On Sun, Jun 8, 2025 at 12:52 PM Branko Čibej <br...@apache.org> wrote: > > On 8. 6. 25 12:32, Timofei Zhakov wrote: > > On Sat, Jun 7, 2025 at 2:52 PM Branko Čibej<br...@apache.org> wrote: > >> On 7. 6. 25 14:11, rinrab (via GitHub) wrote: > >>> rinrab opened a new pull request, #6: > >>> URL:https://github.com/apache/serf/pull/6 > >>> > >>> (all commits are separate changes) > >> We don't have to change the generator to achieve this; MSBuild is > >> already multi-config. The reason why I haven't made the equivalent > >> change yet is that, if debug build or test fail, the release build will > >> never happen. I'm still thinking about the ramifications of that, but I > >> for sure do not want to split the different config builds into two > >> separate jobs. Or maybe I do, I'll have to figure that out. > >> > >> -- Brane > > Yes, this is not mandatory to change the generator. But as I already > > mentioned (I think in the "GitHub actions notifications" thread), > > Ninja generator is much better for automation scripts, like GitHub > > Actions: > > > > [[[ > > However, unrelatedly to the original problem, it's much better to use > > the Ninja generator in build scripts, since it's faster, more > > reliable, simpler, and more customisable in terms of platforms and > > architectures. It could be complicated to set up, but I've done it for > > Subversion, so I can do it for Serf as well. > > ]]] > > Yah. As long as the build takes 10 minutes, of which 9½ are spent > building dependencies from source, I strongly doubt we'd gain much time > by moving to Ninja. Given that this is Windows, we may as well use the > "official" (heh) build tool. I just added '-j' to the builds and they're > almost not sluggish* at all any more. If I could do a cross build and > test of x86 Serf with MSBuild on an ARM64 Windows virtual machine, I'd > say it's customisable enough. Frankly I don't know what you mean by > "more reliable".
I mean that Visual Studio sometimes may work in a way no one would expect it to. For example, that issue with openssl, where it didn't insta-fail, and we couldn't find the source of the problem. Also we can set the same job to build on Linux as well. > Everything else uses generated makefiles (well, except the SCons build > of course). > > -- Brane > > * "almost not sluggish" ... I saw that Microsoft officially supports > clang (llvm) for ARM64, and it's "only" about 5 time faster than MSVC. > But this is about the build tool, not the compiler. -- Timofei Zhakov