Am 22.09.26 um 19:41 schrieb Joe Orton:
On Tue, Sep 22, 2026 at 10:55:40AM +0200, Rainer Jung wrote:
Hi there,
in trunk there's a detection whether OpenSSL can use &errno as a THREADID.
It was added almost 10 years ago in r1791055.
The program used to check for the feature uses APR headers and links against
APR trying to run the result. When APR is not already in the system path,
this fails. For instance on my system, because the compiled test binary can
not load APR due to missing RPATH. If there is no installed APR lib on the
build system when running configure, that test will also fail.
I can work around this when building on my systems, but I wonder, whether it
is a non-pattern to actually use APR during httpd configure tests. It looks
like this test is the only one doing this.
No idea, whether a simpler detection without APR is easily doable.
It's definitely unusual and possibly unwise. TBH if we've got away
without having that test in 2.4.x, we could easily declare OpenSSL < 3
or OpenSSL < 1.1 unsupported on trunk, and get rid of it entirely at
this point. Do you have systems where you actually care about that test
result?
Increasing the supported OpenSSL version is trunk is certainly possible.
It would make backports a bit more hassle.
The test always failed for me, since I only now found out why. So for me
it would be fine to stick with the non-errno thread id implementation.
Best regards,
Rainer
