[Bug] Windows CI builds fail to compile r.flow

[echoix]

Describe the bug

Since last commit 43be353 yesterday r.flow fail on windows CI builds. It also fails the same way for the new PR #3699

I reran the build on the commit and the PR, and still fail at the same place. I'm rerunning again the run on the commit on main.

[echoix]

Here, failing 3 times in a row in the last day https://github.com/OSGeo/grass/actions/runs/9032826748/job/24852093817

[wenzeslaus]

Just a wild guess, but maybe the system includes changed and we are not including memory include explicitly. ``` In file included from D:/a/grass/grass/dist.x86_64-w64-mingw32/include/grass/defs/gis.h:69, 2024-05-11T13:21:16.6173838Z from D:/a/grass/grass/dist.x86_64-w64-mingw32/include/grass/gis.h:725, 2024-05-11T13:21:16.6186814Z from aspect.c:27: 2024-05-11T13:21:16.6195740Z C:\msys64\mingw64\include/sys/stat.h: In function '_fstat64i32': 2024-05-11T13:21:16.6198157Z C:\msys64\mingw64\include/sys/stat.h:80:7: error: implicit declaration of function 'memset' [-Wimplicit-function-declaration] 2024-05-11T13:21:16.6201217Z 80 | memset(_Stat,0,sizeof(struct _stat64i32)); 2024-05-11T13:21:16.6213250Z | ^~~~~~ 2024-05-11T13:21:16.6214461Z C:\msys64\mingw64\include/sys/stat.h:59:1: note: include '<string.h>' or provide a declaration of 'memset' 2024-05-11T13:21:16.6215765Z 58 | #include <_mingw_stat64.h> 2024-05-11T13:21:16.6216473Z +++ |+#include <string.h> 2024-05-11T13:21:16.6217314Z 59 | 2024-05-11T13:21:16.6231741Z In file included from D:/a/grass/grass/dist.x86_64-w64-mingw32/include/grass/segment.h:7, 2024-05-11T13:21:16.6232927Z from r.flow.h:35, 2024-05-11T13:21:16.6233691Z from aspect.c:28: 2024-05-11T13:21:16.6234581Z C:\msys64\mingw64\include/unistd.h: In function 'ftruncate': 2024-05-11T13:21:16.6236423Z C:\msys64\mingw64\include/unistd.h:67:10: error: implicit declaration of function '_chsize'; did you mean '_msize'? [-Wimplicit-function-declaration] 2024-05-11T13:21:16.6238086Z 67 | return _chsize (__fd, __length); 2024-05-11T13:21:16.6238847Z | ^~~~~~~ 2024-05-11T13:21:16.6239445Z | _msize 2024-05-11T13:21:16.6525853Z sed -e ***@***.***@/r.fill.dir/' \ 2024-05-11T13:21:16.6533554Z -e ***@***.***_MAJOR@/8/' \ 2024-05-11T13:21:16.6545704Z -e ***@***.***_MINOR@/4/' \ 2024-05-11T13:21:16.6546812Z -e ***@***.***_RELEASE@/0dev/' \ 2024-05-11T13:21:16.6597583Z ../../mswindows/generic.manifest.rc.in > OBJ.x86_64-w64-mingw32/r.fill.dir.manifest.rc 2024-05-11T13:21:16.6718991Z make[4]: *** [../../include/Make/Compile.make:32: OBJ.x86_64-w64-mingw32/aspect.o] Error 1 2024-05-11T13:21:16.6720135Z make[4]: Leaving directory '/d/a/grass/grass/raster/r.flow' 2024-05-11T13:21:16.6768033Z make -C r.geomorphon || echo /d/a/grass/grass/raster/r.geomorphon >> /d/a/grass/grass/error.log 2024-05-11T13:21:16.7146242Z sed ***@***.***@/r.fill.dir/' ../../mswindows/generic.manifest > OBJ.x86_64-w64-mingw32/r.fill.dir.exe.manifest 2024-05-11T13:21:16.7288490Z make[4]: Entering directory '/d/a/grass/grass/raster/r.geomorphon' 2024-05-11T13:21:16.7706907Z windres --input=OBJ.x86_64-w64-mingw32/r.fill.dir.manifest.rc --input-format=rc --output=OBJ.x86_64-w64-mingw32/r.fill.dir.manifest.res --output-format=coff -I../../mswindows 2024-05-11T13:21:16.7945457Z test -d OBJ.x86_64-w64-mingw32 || mkdir -p OBJ.x86_64-w64-mingw32 ```

[echoix]

I see in the msys2 changelog that the toolchain has been upgraded from GCC 13.2 to GCC 14.1. I will continue to read on what that implies

[echoix]

I tried yesterday enabling extra errors in the Ubuntu builds to match the new errors that are enabled by default in GCC 14 (not gcc 14 only ones), but the jobs didn't fail...

Is there anyone else that runs a rolling release and already has GCC 14 (released a couple days ago) that can confirm if it only windows or on Linux too?

I know that at least a #include <string.h> is needed somewhere, but can't decide the best place for it.

[wenzeslaus]

...but can't decide the best place for it.

I think it simply is: When you need something defined in header A, then include header A even if you are also including header B which include header A (i.e., do not rely on transitive includes).

In this case, it seems like something in r.flow uses memcpy from string.h without including it. Now, when I can better read the message (on desktop), I can see that the compiler provides a helpful message (at least it looks like that):

...note: include '<string.h>' or provide a declaration of 'memset'

On the other hand, the error seems to come from include/sys/stat.h (a system file) not GRASS code:

C:\msys64\mingw64\include/sys/stat.h:80:7: error: implicit declaration of function 'memset'

[nilason]

This is indeed a result of the update to GCC 14. Temporary workaround may be to set -Wno-error=implicit-function-declaration.