Details
-
Task
-
Status: Open (View Workflow)
-
Major
-
Resolution: Unresolved
-
None
Description
Description
Our spec file currently uses redhat-rpm-config's %filter_setup macro to filter auto-generated Provides/Requires. This macro forces RPM's legacy external dependency-generator path instead of the default internal one, which has performance costs, most visible on our largest-file-count subpackages (e.g. MariaDB-test, MariaDB-server).
How %filter_setup works:
%global _use_internal_dependency_generator 0
|
%global __deploop() while read FILE; do echo "${FILE}" | /usr/lib/rpm/rpmdeps -%{1}; done | /bin/sort -u |
%global __find_provides /bin/sh -c "%{?__filter_prov_cmd} %{__deploop P} %{?__filter_from_prov}" |
%global __find_requires /bin/sh -c "%{?__filter_req_cmd} %{__deploop R} %{?__filter_from_req}" |
Setting use_internal_dependency_generator to 0 forces rpmbuild onto the deprecated external-generator branch in rpm's own source. The __deploop line above then forks a brand-new rpmdeps process once per packaged file, thus absorbing its bootstrap performance penalty on every invocation.
Proposed changed:
Migrate our spec's filtering to rpm's native, in-process filtering macros, which don't touch use_internal_dependency_generator at all.
Ref: https://docs.fedoraproject.org/en-US/packaging-guidelines/AutoProvidesAndRequiresFiltering/
With the diff from below, on top of 11.8, I obtained the following
times on .spec files processing during cpack's invocation of rpmbuild.
I don't know why mariadb-devel took longer to process but this is a separate issue that was investigated as part of: https://mariadb.slack.com/archives/C012UURLW9W/p1787747064513859
diff --git a/cmake/cpack_rpm.cmake b/cmake/cpack_rpm.cmake
|
index e27c06534c6..d2dc1409569 100644
|
--- a/cmake/cpack_rpm.cmake
|
+++ b/cmake/cpack_rpm.cmake
|
@@ -125,13 +125,10 @@ SET(CPACK_RPM_SPEC_MORE_DEFINE "
|
|
|
%define pretrans %{nil}
|
|
|
-%{?filter_setup:
|
-%filter_provides_in \\\\.\\\\(test\\\\|result\\\\|h\\\\|cc\\\\|c\\\\|inc\\\\|opt\\\\|ic\\\\|cnf\\\\|rdiff\\\\|cpp\\\\)$
|
-%filter_requires_in \\\\.\\\\(test\\\\|result\\\\|h\\\\|cc\\\\|c\\\\|inc\\\\|opt\\\\|ic\\\\|cnf\\\\|rdiff\\\\|cpp\\\\)$
|
-%filter_from_provides /perl(\\\\(mtr\\\\|My::\\\\)/d
|
-%filter_from_requires /\\\\(perl(\\\\(.*mtr\\\\|My::\\\\|.*HandlerSocket\\\\|Mysql\\\\)\\\\)/d
|
-%filter_setup
|
-}
|
+%global __provides_exclude_from ^.*\\\\.(test|result|h|cc|c|inc|opt|ic|cnf|rdiff|cpp)$
|
+%global __requires_exclude_from ^.*\\\\.(test|result|h|cc|c|inc|opt|ic|cnf|rdiff|cpp)$
|
+%global __provides_exclude ^perl\\\\((mtr|My::)
|
+%global __requires_exclude ^perl\\\\((.*mtr|My::|.*HandlerSocket|Mysql)
|
")
|
|
|
# this creative hack is described here: http://www.cmake.org/pipermail/cmake/2012-January/048416.html |
The resulting packages should be checked for any side-effects, mostly the Requires/Provides headers must be identical between the two methods.
