CI: Update to ubuntu-26.04 by hdiethelm · Pull Request #4477 · LinuxCNC/linuxcnc

hdiethelm · 2026-08-26T21:28:48Z

This solves an issue with po4a that does not translates all the document, see: #4471 (comment)

However, some build jobs fail, WIP:

~~cppcheck~~ Fixed
~~clang build~~ Fixed
tests:
- interp/g71-endless-loop
- interp/g71-endless-loop_2
- interp/g71-with-g70

This solves an issue with po4a that does not translates all the documents. However, some build jobs fail now and will be fixed in the following commits.

BsAtHome · 2026-08-26T21:32:01Z

 //no more used since menu/toolbar added...
-#ifdef AAAAAAAAAAAAAAAAAAAAAA
+#if false
 	hboxbottom = gtk_hbox_new (FALSE,0);


The standard method is #if 0. There is usually no "false" defined and then you could just as well keep it as AAAAAAAAAAAAAAAAAAAAAA.

Thanks, fixed. The AAAAAAAAAAAAAAAAAAAAAA seams to create issues with the new cppcheck that checks also the AAAAAAAAAAAAAAAAAAAAAA case which has a (void*)1 in it that is not worth inspecting and fixing in unused code.

BsAtHome · 2026-08-26T21:32:40Z

+LDFLAGS := -L$(LIB_DIR) -Wl,-rpath,$(LIB_DIR) $(LIBTIRPC_LIBS) $(LDFLAGS) -latomic
 else
-LDFLAGS := -Wl,-rpath-link,../lib $(LIBTIRPC_LIBS) $(LDFLAGS)
+LDFLAGS := -Wl,-rpath-link,../lib $(LIBTIRPC_LIBS) $(LDFLAGS) -latomic
 endif


The question is why is this suddenly necessary?
Also, this only seems to be necessary with clang.

The error is:

Linking rtapi_app /usr/lib/gcc/x86_64-linux-gnu/15/../../../../include/c++/15/bits/atomic_base.h:999:(.text+0x3cd5): undefined reference to `__atomic_compare_exchange' clang++ -std=gnu++20 -rdynamic -o ../bin/rtapi_app objects/rtapi/uspace_rtapi_main.o objects/rtapi/uspace_rtapi_app.o objects/rtapi/uspace_rtapi_parport.o objects/rtapi/uspace_rtapi_string.o objects/rtapi/rtapi_pci.o -pthread -lrt -lfmt -ludev -ldl -lcap -L/home/runner/work/linuxcnc/linuxcnc/lib -Wl,-rpath,/home/runner/work/linuxcnc/linuxcnc/lib -ltirpc -lgpiod /usr/bin/x86_64-linux-gnu-ld.bfd: /usr/lib/gcc/x86_64-linux-gnu/15/../../../../include/c++/15/bits/atomic_base.h:999:(.text+0x3d44): undefined reference to `__atomic_compare_exchange' /usr/bin/x86_64-linux-gnu-ld.bfd: objects/rtapi/uspace_rtapi_main.o: in function `boost::lockfree::detail::fixed_size_freelist<boost::lockfree::queue<message_t, boost::lockfree::capacity<128ul> >::node, boost::lockfree::detail::compiletime_sized_freelist_storage<boost::lockfree::queue<message_t, boost::lockfree::capacity<128ul> >::node, 129ul> >::deallocate_impl(unsigned short)': /usr/lib/gcc/x86_64-linux-gnu/15/../../../../include/c++/15/bits/atomic_base.h:999:(.text+0x3d8d): undefined reference to `__atomic_compare_exchange' /usr/bin/x86_64-linux-gnu-ld.bfd: /usr/lib/gcc/x86_64-linux-gnu/15/../../../../include/c++/15/bits/atomic_base.h:999:(.text+0x3dcf): undefined reference to `__atomic_compare_exchange' /usr/bin/x86_64-linux-gnu-ld.bfd: objects/rtapi/uspace_rtapi_main.o: in function `bool std::__atomic_impl::__compare_exchange<false, boost::lockfree::detail::tagged_index>(boost::lockfree::detail::tagged_index&, std::remove_volatile<boost::lockfree::detail::tagged_index>::type&, std::remove_volatile<boost::lockfree::detail::tagged_index>::type&, bool, std::memory_order, std::memory_order)': /usr/lib/gcc/x86_64-linux-gnu/15/../../../../include/c++/15/bits/atomic_base.h:999:(.text+0x40df): undefined reference to `__atomic_compare_exchange' /usr/bin/x86_64-linux-gnu-ld.bfd: objects/rtapi/uspace_rtapi_main.o:/usr/lib/gcc/x86_64-linux-gnu/15/../../../../include/c++/15/bits/atomic_base.h:999: more undefined references to `__atomic_compare_exchange' follow

So it is most probably boost::lockfree that needs it, could be a boost change that behaves differently with clang. Commit is ready that adds this flag only to rtapi_app.

https://en.cppreference.com/cpp/atomic/atomic
Quote: On GCC and Clang, some of the functionality described here requires
linking against -latomic.

Hm... time to kill boost then, I guess. Moving to pybind11 would probably be the best way to go.

Why pybind? It is this one:

linuxcnc/src/rtapi/uspace_rtapi_main.cc

Line 403 in d47bccc

static boost::lockfree::queue<message_t, boost::lockfree::capacity<128>> rtapi_msg_queue;

Could be rewritten using atomics or an other library can be used but not in this PR.

My mistake, I thought it came from gcodemodule (because it also uses boost).

Anyway, it is not right to add -latomic globally. Only the one failing should require it. And, yes, if it can be changed not to be a problem anymore, then that would actually be better.

I still do not understand how/why this fails now. Only if the size of the exchange exceeded an instruction size, but then the lockfree would be history too.

If it the dependency can be removed, then that is even better.

BsAtHome · 2026-08-26T21:36:07Z

 #ifdef TOOL_NML //{
-  tool_nml_register((CANON_TOOL_TABLE*)& _sai._tools);
+  tool_nml_register(& _sai._tools);
 #else //}{


Why doesn't this trip in emc/task/taskclass.cc, emc/usr_intf/axis/extensions/emcmodule.cc and emc/usr_intf/halui.cc?

& _sai._tools is already type CANON_TOOL_TABLE, no need to cast it seams.

Yes, but why does cppcheck accept the same "old style cast" in those other files?

Where do you see one that is not a cast of NULL? Looks like either casting NULL is allowed or #ifdef / #else is not reached. Anyway, the old cppcheck also did not complain.

BsAtHome · 2026-08-26T21:41:10Z

 CPPCHKOPT=( -j "$nproc" --force "$EXHAUSTIVE" --inline-suppr )
 CPPCHKOPT+=( "--enable=warning,performance,portability" )
 CPPCHKOPT+=( "-I$(realpath "$(dirname "$0")/../include")" )
+CPPCHKOPT+=( "-I$(realpath "$(dirname "$0")/../src/rtapi")" ) #Needed for KERNEL_VERSION and MODULE_VERBOSE macros


Which header is involved?

It is probably not a good idea to have this added for all files. We are trying to isolate public/private headers and the include/search order is important.

It is rtapi.h included from hal_lib.c:

Checking hal/hal_lib_query.c: RTAPI... hal/hal_lib.c:3477:2: error: failed to evaluate #if condition, undefined function-like macro invocation: KERNEL_VERSION( ... ) [syntaxError] #if defined(__KERNEL__) && defined( CONFIG_PROC_FS ) && LINUX_VERSION_CODE < KERNEL_VERSION(3,10,0) ^

And also an issue in mb2hal.c with MODULE_VERBOSE that got solved adding this line but I did not verify why. The error was:

Checking hal/user_comps/mb2hal/mb2hal_hal.c: MODULE_VERBOSE... hal/user_comps/mb2hal/mb2hal.h:36:1: error: syntax error [syntaxError] Checking hal/user_comps/mb2hal/mb2hal_hal.c: RTAPI... MODULE_VERBOSE(emc2, "component:mb2hal:Userspace HAL component to communicate with one or more Modbus devices"); ^

It seems that there is a wrong assumption here because, f.ex., hal_lib.c needs to compile once as a user-space library and once as a RT library. These have different points of view. That means that the error must be dealt with locally and the real cause must be found. There is probably an assumption in there that cppcheck chokes on.

Hmm, it looks like with adding this include path, cppckeck does not check any more the MODULE_VERBOSE case.

I greped the whole repo, /usr/include, /usr/src and /usr/realtime-: MODULE_VERBOSE exists only in mb2hal.h. Can this this might be just be deleted? Looks to me like a long gone makro.

cppcheck passes now. Replace #ifdef AAAAA with #if 0 so cppcheck does not check unused code.

clang needs -latomic: See https://en.cppreference.com/cpp/atomic/atomic Quote: On GCC and Clang, some of the functionality described here requires linking against -latomic.

CI: Update to ubuntu-26.04

cba7a95

This solves an issue with po4a that does not translates all the documents. However, some build jobs fail now and will be fixed in the following commits.

hdiethelm mentioned this pull request Aug 26, 2026

ci: Enable fail on htmlcheck error #4471

Draft

hdiethelm marked this pull request as draft August 26, 2026 21:31

BsAtHome reviewed Aug 26, 2026

View reviewed changes

hdiethelm force-pushed the ci_update_ubuntu_26.04 branch from 28dbfc4 to e668eb2 Compare August 26, 2026 21:56

hdiethelm added 2 commits August 27, 2026 00:14

CI: Update to ubuntu-26.04: cppcheck

156851d

cppcheck passes now. Replace #ifdef AAAAA with #if 0 so cppcheck does not check unused code.

CI: Update to ubuntu-26.04: clang build

a5d5858

clang needs -latomic: See https://en.cppreference.com/cpp/atomic/atomic Quote: On GCC and Clang, some of the functionality described here requires linking against -latomic.

hdiethelm force-pushed the ci_update_ubuntu_26.04 branch from e668eb2 to a5d5858 Compare August 26, 2026 22:16

Uh oh!

Conversation

hdiethelm commented Aug 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

hdiethelm commented Aug 26, 2026 •

edited

Loading