Build failures on RTEMS 7

[kiwichris]

I am wanting to test #61 with RTEMS 7. I am seeing build errors. They are:

../devIocStatsAnalog.c:286:37: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(int)' [-Wincompatible-pointer-types]
  286 | aStats devAiStats = {6,       NULL, ai_init, ai_init_record, ai_ioint_info,
      |                                     ^~~~~~~
../devIocStatsAnalog.c:286:37: note: (near initialization for 'devAiStats.init')
../devIocStatsAnalog.c:180:13: note: 'ai_init' declared here
  180 | static long ai_init(int pass);
      |             ^~~~~~~
../devIocStatsAnalog.c:286:46: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(aiRecord *)' [-Wincompatible-pointer-types]
  286 | aStats devAiStats = {6,       NULL, ai_init, ai_init_record, ai_ioint_info,
      |                                              ^~~~~~~~~~~~~~
../devIocStatsAnalog.c:286:46: note: (near initialization for 'devAiStats.init_record')
../devIocStatsAnalog.c:181:13: note: 'ai_init_record' declared here
  181 | static long ai_init_record(aiRecord *);
      |             ^~~~~~~~~~~~~~
../devIocStatsAnalog.c:286:62: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(int,  aiRecord *, struct ioscan_head **)' [-Wincompatible-pointer-types]
  286 | aStats devAiStats = {6,       NULL, ai_init, ai_init_record, ai_ioint_info,
      |                                                              ^~~~~~~~~~~~~
../devIocStatsAnalog.c:286:62: note: (near initialization for 'devAiStats.get_ioint_info')
../devIocStatsAnalog.c:183:13: note: 'ai_ioint_info' declared here
  183 | static long ai_ioint_info(int cmd, aiRecord *pr, IOSCANPVT *iopvt);
      |             ^~~~~~~~~~~~~
../devIocStatsAnalog.c:287:22: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(aiRecord *)' [-Wincompatible-pointer-types]
  287 |                      ai_read, NULL};
      |                      ^~~~~~~
../devIocStatsAnalog.c:287:22: note: (near initialization for 'devAiStats.read_write')
../devIocStatsAnalog.c:182:13: note: 'ai_read' declared here
  182 | static long ai_read(aiRecord *);
      |             ^~~~~~~
../devIocStatsAnalog.c:289:37: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(aoRecord *)' [-Wincompatible-pointer-types]
  289 | aStats devAoStats = {6, NULL, NULL, ao_init_record, NULL, ao_write, NULL};
      |                                     ^~~~~~~~~~~~~~
../devIocStatsAnalog.c:289:37: note: (near initialization for 'devAoStats.init_record')
../devIocStatsAnalog.c:189:13: note: 'ao_init_record' declared here
  189 | static long ao_init_record(aoRecord *pr);
      |             ^~~~~~~~~~~~~~
../devIocStatsAnalog.c:289:59: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(aoRecord *)' [-Wincompatible-pointer-types]
  289 | aStats devAoStats = {6, NULL, NULL, ao_init_record, NULL, ao_write, NULL};
      |                                                           ^~~~~~~~
../devIocStatsAnalog.c:289:59: note: (near initialization for 'devAoStats.read_write')
../devIocStatsAnalog.c:190:13: note: 'ao_write' declared here
  190 | static long ao_write(aoRecord *);
      |             ^~~~~~~~
../devIocStatsAnalog.c:292:14: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(int)' [-Wincompatible-pointer-types]
  292 |     6, NULL, ai_clusts_init, ai_clusts_init_record, NULL, ai_clusts_read, NULL};
      |              ^~~~~~~~~~~~~~
../devIocStatsAnalog.c:292:14: note: (near initialization for 'devAiClusts.init')
../devIocStatsAnalog.c:185:13: note: 'ai_clusts_init' declared here
  185 | static long ai_clusts_init(int pass);
      |             ^~~~~~~~~~~~~~
../devIocStatsAnalog.c:292:30: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(aiRecord *)' [-Wincompatible-pointer-types]
  292 |     6, NULL, ai_clusts_init, ai_clusts_init_record, NULL, ai_clusts_read, NULL};
      |                              ^~~~~~~~~~~~~~~~~~~~~
../devIocStatsAnalog.c:292:30: note: (near initialization for 'devAiClusts.init_record')
../devIocStatsAnalog.c:186:13: note: 'ai_clusts_init_record' declared here
  186 | static long ai_clusts_init_record(aiRecord *);
      |             ^~~~~~~~~~~~~~~~~~~~~
../devIocStatsAnalog.c:292:59: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(aiRecord *)' [-Wincompatible-pointer-types]
  292 |     6, NULL, ai_clusts_init, ai_clusts_init_record, NULL, ai_clusts_read, NULL};
      |                                                           ^~~~~~~~~~~~~~
../devIocStatsAnalog.c:292:59: note: (near initialization for 'devAiClusts.read_write')
../devIocStatsAnalog.c:187:13: note: 'ai_clusts_read' declared here
  187 | static long ai_clusts_read(aiRecord *);
      |             ^~~~~~~~~~~~~~

../devIocStatsString.c:189:14: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(int)' [-Wincompatible-pointer-types]
  189 |     5, NULL, stringin_init, stringin_init_record, NULL, stringin_read};
      |              ^~~~~~~~~~~~~
../devIocStatsString.c:189:14: note: (near initialization for 'devStringinStats.init')
../devIocStatsString.c:139:13: note: 'stringin_init' declared here
  139 | static long stringin_init(int pass);
      |             ^~~~~~~~~~~~~
../devIocStatsString.c:189:29: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(stringinRecord *)' [-Wincompatible-pointer-types]
  189 |     5, NULL, stringin_init, stringin_init_record, NULL, stringin_read};
      |                             ^~~~~~~~~~~~~~~~~~~~
../devIocStatsString.c:189:29: note: (near initialization for 'devStringinStats.init_record')
../devIocStatsString.c:140:13: note: 'stringin_init_record' declared here
  140 | static long stringin_init_record(stringinRecord *);
      |             ^~~~~~~~~~~~~~~~~~~~
../devIocStatsString.c:189:57: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(stringinRecord *)' [-Wincompatible-pointer-types]
  189 |     5, NULL, stringin_init, stringin_init_record, NULL, stringin_read};
      |                                                         ^~~~~~~~~~~~~
../devIocStatsString.c:189:57: note: (near initialization for 'devStringinStats.read_stringin')
../devIocStatsString.c:141:13: note: 'stringin_read' declared here
  141 | static long stringin_read(stringinRecord *);
      |             ^~~~~~~~~~~~~
../devIocStatsString.c:190:53: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(stringinRecord *)' [-Wincompatible-pointer-types]
  190 | sStats devStringinEnvVar = {5,    NULL,       NULL, envvar_init_record,
      |                                                     ^~~~~~~~~~~~~~~~~~
../devIocStatsString.c:190:53: note: (near initialization for 'devStringinEnvVar.init_record')
../devIocStatsString.c:142:13: note: 'envvar_init_record' declared here
  142 | static long envvar_init_record(stringinRecord *);
      |             ^~~~~~~~~~~~~~~~~~
../devIocStatsString.c:191:35: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(stringinRecord *)' [-Wincompatible-pointer-types]
  191 |                             NULL, envvar_read};
      |                                   ^~~~~~~~~~~
../devIocStatsString.c:191:35: note: (near initialization for 'devStringinEnvVar.read_stringin')
../devIocStatsString.c:143:13: note: 'envvar_read' declared here
  143 | static long envvar_read(stringinRecord *);
      |             ^~~~~~~~~~~
../devIocStatsString.c:192:43: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(stringinRecord *)' [-Wincompatible-pointer-types]
  192 | sStats devStringinEpics = {5, NULL, NULL, epics_init_record, NULL, epics_read};
      |                                           ^~~~~~~~~~~~~~~~~
../devIocStatsString.c:192:43: note: (near initialization for 'devStringinEpics.init_record')
../devIocStatsString.c:144:13: note: 'epics_init_record' declared here
  144 | static long epics_init_record(stringinRecord *);
      |             ^~~~~~~~~~~~~~~~~
../devIocStatsString.c:192:68: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(stringinRecord *)' [-Wincompatible-pointer-types]
  192 | sStats devStringinEpics = {5, NULL, NULL, epics_init_record, NULL, epics_read};
      |                                                                    ^~~~~~~~~~
../devIocStatsString.c:192:68: note: (near initialization for 'devStringinEpics.read_stringin')
../devIocStatsString.c:145:13: note: 'epics_read' declared here
  145 | static long epics_read(stringinRecord *);
      |             ^~~~~~~~~~

../devIocStatsWaveform.c:120:14: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(int)' [-Wincompatible-pointer-types]
  120 |     5, NULL, waveform_init, waveform_init_record, NULL, waveform_read};
      |              ^~~~~~~~~~~~~
../devIocStatsWaveform.c:120:14: note: (near initialization for 'devWaveformStats.init')
../devIocStatsWaveform.c:103:13: note: 'waveform_init' declared here
  103 | static long waveform_init(int pass);
      |             ^~~~~~~~~~~~~
../devIocStatsWaveform.c:120:29: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(waveformRecord *)' [-Wincompatible-pointer-types]
  120 |     5, NULL, waveform_init, waveform_init_record, NULL, waveform_read};
      |                             ^~~~~~~~~~~~~~~~~~~~
../devIocStatsWaveform.c:120:29: note: (near initialization for 'devWaveformStats.init_record')
../devIocStatsWaveform.c:104:13: note: 'waveform_init_record' declared here
  104 | static long waveform_init_record(waveformRecord *);
      |             ^~~~~~~~~~~~~~~~~~~~
../devIocStatsWaveform.c:120:57: error: initialization of 'long int (*)(void)' from incompatible pointer type 'long int (*)(waveformRecord *)' [-Wincompatible-pointer-types]
  120 |     5, NULL, waveform_init, waveform_init_record, NULL, waveform_read};
      |                                                         ^~~~~~~~~~~~~
../devIocStatsWaveform.c:120:57: note: (near initialization for 'devWaveformStats.read_waveform')
../devIocStatsWaveform.c:105:13: note: 'waveform_read' declared here
  105 | static long waveform_read(waveformRecord *);
      |             ^~~~~~~~~~~~~

[simon-ess]

What version c standard are you using? See #65 for a similar issue.

[kiwichris]

What version c standard are you using?

RTEMS 7 is on gcc 15 and default language standard is C23 for C and C++20 for C++.

See #65 for a similar issue.

I am sorry for raising a second report on this. I did look and did not think that issue was related.

I searched epics-base for solutions as DEVSUPFUN is defined there and borrowing the following has iocStats building:

diff --git a/devIocStats/devIocStatsAnalog.c b/devIocStats/devIocStatsAnalog.c
index 62d3c53..b38a7ce 100644
--- a/devIocStats/devIocStatsAnalog.c
+++ b/devIocStats/devIocStatsAnalog.c
@@ -283,13 +283,16 @@ static validGetParms statsGetParms[] = {
     {"cbHighQueueOverruns", statsCbHighQOverruns, QUEUE_TYPE},
     {NULL, NULL, 0}};

-aStats devAiStats = {6,       NULL, ai_init, ai_init_record, ai_ioint_info,
-                     ai_read, NULL};
+aStats devAiStats = {6,       NULL, (DEVSUPFUN) ai_init,
+                     (DEVSUPFUN) ai_init_record, (DEVSUPFUN) ai_ioint_info,
+                     (DEVSUPFUN) ai_read, NULL};
 epicsExportAddress(dset, devAiStats);
-aStats devAoStats = {6, NULL, NULL, ao_init_record, NULL, ao_write, NULL};
+aStats devAoStats = {6, NULL, NULL, (DEVSUPFUN) ao_init_record, NULL,
+                     (DEVSUPFUN) ao_write, NULL};
 epicsExportAddress(dset, devAoStats);
 aStats devAiClusts = {
-    6, NULL, ai_clusts_init, ai_clusts_init_record, NULL, ai_clusts_read, NULL};
+    6, NULL, (DEVSUPFUN) ai_clusts_init, (DEVSUPFUN) ai_clusts_init_record,
+    NULL, (DEVSUPFUN) ai_clusts_read, NULL};
 epicsExportAddress(dset, devAiClusts);

 static memInfo meminfo = {0.0, 0.0, 0.0, 0.0, 0.0, 0.0};

Plus the same in a couple of other files. I can generate a PR if needed?

[anjohnson]

@kiwichris The C templates used for most EPICS device and drivers support code were developed a long time ago and those originals (including the DEVSUPFUN and DRVSUPFUN macros) are no longer valid C by the latest standards. Your changes above aren't really how I would like these to be fixed. We have modified the device support files in epics-base to a good solution, but many external support modules like this one haven't been converted yet — the EPICS community doesn't generally work with the bleeding-edge compilers so we have quite a lot of technical debt like this.

My comments in issue #65 which referred to the "Type-safe Device and Driver Support Tables" section of the 3.16.2 Release Notes provide the basic outline for the changes, but a combination of the changes in this commit and this later one to the ipac driver are probably the best way to reach the preferred format.

I think there's a tool (Cochinelle?) that can do significant mechanical reformatting of C code as is needed here (I'm sure AI could be used similarly) but I've never tried it and I'm not sure if it's available for free; if it was, it could probably be applied to many EPICS support modules to save lots of work for the community.

[kiwichris]

Thanks and I understand you not wanting to bury these issues as I hacked. I did this to be able to continue testing so we can get these outstanding PRs merged.

RTEMS tracks recent GCC versions and GCC release have started to move the base language standard so GCC 15 is C17 which RTEMS 7 uses and RTEMS 8 will be on GCC 16 on C23 and that is even harder. @joelsherrill keeps us up to date as a community service to GCC for embedded work.

[joelsherrill]

My starting advice is to consciously pick the language version (C, C++, etc) that you will use and specify -std= on the command line. This at least avoids introduction of issues when a new GCC version changes the default. It is likely that EPICS uses a wide enough range of OSes that the default varies. This shows the defaults for a range of GCC versions. Note that gnu instead of c or c++ indicates that GNU extensions are allowed and more APIs are enabled. -std=c17 would strictly follow the C Language Standard and limit you to the library without POSIX.

• GCC 15: gnu23, gnu++17
• GCC 11 – 14: gnu17, gnu++17
• GCC 6 – 10: gnu11, gnu++14
• GCC 5: gnu11, gnu++98
• GCC 4.9 & older, gnu90, gnu++98

If it helps at all, C17 is just C11 with bug fixes and C++14 is C++11 with bug fixes. C++17 and C23 have feature additions.

EPICS also needs to consider what the oldest compiler/OS to be supported is. That sets the floor. For RTEMS 7, that is around gcc 12 on just the Microblaze, 14 for the nios2, and 15 for everything else.

The newer language versions tend to clarify undefined and under-defined cases which leads to new warnings, Fixing those warnings normally will not break builds on older language versions. At least I don't remember an example of this.

For RTEMS, I addressed 1000s of warnings over the course of months. Many were the same mistake over and over which was easy to fix. I tended to try to fix all warnings of a particular type and not worry if all warnings were gone in a file. Some were lurking bugs and required help. I started with -Wall with some disabled so I could turn on -Werror. I worked to fix the disabled warnings, then added -Wextra and similarly disabled some warnings which took time to resolve. Today we build with -Wall -Wextra -Werror and many -pedantic warnings are also resolved.

This required building RTEMS for all BSPs with multiple settings. Did the warning show up only in debug mode? etc.

Critically, you need to really fix the warning. Do not paper over it with a cast or something else questionable.

In theory, now that -Werror is user facing, we shouldn't get submissions with warnings. This was an important bridge to cross. Otherwise, warnings return over time. I have fought this before multiple times.

There is no good time to tackle this. I decided to fix what I could before 7 branched and the GCC 15 moving to C23 was the trigger for me.

For EPICS, I would recommend picking a base language version (e.g. set -std=XXX) for the entire project with overrides if needed for specific modules. Try -Wall and see how bad it is on a few core modules. Grow the set of modules and warnings. Encourage module maintainers to do the same.

It might make for a good EPICS Hackathon theme. I am available at least for guidance.