Script 'mail_helper' called by obssrc
Hello community,
here is the log from the commit of package python-apache-tvm-ffi for
openSUSE:Factory checked in at 2026-08-01 18:33:05
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Comparing /work/SRC/openSUSE:Factory/python-apache-tvm-ffi (Old)
and /work/SRC/openSUSE:Factory/.python-apache-tvm-ffi.new.16738 (New)
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Package is "python-apache-tvm-ffi"
Sat Aug 1 18:33:05 2026 rev:2 rq:1368783 version:0.1.13
Changes:
--------
---
/work/SRC/openSUSE:Factory/python-apache-tvm-ffi/python-apache-tvm-ffi.changes
2026-07-15 16:55:36.890960265 +0200
+++
/work/SRC/openSUSE:Factory/.python-apache-tvm-ffi.new.16738/python-apache-tvm-ffi.changes
2026-08-01 18:36:29.682629097 +0200
@@ -1,0 +2,26 @@
+Fri Jul 31 07:52:41 UTC 2026 - Martin Pluskal <[email protected]>
+
+- Update to 0.1.13:
+ * Breaking: dataclasses field resolution was separated from native
+ lifetimes, native enum state was unified and reflection now
+ requires exact field and method name matches
+ * Optional and Variant are now backed by TVMFFIAny for a stable
+ ABI layout, breaking code compiled against 0.1.12
+ * Minimum supported Python is now 3.9
+ * Add structural visitor and typed structural walk APIs, type
+ subsumption relations and an Expected<void> specialisation
+ * Tie the Python wrapper lifetime to the underlying C++ FFI object
+ * Support pickling of FFI String and Bytes, expose Tensor data_ptr
+ * Expand dtype parsing and introspection, require canonical Python
+ device names
+ * Refactor stubgen and improve typing of overloaded methods
+ * Rust bindings gained Map, Optional and the match_any! macro
+ * ORCJIT addon gained a shared session, a high-level load_module
+ and an embedded liborc_rt
+ * Core wheels are now tagged per Python version instead of abi3
+ * Various fixes: exception-safe object construction, uint1 tensor
+ packing, JIT GPU backend detection, hidden static library symbols
+- Require Cython >= 3.2.8, CMake >= 3.26, scikit-build-core >= 0.10.0
+ and ninja >= 1.11 as mandated by upstream
+
+-------------------------------------------------------------------
Old:
----
apache_tvm_ffi-0.1.12.tar.gz
New:
----
apache_tvm_ffi-0.1.13.tar.gz
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Other differences:
------------------
++++++ python-apache-tvm-ffi.spec ++++++
--- /var/tmp/diff_new_pack.9kZHFb/_old 2026-08-01 18:36:31.642696575 +0200
+++ /var/tmp/diff_new_pack.9kZHFb/_new 2026-08-01 18:36:31.654696988 +0200
@@ -18,25 +18,25 @@
%{?sle15_python_module_pythons}
Name: python-apache-tvm-ffi
-Version: 0.1.12
+Version: 0.1.13
Release: 0
Summary: Minimal FFI runtime and ABI for machine learning systems
License: Apache-2.0
URL: https://github.com/apache/tvm-ffi
Source:
https://files.pythonhosted.org/packages/source/a/apache_tvm_ffi/apache_tvm_ffi-%{version}.tar.gz
Source99: %{name}-rpmlintrc
-BuildRequires: %{python_module Cython >= 3.0}
+BuildRequires: %{python_module Cython >= 3.2.8}
BuildRequires: %{python_module devel}
BuildRequires: %{python_module pip}
-BuildRequires: %{python_module scikit-build-core}
+BuildRequires: %{python_module scikit-build-core >= 0.10.0}
BuildRequires: %{python_module setuptools-scm}
BuildRequires: %{python_module setuptools}
BuildRequires: %{python_module typing_extensions >= 4.5}
BuildRequires: %{python_module wheel}
-BuildRequires: cmake
+BuildRequires: cmake >= 3.26
BuildRequires: fdupes
BuildRequires: gcc-c++
-BuildRequires: ninja
+BuildRequires: ninja >= 1.11
BuildRequires: python-rpm-macros
Requires: python-typing_extensions >= 4.5
Requires(post): update-alternatives
@@ -71,9 +71,10 @@
%python_expand %fdupes %{buildroot}%{$python_sitearch}/tvm_ffi
%check
-# The upstream pytest suite requires torch, numpy and ml_dtypes (unpackaged
-# here) and generates/loads native modules at runtime, so it is not runnable
-# in the build root. Fall back to an import smoke test that loads the compiled
+# The upstream pytest suite pulls the whole torch stack into the build root
+# (its "test" dependency group includes torch, numpy and ml_dtypes) and
+# JIT-compiles and loads native modules while the tests run, so it is not
+# practical here. Fall back to an import smoke test that loads the compiled
# Cython core extension and the bundled runtime shared library.
%python_expand PYTHONPATH=%{buildroot}%{$python_sitearch} $python -Bc "import
tvm_ffi; print(tvm_ffi.__version__)"
@@ -91,7 +92,7 @@
%python_alternative %{_bindir}/tvm-ffi-config
%python_alternative %{_bindir}/tvm-ffi-stubgen
# The whole tvm_ffi tree is shipped intact: besides the Python package and the
-# compiled core.abi3.so it carries lib/libtvm_ffi*.so, the C++ headers under
+# compiled core extension it carries lib/libtvm_ffi*.so, the C++ headers under
# include/, the JIT sources under src/ and the CMake package config under
# share/cmake/tvm_ffi/ that downstream consumers (e.g. xgrammar) discover via
# find_package(tvm_ffi CONFIG). Keeping the headers/CMake config in the Python
++++++ apache_tvm_ffi-0.1.12.tar.gz -> apache_tvm_ffi-0.1.13.tar.gz ++++++
++++ 37029 lines of diff (skipped)
