. Changing version to 39.