On 07.03.2018 17:15, Nathan Froyd wrote:
FWIW, the Android team is planning on making standalone toolchains
obsolete in the relatively near future (the roadmap says later this
year):
https://android.googlesource.com/platform/ndk/+/master/docs/Roadmap.md#make-standalone-toolchains-obsolete
Actually, this should have been done aeons ago - and the whole idea of
all these "sdk"'s is pretty ridiculous. Instead there should be an clean
separation between individual packages (incl. kernel, bootloader, etc),
toolchains (compiler, binutils, ...) and the distro buildsystem system
(eg. debuild, buildd, ...) and target- as well as dist-specific
confguration. That's pretty much what we're doing w/ debian world as
well es embedded kits like ptxdist.
For an android-alike system that primarily means:
* different c-library (which nowadays wouldn't be necessary anymore,
as non-ancient mobile handsets are strong enough for running plain
glibc+friends) ... historically there had been a strong connection
between libc and gcc (had to be compiled together, to fit each other),
but that also had been lifted long ago (IIRC clang never had that
problem)
* customized kernel, bootloader, image layouts, etc (basically what
ptxdist calls the platform configuration)
* additional java->dalvik compiler
All the rest is just usual sw packages ... pretty boring from build
engineering pov. Unfortunately, Google doesn't wan't an simple and
straightforward, easily understandable and customizable system.
(it's really time for creating an own mobile/handheld distro)
For servo, we'd have to add the rust toolchain (configured for the
particular target - within an arch, it should be pretty generic).
It's long ago since I touched android the last time, but IMHO, the
first step should be integrating rust into several distros - up to
the esoteric ones like android - then treat servo as just yet
another library package.
Actually, servo shouldn't be a library, but a dedicated (server)
application, which the integrating application just talks to, but
doesn't link against (the Plan9 / suckless approach). I did some
experiments in that direction several months ago, but yet lacking
time for deeper involvement.
One thing's absolutely clear for me: give up the whole concept of
SDKs as such. It never fits well and just horrible to maintain.
--mtx
--
Enrico Weigelt, metux IT consult
Free software and Linux embedded engineering
i...@metux.net -- +49-151-27565287
_______________________________________________
dev-servo mailing list
dev-servo@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-servo
