Bump Boost shipped for Windows to v1.87

[Al2Klimov]

No description provided.

[Al2Klimov]

[Al2Klimov]

@julianbrost @yhabteab @oxzi In your own interest, please pin Boost to < v1.87 for now.

boost::asio::spawn() doesn't take boost::coroutines::attributes anymore.

[Al2Klimov]

... which is pretty annoying, seems that it works only with fibers now: boostorg/asio@df973a8#diff-4f46281483e5218aa919801a401dcd5b4c19704daa16f036e52da35a2860109dL1320

Definitely nothing for v2.14.4!

[julianbrost]

@julianbrost @yhabteab @oxzi In your own interest, please pin Boost to < v1.87 for now.

That's only a very short-term solution. This will turn into a bigger problem as soon as any distro starts shipping that version. So this needs a proper solution in any case.

[Al2Klimov]

That's only a very short-term solution.

Exactly. A bridge technology.

I mean, aren't you happy that I'm urging NOT urging yet another PR of mine into the current milestone?

This will turn into a bigger problem as soon as any distro starts shipping that version.

Yet, both Ubuntu 25 and Fedora 42 have Boost 1.83. OpenBSD 7.7 has v1.84 yet, on NixOS pinning is easiest of all I guess and MacOS isn't even officially supported.

So this needs a proper solution in any case.

Sure. A long-term solution.

[julianbrost]

I don't say this must go into 2.14.4, I'm just saying that we shouldn't wait until "oh fuck, Fedora 44 is here and we can't build packages".

[Al2Klimov]

Note

• Due to the unit tests from Timeout: use less resources, clean them up better and make cancellation deterministic #10254 we know that Boost 1.87 has broken our Timeout: https://github.com/Icinga/icinga2/actions/runs/12708854995/job/35426656579?pr=10278

[Al2Klimov]

we shouldn't wait until "oh fuck, Fedora 44 is here and we can't build packages".

A theoretical option is still doing the AL2-move, i.e building Icinga-boost 1.86 packages for that one or two $OS. (It's funny because normally that's been done due to Boost being too OLD, not too NEW.😅) Sound radical, but:

1. can still be undone for any future Icinga version
2. other OS have even older Boost versions
3. Fedora has the shortest per-version support lifetime
4. even the current PR state I'd rather not backport, if not necessary (or you insist), until a .0 and a .1 version of Icinga 2 with this PR

Note on the test failure

Unmodified

(WHERE master = 1065d3b)

.../icinga2/test/base-io-engine.cpp:38: error: in "base_io_engine/timeout_run": check called == 1 has failed [0 != 1]

This just says us the Timeout callback, [&called] { ++called; }, hasn't been called.

Has the coroutine even been started?

--- a/test/base-io-engine.cpp
+++ b/test/base-io-engine.cpp
@@ -20,2 +20,3 @@ BOOST_AUTO_TEST_CASE(timeout_run)
        boost::asio::spawn(strand, [&](boost::asio::yield_context yc) {
+               BOOST_CHECK_EQUAL(called, 1337);
                boost::asio::deadline_timer timer (io);

Now the test fails exactly like above, not with 1337, so the answer is no.

We need to go deeper...

--- a/test/base-io-engine.cpp
+++ b/test/base-io-engine.cpp
@@ -2,40 +2,46 @@

 #include "base/io-engine.hpp"
 #include "base/utility.hpp"
 #include <boost/asio.hpp>
 #include <boost/date_time/posix_time/posix_time.hpp>
 #include <BoostTestTargetConfig.h>
+#include <memory>
 #include <thread>

 using namespace icinga;

 BOOST_AUTO_TEST_SUITE(base_io_engine)

 BOOST_AUTO_TEST_CASE(timeout_run)
 {
        boost::asio::io_context io;
        boost::asio::io_context::strand strand (io);
        int called = 0;
+       auto workGuard (std::make_unique<decltype(boost::asio::make_work_guard(io))>(boost::asio::make_work_guard(io)));

        boost::asio::spawn(strand, [&](boost::asio::yield_context yc) {
+               BOOST_CHECK_EQUAL(called, 1337);
                boost::asio::deadline_timer timer (io);

                Timeout timeout (strand, boost::posix_time::millisec(300), [&called] { ++called; });
                BOOST_CHECK_EQUAL(called, 0);

                timer.expires_from_now(boost::posix_time::millisec(200));
                timer.async_wait(yc);
                BOOST_CHECK_EQUAL(called, 0);

                timer.expires_from_now(boost::posix_time::millisec(200));
                timer.async_wait(yc);
+               workGuard = nullptr;
        });

        std::thread eventLoop ([&io] { io.run(); });
+       BOOST_CHECK_EQUAL(called, 23);
        io.run();
+       BOOST_CHECK_EQUAL(called, 42);
        eventLoop.join();

        BOOST_CHECK_EQUAL(called, 1);
 }

 BOOST_AUTO_TEST_CASE(timeout_cancelled)

This one was my previous printf-debugger and it hangs at 0 != 23. I.e run() awaits workGuard, but the coroutine never stops that.

[Al2Klimov]

For the record, this affects only Windows, currently. (Remember! WE control what we ship there.)

[Al2Klimov]

This also affects:

• Fedora 39+ (all)
• Ubuntu 24.04+