Conversation
376ba27 to
d422a9a
Compare
|
Maybe just add |
|
And maybe |
a1cc392 to
60815d2
Compare
|
Murdock is not happy yet. |
|
Maybe we should split out the thread size configuration into a separate header, so that IMO, C++ code should never see any low level headers and we just provide C++ compatibility for the public API. That way, C++ becomes quite a bit less pain in the ass to maintain. |
Yes! I have a commit somewhere doing that. I'm on holidays this week, will dig out when back. |
I now remembered why I suggested this: The issue is that |
This has bitten me on multiple occasions already. |
... like, the inlined thread_yield_higher() in thread_arch.h also depends on cpu_conf.h ... |
|
Is there a fix for: |
Yes, not including the system headers from C++ :-) Sadly, this is not so trivial. It would be best if we could avoid including vendor files from "public" header files (directly or indirectly) altogether, so that only the C source in |
```
error: register allocation failed: maximum depth for recoloring reached. Use -fexhaustive-register-search to skip cutoffs
In file included from /home/kaspar/src/riot/build/pkg/micro-ecc/uECC.c:182:
/home/kaspar/src/riot/build/pkg/micro-ecc/asm_arm.inc:62:9: error: inline assembly requires more registers than available
".syntax unified \n\t"
^
```
60815d2 to
8e8c24a
Compare
|
I'm not going to work on this anymore. |
|
Oh, it has been fixed anyway in the meantime |
|
But it seems murdock is still not building some boards with llvm: https://github.com/RIOT-OS/RIOT/blob/master/.murdock#L40 |
It has been re-enabled in 52cf2b4 - I forgot to update the comment above, though. That only a subset of the boards are build with LLVM is intentional to avoid increasing CI time by factor 2. I dropped the obsolete comment and added a bit of doc in #20482 |
Contribution description
This PR bundles some llvm changes necessary to build with llvm 11.
Testing procedure
Issues/PRs references