Tree clean-up & fixes before 4.5.0 #24
Conversation
this was not done for newlib-xtensa sources since newlib-1.19
amends c0f3596 using dummy implementation already used in esp8266/Arduino
remove #if 0 from newlib string.h implementations, explicitly list files that need overrides amends 42f6e16
Change newlib header to allow redefinitions instead of implicitly requiring sys/pgmspace.h
Keeps newlib sys/stdio.h, avoid c/p its code
Remove unused sys/xtensa/include/config.h Update sys/config.h instead
Redundant and unlike AVR this is a plain type w/o any extra attributes
don't interpret pointer value as float or double sync with sys/xtensa/sys/pgmspace.h & don't swallow original strings
Don't link the whole .o when only one function is used
| -H "Authentication: ${{ secrets.GITHUB_TOKEN }}" \ | ||
| -H "Accept: application/vnd.github.VERSION.raw" \ | ||
| -o "$flocal" \ | ||
| https://api.github.com/repos/esp8266/Arduino/contents/$fremote |
There was a problem hiding this comment.
I don't think we need any authentication here. This is only ever getting the master version of the 2 files, no?
There was a problem hiding this comment.
It is not technically required, it is helpful for cases when CI breaks and has to be re-run multiple times. Otherwise, user has to wait for an hour to continue after performing 60 requests.
- https://api.github.com/rate_limit
- https://docs.github.com/en/rest/rate-limit/rate-limit?apiVersion=2022-11-28#get-rate-limit-status-for-the-authenticated-user
I also c/p the request from my other incorrectly written workflow, it is actually authorization: bearer $TOKEN not authenticate: $TOKEN
|
Nice work, cleans things up a lot. Did you decide to just punt on printf-nano and go full-fat for 4.5.0? |
Same as `gh-api`, allow 5000 vs. 60 api requests per hour ref. https://docs.github.com/en/rest/rate-limit/rate-limit
Reverting 3b97a5e plus this patch to move ssputws_r and sswprint_r out of nano-formatted-io if-else seems ok diff --git a/newlib/libc/stdio/Makefile.inc b/newlib/libc/stdio/Makefile.inc
index e25680212..09a508928 100644
--- a/newlib/libc/stdio/Makefile.inc
+++ b/newlib/libc/stdio/Makefile.inc
@@ -39,9 +39,7 @@ libc_a_SOURCES += \
%D%/sprint_r.c \
%D%/swprint_r.c \
%D%/ssputs_r.c \
- %D%/ssputws_r.c \
- %D%/ssprint_r.c \
- %D%/sswprint_r.c
+ %D%/ssprint_r.c
endif
libc_a_SOURCES += \
@@ -99,6 +97,8 @@ libc_a_SOURCES += \
%D%/snprintf.c \
%D%/sprintf.c \
%D%/sscanf.c \
+ %D%/ssputws_r.c \
+ %D%/sswprint_r.c \
%D%/stdio.c \
%D%/svfiwprintf.c \
%D%/svfiwscanf.c \
--
2.43.0After patch build files list matches 4.0.0, so it should build exactly the same result. Fingers crossed, nothing else breaks, given that it still builds & links No idea how this is used in the wild w/ anyone else's esp8266 builds, btw. wchar_t is necessary for "C99 support" label, but not sure how much of it is actually used. libstdc++-v3 linkage is a nasty side-effect, as library tries to implicitly create various objects using wchar_t. e.g. There is also There is also https://github.com/picolibc/picolibc which is based on newlib, but it also includes tinystdio port that fully supports wide chars. tbh that might be even better, just have to rebase it yet again |
mcspr
force-pushed
the
xtensa-4_0_0-lock-arduino
branch
2 times, most recently
from
49ac679 to
07c6da0
Compare
|
It's a holiday week here and so can't dig too much into this, but looks good at a glance, still. I found some time formatting issues also WRT nano-printf. strftime or something similar, would need to see if I have the logs. There was one example that formatted a date using it that also failed to link in CI w/nano-printf. AdvancedWebServer, maybe? It was a WebServer example for sure... |
sync w/ picolibc va_list typing (which is often confusing for clang, too) sys/_stdio.h & others, editor should see stdarg.h dependency
Same as memmove_P check, but for libc func Note that esp8266/Arduino currently re-maps ROM functions > PROVIDE ( strcpy = 0x4000bec8 ); > PROVIDE ( strncpy = 0x4000c0a0 ); But, both are still available via `ets_...`
Another c/p of existing source
4.0.0 build system does not prepend library name to outputs in libc/sys/ 4.5.0 does (but, libc/machine is preferred) move atexit.c & previously missing time.c amends e9fa704
mcspr
force-pushed
the
xtensa-4_0_0-lock-arduino
branch
from
07c6da0 to
961364f
Compare
|
btw sorry for force-pushes above. I intended to work on a test branch, but it has gone horribly wrong... From public arduino-pico builds I've seen at least one erroneous PR run. Both arm & riscv fail with UploadHudeFile.ino And it is also related to Using the same idea, I've repackaged 230224 with lib{c,g,m}.a and Since WebServer still uses regex.h and not C++ header, instead I rigged Happy holidays :) |
|
If you checked the sources here, pls also take a look at merge & release branches for 4.5.0 I could put up it as PR, but these probably could be just fetched manually. nb. for some weird reason, merging one version tag at a time is much easier than attempting to jump from 4.0.0 to 4.5.0 |
2 participants
Hiding these behind a merge-commit does not seem sensible, and I don't want to lose history info. Hope messages below are pretty self-descriptive.
CI is updated to reference up-to-date toolchain and produces a .zip with filesystem after
make installedit: https://github.com/mcspr/newlib-xtensa/actions/runs/15948100054/job/44984645502
Functional changes are limited to splitting
*_pgmspace.c& str{,n}cpy into separate .c to provide .o per function and avoiding messing with the newlib sources too muchRest mostly deals with crusted over automake/conf sources, port file overrides, header chain order tweaks, removing some unused files coming from original patch and correcting files that were changed after upstream merges over the years.
ps. would worth a toolchain bump, also including the 'always-inline' commit for sys/pgmspace.h
could also push for 4.5.0, since I don't really plan on changing that much as I thought originally
ps2. newlib-4.0.0 used ancient autoconf-2.64 and automake-1.11 (which btw ubuntu ships w/ a broken debian patch even to this day)