gklib @20230327: error: 'lnlen' may be used uninitialized in this function

[rmottola (Riccardo)]

I suppose it is not a specific PPC issue, but a fact that GCC is being used instead of clang.

 CMakeFiles/GKlib.dir/itemsets.c.o.d -o CMakeFiles/GKlib.dir/itemsets.c.o -c /opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_macports_release_tarballs_ports_math_gklib/gklib/work/GKlib-8bd6bad750b2b0d90800c632cf18e8ee93ad72d7/itemsets.c
/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_macports_release_tarballs_ports_math_gklib/gklib/work/GKlib-8bd6bad750b2b0d90800c632cf18e8ee93ad72d7/io.c: In function 'gk_readfile':
/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_macports_release_tarballs_ports_math_gklib/gklib/work/GKlib-8bd6bad750b2b0d90800c632cf18e8ee93ad72d7/io.c:113:24: error: 'lnlen' may be used uninitialized in this function [-Werror=maybe-uninitialized]
   if (*lineptr == NULL || *n == 0) {
       ~~~~~~~~~~~~~~~~~^~~~~~~~~~

[rmottola (Riccardo)]

what is the best way to reduce warnings in the portfile? Maybe this is a false flag-

[jmroot (Joshua Root)]

I'm not sure if it works to add something like -Wno-error without specifying a warning class. You might have to patch the CMakeLists.txt to not add -Werror in the first place. Of course if the variable really can be used uninitialized, that should be reported upstream.

[rmottola (Riccardo)]

I have this error too. Compilation is being done with gcc7

/opt/local/bin/gcc-mp-7 -DGKlib_EXPORTS -I/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_macports_release_tarballs_ports_math_gklib/gklib/work/GKlib-8bd6bad750b2b0d90800c632cf18e8ee93ad72d7/. -I/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_macports_release_tarballs_ports_math_gklib/gklib/work/GKlib-8bd6bad750b2b0d90800c632cf18e8ee93ad72d7/test -pipe -Os -DNDEBUG -I/opt/local/include -DLINUX -D_FILE_OFFSET_BITS=64 -std=c99 -fno-strict-aliasing -mtune=native -fPIC -Werror -Wall -pedantic -Wno-unused-function -Wno-unused-but-set-variable -Wno-unused-variable -Wno-unknown-pragmas -Wno-unused-label -DNO_X86=ON -DNDEBUG -DNDEBUG2 -DHAVE_EXECINFO_H -O3 -arch ppc -mmacosx-version-min=10.5 -fPIC -MD -MT CMakeFiles/GKlib.dir/io.c.o -MF CMakeFiles/GKlib.dir/io.c.o.d -o CMakeFiles/GKlib.dir/io.c.o -c /opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_macports_release_tarballs_ports_math_gklib/gklib/work/GKlib-8bd6bad750b2b0d90800c632cf18e8ee93ad72d7/io.c
/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_macports_release_tarballs_ports_math_gklib/gklib/work/GKlib-8bd6bad750b2b0d90800c632cf18e8ee93ad72d7/io.c: In function 'gk_readfile':
/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_macports_release_tarballs_ports_math_gklib/gklib/work/GKlib-8bd6bad750b2b0d90800c632cf18e8ee93ad72d7/io.c:113:24: error: 'lnlen' may be used uninitialized in this function [-Werror=maybe-uninitialized]
   if (*lineptr == NULL || *n == 0) {

I doubt this is a PPC only, but more a compiler issue which spits out a warning. Maybe on intel we usually have a different gcc or clang version.

Should maybe warning=errors should be removed or specific initialization warnings suppressed?

[rmottola (Riccardo)]

Out of curiosity I tried using gcc 4.8 (only other available MP compiler there)

gcc-mp-4.8: error: unrecognized command line option '-arch'

[ryandesign (Ryan Carsten Schmidt)]

Please attach the main.log file.

I am not able to reproduce the problem on macOS 12 building using Xcode clang or MacPorts gcc 13. I can't test gcc 7 because it's too old for this OS.

Replying to rmottola:

io.c: In function 'gk_readfile':
io.c:113:24: error: 'lnlen' may be used uninitialized in this function [-Werror=maybe-uninitialized]
   if (*lineptr == NULL || *n == 0) {
       ~~~~~~~~~~~~~~~~~^~~~~~~~~~

gk_readfile uses lnlen only once, at the end of this chunk:

​https://github.com/KarypisLab/GKlib/blob/8bd6bad750b2b0d90800c632cf18e8ee93ad72d7/io.c#L147-L159

char **gk_readfile(char *fname, size_t *r_nlines)
{
  size_t lnlen, nlines=0;
  char *line=NULL, **lines=NULL;
  FILE *fpin;

  gk_getfilestats(fname, &nlines, NULL, NULL, NULL);
  if (nlines > 0) {
    lines = (char **)gk_malloc(nlines*sizeof(char *), "gk_readfile: lines");

    fpin = gk_fopen(fname, "r", "gk_readfile");
    nlines = 0;
    while (gk_getline(&line, &lnlen, fpin) != -1) {

As we see at the top of the function, at this point line is a NULL pointer and lnlen is uninitialized.

gk_getline in turn does this:

​https://github.com/KarypisLab/GKlib/blob/8bd6bad750b2b0d90800c632cf18e8ee93ad72d7/io.c#L101C1-L113

ssize_t gk_getline(char **lineptr, size_t *n, FILE *stream)
{
#ifdef HAVE_GETLINE
  return getline(lineptr, n, stream);
#else
  size_t i;
  int ch;

  if (feof(stream))
    return -1;  

  /* Initial memory allocation if *lineptr is NULL */
  if (*lineptr == NULL || *n == 0) {

The way I read this, in this call, *lineptr (aka line) is NULL, so the second part of the condition checking if *n (aka lnlen which is uninitialized) is 0 is not evaluated, so I don't think the variable is being used uninitialized. You could still report the problem to the developers; it would be simple to just initialize the variable anyway.

HAVE_GETLINE is true on Mac OS X 10.7 or later, so this problem only affects Mac OS X 10.6 and earlier, which is probably why not more people have reported it.

I tried removing these lines from GKlibSystem.cmake (because I could find no way to override this result on the command line):

check_function_exists(getline HAVE_GETLINE)
if(HAVE_GETLINE)
  set(GKlib_COPTIONS "${GKlib_COPTIONS} -DHAVE_GETLINE")
endif(HAVE_GETLINE)

Even so, I did not see a build error on macOS 12 with Xcode clang or MacPorts gcc 13. Maybe this warning was improved in later GCC versions to be more accurate.

The build system sets -Werror only when the compiler is GCC. I submitted a PR to stop doing that:

​https://github.com/KarypisLab/GKlib/pull/36