On 1/26/26 13:00, enh wrote:
i assume you've already come across
`-fdebug-prefix-map=/proc/self/cwd=`
I had not!
and there's some reason you can't
inject that into your toolchain builds?
In theory I can just throw a "cd /proc/self/cwd" into
scripts/mcm-buildall.sh...
I'm trying to get away from that and use a cross linux from scratch
style toolchain build (they added support for gcc 15.1 but did NOT add
support for musl-1.2, but my city got invaded by the SS recently, which
has been a bit of a drain on available spoons. ELC is here in
minneapolis in may assuming there still is one by then, and the talk
proposal deadline's the 8th. Trying to scrape up the enthusiasm to do a
couple proposals...)
this is what we use in Android
builds, primarily as a build hermeticity/OTA diff reduction strategy:
https://source.android.com/docs/core/ota/reduce_size#build_directory
I'm all for it.
Alas, trying the naieve cd /proc/self/cwd thing didn't work because both
/proc/self and the cwd under it change dynamically and thus subprocesses
got really confused. It works with /proc/$$/cwd but alas that's not a
stable path between builds. (Less bad than gnu/dammit leaking an
arbitrary build directory into the resulting binaries because gnu, but
still...)
Trying to feed the above build flag into five nested package builds
(gcc, binutils, gmp, mpc, and mpfr) sounds unpleasant even before you
get to the "and we recursively call ourselves in subdirectories"
shenanigans of autoconf and gmake, but lemme see what I can do. (I can
grep for "landley" in the resulting binaries is a good check, I guess...)
Thanks for the heads up,
Rob
_______________________________________________
Toybox mailing list
[email protected]
http://lists.landley.net/listinfo.cgi/toybox-landley.net
