Request 1129091 (accepted)

This request supersedes: request 1124429 (Show diff)

Overview

Request 1129091 accepted

- update to 3.27.8 (bsc#1215888, CVE-2023-38545):
* cmComputeLinkDepends: also copy the target from object link items
* cmComputeLinkInformation: skip over linking to items for object purposes
* Fortran: Restore support for TARGET_OBJECTS providing modules
* curl: Backport SOCKS5 heap buffer overflow fix from curl 8.4.0
* GetPrerequisites: Revert “Ignore relative paths printed by ldd”
* cmCommonTargetGenerator: Factor out GetLinkedTargetDirectories loop body
* cmComputeLinkInformation: Track targets named by TARGET_OBJECTS sources
* FindOpenMP: Add support for openmp 5.1 (llvm 17) and 5.2

- modify shebang to the actual python interpreter on install
target_sources() command.
and turn off LTO
fix may be restored in a future version of CMake via a policy.
CMAKE_CONFIGURATION_TYPES
CMAKE_INSTALL_MODE
* Multiple further bug fixes
race
compiler generated headers.
* The Compile Features functionality was updated for MSVC 19.28
https://blog.kitware.com/cmake-3-17-0-available-for-download/
* In CMake 3.15.0 through 3.15.3, the EXCLUDE_FROM_ALL directory
within the directory to be excluded even from its own “all”.
This has been fixed. The bug also existed in 3.14.0 through
- Use pkgconfig for dependencies where possible (required due to
find_package() command to fail on a missing package even
* reverted change to pass compiler implicit include directories
* Cmake(1) build tool (cmake--build) gained --parallel []
* CMAKE_FOLDER variable was added to initialze the FOLDER property
* ChangeLog, see: https://cmake.org/cmake/help/v3.12/release/3.12.html

Created by dirkmueller 7 months ago
In state accepted
Supersedes 1124429

Submit cmake

Comments for request 1129091 0
Comments for request 1124429 5

Ana Guerrero (anag+factory) - 8 months ago

reviewer

@simotek what's the difference with https://build.opensuse.org/request/show/1120469 ?

Dominique Leuenberger (dimstar) - 8 months ago

reviewer

IMHO we should skip this NDEBUG madness and revert the macro changes..

Even the referenced redhat bug is kinda not promising:

 Peter Hutterer 2019-04-08 06:58:05 UTC

This change is causing packages that rely on assert() to break, see e.g. bug 1696264.

Any reason you can't just add -Db_ndebug to the mesa build, instead of changing the behaviour of the build system?

Comment 13 Michael Catanzaro 2019-04-08 13:02:20 UTC

Traditionally, some packages expect to be compiled with NDEBUG and others simply don't. It's kinda expected that this change will break packages that haven't tested release builds. Keep in mind this is just making plain builds (distro builds) more similar to release builds.

Jan Engelhardt (jengelh) - 8 months ago

reviewer

we could also just remove llvm16 and rust 1.71 (llvm17 & rust 1.73 are fine), something that would normally happen in due time anyway

Dominique Leuenberger (dimstar) - 8 months ago

reviewer

we could also just not break things as long as we don't have to

Dominique Leuenberger (dimstar) - 8 months ago

reviewer

For completeness:

> dependson llvm16
llvm16 :
      MozillaFirefox
      MozillaThunderbird
      cnf-rs
      flux
      include-what-you-use
      river
      rpcs3
      rust1.70
      rust1.70:test
      rust1.71
      rust1.71:test
      rust1.72
      rust1.72:test
      velociraptor
      velociraptor:client
      waylock
      zig

There is quite a bit behind llvm16