Build warnings in the Windows Meson build we should fix

[rgommers]

@Smit-create here are all the relevant build warnings from today's Windows build against numpy 1.21.5. Could you look into these?

[325/1557] Compiling C object scipy/sparse/linalg/_dsolve/libsuperlu_lib.a.p/SuperLU_SRC_cmemory.c.obj
../scipy/sparse/linalg/_dsolve/SuperLU/SRC/cmemory.c: In function 'cStackCompress':
../scipy/sparse/linalg/_dsolve/SuperLU/SRC/cmemory.c:649:24: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast]
  649 |     Glu->stack.used -= (long int) fragment;
      |                        ^
../scipy/sparse/linalg/_dsolve/SuperLU/SRC/cmemory.c:650:24: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast]
  650 |     Glu->stack.top1 -= (long int) fragment;
      |                        ^

The above one also happens in dStackCompress and the s/z variants.

[357/1557] Compiling C++ object scipy/special/_ufuncs_cxx.cp39-win_amd64.pyd.p/meson-generated__ufuncs_cxx.cpp.obj
scipy/special/_ufuncs_cxx.cp39-win_amd64.pyd.p/_ufuncs_cxx.cpp:659: warning: "_USE_MATH_DEFINES" redefined
  659 |   #define _USE_MATH_DEFINES
      | 
<command-line>: note: this is the location of the previous definition
In file included from C:\hostedtoolcache\windows\Python\3.9.9\x64\lib\site-packages\numpy\core\include/numpy/npy_math.h:8,
                 from ..\scipy\special/_complexstuff.h:5,
                 from scipy/special/_ufuncs_cxx.cp39-win_amd64.pyd.p/_ufuncs_cxx.cpp:694:

[516/1557] Compiling C object scipy/stats/_unuran/unuran_wrapper.cp39-win_amd64.pyd.p/.._..__lib_unuran_unuran_src_methods_mvtdr.c.obj
In file included from ../scipy/_lib/unuran/unuran/src/methods/mvtdr.c:330:
In function '_unur_mvtdr_etable_new',
    inlined from '_unur_mvtdr_triangulate' at ../scipy/_lib/unuran/unuran/src/methods/mvtdr_init.h:978:11:
../scipy/_lib/unuran/unuran/src/methods/mvtdr_init.h:1760:17: warning: argument 1 value '18446744073709551608' exceeds maximum object size 9223372036854775807 [-Walloc-size-larger-than=]
 1760 |   GEN->etable = malloc( size * sizeof(E_TABLE*) );
      |                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from ..\scipy\_lib\unuran\unuran\src/unur_source.h:71,
                 from ../scipy/_lib/unuran/unuran/src/methods/mvtdr.c:50:
../scipy/_lib/unuran/unuran/src/methods/mvtdr_init.h: In function '_unur_mvtdr_triangulate':
c:/rtools40/ucrt64/x86_64-w64-mingw32/include/stdlib.h:531:17: note: in a call to allocation function 'malloc' declared here
  531 |   void *__cdecl malloc(size_t _Size);
      |                 ^~~~~~

[564/1557] Compiling C object scipy/special/cython_special.cp39-win_amd64.pyd.p/meson-generated_cython_special.c.obj
scipy/special/cython_special.cp39-win_amd64.pyd.p/cython_special.c:645: warning: "_USE_MATH_DEFINES" redefined
  645 |   #define _USE_MATH_DEFINES
      | 
<command-line>: note: this is the location of the previous definition
In file included from C:\hostedtoolcache\windows\Python\3.9.9\x64\lib\site-packages\numpy\core\include/numpy/ndarraytypes.h:4,
                 from C:\hostedtoolcache\windows\Python\3.9.9\x64\lib\site-packages\numpy\core\include/numpy/ndarrayobject.h:12,
                 from C:\hostedtoolcache\windows\Python\3.9.9\x64\lib\site-packages\numpy\core\include/numpy/arrayobject.h:4,
                 from scipy/special/cython_special.cp39-win_amd64.pyd.p/cython_special.c:681:

scipy/signal/_upfirdn_apply.cp39-win_amd64.pyd.p/_upfirdn_apply.c: In function '__pyx_fuse_1__pyx_pw_5scipy_6signal_14_upfirdn_apply_21_apply':
scipy/signal/_upfirdn_apply.cp39-win_amd64.pyd.p/_upfirdn_apply.c:14631:14: warning: '__pyx_v_row_size_bytes' may be used uninitialized in this function [-Wmaybe-uninitialized]
14631 |       (void)(memset(__pyx_v_output_row, 0, __pyx_v_row_size_bytes));
      |             ~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
scipy/signal/_upfirdn_apply.cp39-win_amd64.pyd.p/_upfirdn_apply.c:14056:10: note: '__pyx_v_row_size_bytes' was declared here
14056 |   size_t __pyx_v_row_size_bytes;
      |          ^~~~~~~~~~~~~~~~~~~~~~

Many warnings like:

[936/1557] Compiling Fortran object scipy/sparse/linalg/_eigen/arpack/libarpack_lib.a.p/ARPACK_SRC_sgetv0.f.obj
../scipy/sparse/linalg/_eigen/arpack/ARPACK/SRC/sgetv0.f:404:32:

  367 |           call svout (logfil, 1, rnorm0, ndigit,
      |                                 2
......
  404 |          call svout (logfil, n, resid, ndigit,
      |                                1
Warning: Rank mismatch between actual argument at (1) and actual argument at (2) (scalar and rank-1)

These rank mismatch warnings should be silenced by using -fallow-argument-mismatch, see the TODO comment in the top-level meson.build.

Also recording this one, but needs someone who speaks Fortran:

[999/1557] Compiling Fortran object scipy/stats/_mvn.cp39-win_amd64.pyd.p/mvndst.f.obj
../scipy/stats/mvndst.f:233:39:

  233 |       DOUBLE PRECISION COV(NL*(NL+1)/2), A(NL), B(NL), Y(NL)
      |                                       1
Warning: Array 'cov' at (1) is larger than limit set by '-fmax-stack-var-size=', moved from stack to static storage. This makes the procedure unsafe when called recursively, or concurrently from multiple threads. Consider using '-frecursive', or increase the '-fmax-stack-var-size=' limit, or change the code to use an ALLOCATABLE array. [-Wsurprising]

Not sure what the deal is with this one:

[1025/1557] Linking target scipy/stats/_biasedurn.cp39-win_amd64.pyd
Warning: corrupt .drectve at end of def file
Warning: corrupt .drectve at end of def file
Warning: corrupt .drectve at end of def file

../scipy/odr/__odrpack.c: In function 'odr':
../scipy/odr/__odrpack.c:1121:21: warning: format '%ld' expects argument of type 'long int', but argument 2 has type 'npy_intp' {aka 'long long int'} [-Wformat=]
 1121 |           printf("%ld %lld\n", PyArray_DIMS(work)[0], (long long)lwork);
      |                   ~~^          ~~~~~~~~~~~~~~~~~~~~~
      |                     |                            |
      |                     long int                     npy_intp {aka long long int}
      |                   %lld
../scipy/odr/__odrpack.c:1121:21: warning: format '%ld' expects argument of type 'long int', but argument 2 has type 'npy_intp' {aka 'long long int'} [-Wformat=]
 1121 |           printf("%ld %lld\n", PyArray_DIMS(work)[0], (long long)lwork);
      |                   ~~^          ~~~~~~~~~~~~~~~~~~~~~
      |                     |                            |
      |                     long int                     npy_intp {aka long long int}
      |                   %lld

Let's prioritize this one, since currently linprog (which uses HiGHS) is segfaulting:

[1474/1557] Compiling C++ object scipy/optimize/_highs/libhighs.a.p/.._..__lib_highs_src_mip_HighsImplications.cpp.obj
../scipy/_lib/highs/src/mip/HighsImplications.cpp: In member function 'bool HighsImplications::runProbing(HighsInt, HighsInt&)':
../scipy/_lib/highs/src/mip/HighsImplications.cpp:167:46: warning: 'implicsdown' may be used uninitialized in this function [-Wmaybe-uninitialized]
  167 |       if (implicsup[u].column < implicsdown[d].column)
      |  

[Smit-create]

Could you look into these?

Yeah, will look into it.

[Smit-create]

With reference to this: https://github.com/rgommers/scipy/runs/5123938429?check_suite_focus=true

1. This is the warning from autogenerated code:

scipy/linalg/_interpolativemodule.c:554:1: warning: 'thread' attribute directive ignored [-Wattributes]
  554 | static F2PY_THREAD_LOCAL_DECL cb_matvect_in_idd__user__routines_t *_active_cb_matvect_in_idd__user__routines = NULL;
      | ^~~~~~

[rgommers]

Okay, let's leave that one alone for now then. I think it is fixed in a more recent NumPy version already. At least we fixed one thread-related warning in f2py recently.

[Smit-create]

1760 | GEN->etable = malloc( size * sizeof(E_TABLE*) );

This comes from the _unuran submodule. A probable fix for this is:

diff --git a/unuran/src/methods/mvtdr_init.h b/unuran/src/methods/mvtdr_init.h
index 333b693..42ca01e 100644
--- a/unuran/src/methods/mvtdr_init.h
+++ b/unuran/src/methods/mvtdr_init.h
@@ -1757,7 +1757,7 @@ _unur_mvtdr_etable_new( struct unur_gen *gen, int size )
   GEN->etable_size = size;
 
   /* make root */
-  GEN->etable = malloc( size * sizeof(E_TABLE*) );
+  GEN->etable = malloc(_unur_min(size * sizeof(E_TABLE*), 9223372036854775807));
   if (GEN->etable==NULL) {
     _unur_error(gen->genid,UNUR_ERR_MALLOC,""); return UNUR_ERR_MALLOC; }

So, do I need to fix that here https://github.com/scipy/unuran.git?

[rgommers]

@Smit-create yes indeed, that is the place to make a change. That fix doesn't look correct though (using a hardcoded number like that is almost always a sign of trouble). The problem here seems to be that in this line:

GEN->etable = malloc( size * sizeof(E_TABLE*) );

E_TABLE isn't defined. Searching further in that file:

VERTEX *
_unur_mvtdr_etable_find_or_insert( struct unur_gen *gen, VERTEX **vidx )
...
  E_TABLE *pet, *pet_last;  /* table entry */
  int idx[2];               /* store indices */
  int hidx;                 /* hash number */
...
pet = malloc( sizeof(E_TABLE) );

However, that's the wrong E_TABLE struct. There's a global definition in mvtdr_struct.h:

typedef struct s_edge_table       /* -- hash table for edges --------------- */
{
  int  index[2];                  /* index of incident vertices              */
  VERTEX *vertex;                 /* index of corresponding vertex (=barycenter) */
  struct s_edge_table *next;      /* next entry in list                      */
} E_TABLE;

The code seems to rely on both of these files being included in the correct order in mvtdr.c.

I'm guessing the fix is to use sizeof(E_TABLE) instead of sizeof(E_TABLE*). It's not clear to me though why this code works, and only gives a build warning on Windows.

Cc @tirthasheshpatel, in case this code is familiar to you.

[rgommers]

Looks like we've fixed what needs fixing for now (modulo some discussion on the interpolative warnings PR). There's some warnings left, TBD later whether they need silencing. I'll close this.