Re: Subversion 1.15.0-rc3 up for testing/signing

[Branko Čibej]

On 14. 7. 2026 15:52, Ivan Zhakov wrote:
On Tue, 14 Jul 2026 at 16:12, Branko Čibej <br...@apache.org> wrote:
    On 13. 7. 2026 13:37, Evgeny Kotkov via dev wrote:
    The 1.15.0-rc3 release artifacts are now available for testing/signing.
    Please get the tarballs from
       https://dist.apache.org/repos/dist/dev/subversion
    and add your signatures there.

    Thanks!

    I found this difference in the generated targets.cmake between the
    tar.gz and zip archives. There's no semantic difference but I
    wonder if it's relevant, given that the generator is supposed to
    be the same, the conditions shouldn't be reversed like this:

    --- subversion-1.15.0-rc3/build/cmake/targets.cmake     2026-07-13 12:31:42
    +++ zip/build/cmake/targets.cmake       2026-07-13 12:31:23
    @@ -193,7 +193,7 @@
      endif()
      
      # Test changes in libsvn_fs_base
    -if (SVN_ENABLE_PROGRAMS AND SVN_ENABLE_FS_BASE)
    +if (SVN_ENABLE_FS_BASE AND SVN_ENABLE_PROGRAMS)
        add_executable(changes-test
          subversion/tests/libsvn_fs_base/changes-test.c
        )
    @@ -747,7 +747,7 @@
      endif()
      
      # Tests for *public* fs API (svn_fs.h)
    -if (SVN_ENABLE_PROGRAMS AND SVN_ENABLE_FS_BASE)
    +if (SVN_ENABLE_FS_BASE AND SVN_ENABLE_PROGRAMS)
        add_executable(fs-base-test
          subversion/tests/libsvn_fs_base/fs-base-test.c
        )
    @@ -3541,7 +3541,7 @@
      endif()
      
      # Test strings/reps in libsvn_fs_base
    -if (SVN_ENABLE_PROGRAMS AND SVN_ENABLE_FS_BASE)
    +if (SVN_ENABLE_FS_BASE AND SVN_ENABLE_PROGRAMS)
        add_executable(strings-reps-test
          subversion/tests/libsvn_fs_base/strings-reps-test.c
        )
    @@ -4091,7 +4091,7 @@
      endif()
      
      # Subversion Server
    -if (SVN_ENABLE_PROGRAMS AND SVN_ENABLE_RA_SVN)
    +if (SVN_ENABLE_RA_SVN AND SVN_ENABLE_PROGRAMS)
        add_executable(svnserve
          subversion/svnserve/cyrus_auth.c
          subversion/svnserve/log-escape.c



I can confirm this issue, though it doesn't affect the build itself. 
It should be fixed in r1936132.

Hah, interesting, set ordering is different. Nice catch.

Longer term, we could think about getting to a state where our 
.tar.gz, .tar.bz2, and .zip all have identical content (generated 
once). Otherwise we risk potential inconsistencies: for example, if 
someone on Windows downloads subversion-1.14.5.tar.bz2 instead of 
.zip. And it also makes the release process more complicated, because 
there is at least 2x different content to check/verify.

Unix tarballs are libtoolized and contain generated swig binding code 
and such. We don't need the libtool bits and don't support pre-generated 
bindings in the Windows build, so there's a reason why contents are 
somewhat different. And all the files within the .zip with 
svn:eol-style=native have CRLF line endings (that's why I use 'diff 
-qrw` to compare). The latter used to be required because 'cl' and 
'msdev' and 'cmd.exe' didn't use to like just LF newlines in source and 
project files. I suppose that's no longer a concern?

-- Brane