On 20. 5. 2026 15:30, Branko Čibej wrote:
On Wed, 20 May 2026, 13:13 Evgeny Kotkov via dev,
<[email protected]> wrote:
Evgeny Kotkov <[email protected]> writes:
> The 1.15.0-rc2 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.
Given the current responses, I guess that rolling an RC3 build is
warranted.
Here is a summary of our current state and the reported issues:
1) The macOS build issues.
These should be resolved by updating libtool to 2.4.7 (r1934402).
The RC3 build will use the updated libtool version.
2) The venv-related issues when running the test suite on Windows.
Based on the description, this appears to be a regression from
1.14.5.
3) On Windows, the vcxproj generator fails to link against z.lib
(the newer
library name starting from zlib 1.3.2 (or 1.3?)).
The fix is still being discussed, but maybe we can get it in
place for RC3.
This is not a regression compared to 1.14.5, so not a strict
blocker.
4) Several failing Windows tests reported by jun66j5, potentially
related
to packed FSFS repositories.
It seems that we still need to determine if this is a
regression and
identify the root cause.
My current plan is to wait for a resolution on item (2) and for
information
on item (4) to confirm whether it is a regression or not. After
that, I am
hoping to roll the RC3 build.
If I have overlooked any details from the thread, please let me know.
Thanks,
Evgeny Kotkov
On 2): I'm working on a different approach that would download and
install the dependencies (in svn-test-work) only if explicitly requested.
That should solve the reported issues. XML schema validation is
already optional and won't be performed if lxml and the schema
converter aren't available, without affecting test results.
In the CI workflows, we can install python-lxml etc.
Earlier today, I unpacked the -rc2 tarball, removed all Swig-generated
files and ran `autogen.sh --release` in it with my PATH set up so that
it found libtool-2.4.7, autoconf-2.69 and swig-3.0.12. Then I built
that, ran all check-swig-* and check-javahl and check and all build and
tests passed.
So, I think we're good on macOS with the RM libtool upgrade. However, we
still have to decide what to do about the default for "download and
install Python modules for XML validation in tests". I've installed
those dependencies independent of the test suite on almost all CI
runners – CMake macOS is the exception – and I've seen no issues, the
XML validation runs as expected and nothing gets installed on the fly.
We still have a number of backports proposed in STATUS, most of which I
can't test and a couple that I proposed (especially the one about not
requiring 'venv'). I think all of them should be in the next release
candidate so please review, thanks!
-- Brane
P.S.: Regarding the r1931334 group proposal: I ran some very informal
benchmarks that showed up to 30% higher throughput when using the new
code optimised for adler32_z(). Sorry, I don't have the test program
around any more, but yes, I was surprised with the result, too.
