On Wed, Sep 13, 2023 at 04:44:05PM +0100, Stuart Henderson wrote: > On 2023/09/10 11:54, Theo Buehler wrote: > > On Sun, Sep 10, 2023 at 11:46:24AM +0200, Theo Buehler wrote: > > > On Fri, Sep 08, 2023 at 11:53:15PM +0200, Theo Buehler wrote: > > > > Here is an updated version of my previous abandoned attempt at updating > > > > this google monster: https://marc.info/?t=167689329100001&r=1&w=2 > > > > > > > > A major annoyance comes from abseil-cpp shipping 20 gazillion components > > > > combined with a misdesign in our pkg-config. The result is that protobuf > > > > detection in configure scripts takes a ridiculous amount of time. > > > > Configuring libphonenumber takes 10 minutes on a fast amd64 box and over > > > > half an hour on a fast sparc64 box. pkgconf is a lot faster. > > > > > > > > Remaining fallout: audio/clementine, devel/protobuf-c, games/pokerth > > > > > > > > * clementine and cmake make it hard to pass -std=gnu++17 to the various > > > > components. There are some leads on GH. I haven't tried myself, and > > > > I won't: https://github.com/clementine-player/Clementine/issues/7272 > > > > > > > > * pokerth is similarly obtuse. Again this can be fixed by someone else. > > > > > > > > * I sent out a diff for protobuf-c earlier today. All its consumers > > > > build. > > > > > > > > py3-protobuf works for me, the only thing I couldn't figure out is how > > > > to make tests pass when the package is installed. > > > > > > > > I have tested this on amd64, arm64 and sparc64. It ran through an amd64 > > > > bulk (as 4.23.2, a bulk with 4.23.3 is ongoing). Unless the pkg-config > > > > vs abseil-cpp thing is considered a blocker (I'd argue it will give more > > > > incentive to fix pkg-config which needs to be done anyway), I'd like to > > > > land this modulo tweaks from review. > > > > > > Here's a better diff that disables the tests. This simplifies the port > > > and also avoids installing libgmock for no good reason. Additionally, > > > switch to shared libraries for libutf8_{range,validity}. > > > > Missed cvs add for utf8_range's CMakeLists.txt. > > "Re: path: speed-up pkg-config" from tech@ probably wants to be landed > fairly soon when this is in - it's still pretty slow with that, but much > less bad than -current.
Indeed. > There will be a load of new WANTLIB vomit in dependent ports (see below) > but I don't suppose there's much we can do about that (output from > "pkg-config --libs protobuf" looks about the same on other OS). > Thanks Google. Ugh. Sorry for missing this. > I'm broadly OK with this but as it's getting a bit tight for > release timing we probably need to throw some rocks at > pokerth/clementine build process until they can be made to work > Now that I have it built I suppose I can do that, though won't > be able to do much in the way of run testing for those. Let's not rush it in before release then. It's not that important. This gives me time to prepare diffs for all that wantlib churn, we can ship pokerth/clementine and whoever cares about these can fix them in the coming months.
