warnings from a failed flang build trial

[jalvesz]

Description

I tried to build stdlib with flang 21.1.8 from a conda installation on windows with the following cmake options:

cmake -B build -G Ninja -DBUILD_TESTING=on -DCMAKE_MAXIMUM_RANK:String=4 -DCMAKE_BUILD_TYPE=Release -DCMAKE_Fortran_COMPILER=flang -DCMAKE_C_COMPILER=clang-cl -DWITH_XDP=off -DWITH_QP=off

installation procedure: https://fortran-lang.discourse.group/t/building-flang/10697/14

I got some warnings and errors that I thought are worth noting even if flang is not yet fully production ready:

Deprecation and unsafe Windows functions

[146/2261] Building C object src\system\CMakeFiles\fortran_stdlib_system.dir\stdlib_system.c.obj
C:\\...\fortran_lang\stdlib\src\system\stdlib_system.c(51,17): warning: 'strerror' is deprecated: This function or variable may be unsafe. Consider using strerror_s instead. To disable deprecation, use _CRT_SECURE_NO_WARNINGS. See online help for details. [-Wdeprecated-declarations]
   51 |     char* err = strerror(errno);
      |                 ^
C:\Program Files (x86)\Windows Kits\10\include\10.0.22621.0\ucrt\string.h(177,16): note: 'strerror' has been explicitly marked deprecated here
  177 | _Check_return_ _CRT_INSECURE_DEPRECATE(strerror_s)
      |                ^
C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.37.32822\include\vcruntime.h(355,55): note: expanded from macro '_CRT_INSECURE_DEPRECATE'
  355 |         #define _CRT_INSECURE_DEPRECATE(_Replacement) _CRT_DEPRECATE_TEXT(    \
      |                                                       ^
C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.37.32822\include\vcruntime.h(345,47): note: expanded from macro '_CRT_DEPRECATE_TEXT'
  345 | #define _CRT_DEPRECATE_TEXT(_Text) __declspec(deprecated(_Text))
      |                                               ^
1 warning generated.

[152/2261] Building C object src\system\CMakeFiles\fortran_stdlib_system.dir\stdlib_system_subprocess.c.obj
C:\\...\fortran_lang\stdlib\src\system\stdlib_system_subprocess.c(50,20): warning: 'fopen' is deprecated: This function or variable may be unsafe. Consider using fopen_s instead. To disable deprecation, use _CRT_SECURE_NO_WARNINGS. See online help for details. [-Wdeprecated-declarations]     
   50 |         stdin_fp = fopen(stdin_file, "w");
      |                    ^
C:\Program Files (x86)\Windows Kits\10\include\10.0.22621.0\ucrt\stdio.h(212,20): note: 'fopen' has been explicitly marked deprecated here
  212 |     _Check_return_ _CRT_INSECURE_DEPRECATE(fopen_s)
      |                    ^
C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.37.32822\include\vcruntime.h(355,55): note: expanded from macro '_CRT_INSECURE_DEPRECATE'
  355 |         #define _CRT_INSECURE_DEPRECATE(_Replacement) _CRT_DEPRECATE_TEXT(    \
      |                                                       ^
C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.37.32822\include\vcruntime.h(345,47): note: expanded from macro '_CRT_DEPRECATE_TEXT'
  345 | #define _CRT_DEPRECATE_TEXT(_Text) __declspec(deprecated(_Text))
      |                                               ^
1 warning generated.

Warnings:

[144/2261] Building Fortran object src\system\CMakeFiles\fortran_stdlib_system.dir\stdlib_system.F90.obj
C:\\...\\fortran_lang\\stdlib\\src\\system\\stdlib_system.F90:1248:16: portability: Reference to generic function 'state_type' (resolving to specific 'new_state_nowhere') is ambiguous with a structure constructor of the same name [-Wambiguous-structure-constructor]
          err0 = state_type(STDLIB_SUCCESS,path,' not deleted: file does n&

Ambiguous contructors

[262/2261] Building Fortran object src\stringlist\CMakeFiles\fortran_stdlib_stringlist.dir\stdlib_stringlist_type.f90.obj
C:\\...\\fortran_lang\\stdlib\\src\\stringlist\\stdlib_stringlist_type.f90:152:21: warning: Function result is never defined [-Wundefined-function-result]
        pure function new_stringlist()
                      ^^^^^^^^^^^^^^
C:\\...\\fortran_lang\\stdlib\\src\\stringlist\\stdlib_stringlist_type.f90:169:33: portability: Reference to generic function 'stringlist_type' (resolving to specific 'new_stringlist_sarray') is ambiguous with a structure constructor of the same name [-Wambiguous-structure-constructor]    
          new_stringlist_carray = stringlist_type( sarray )
                                  ^^^^^^^^^^^^^^^^^^^^^^^^^
C:\\...\\fortran_lang\\stdlib\\src\\stringlist\\stdlib_stringlist_type.f90:179:33: portability: Reference to generic function 'stringlist_type' (resolving to specific 'new_stringlist') is ambiguous with a structure constructor of the same name [-Wambiguous-structure-constructor]
          new_stringlist_sarray = stringlist_type()

Underflow

Could be replace by ieee_nan ?

Building Fortran object src\lapack\CMakeFiles\fortran_stdlib_lapack.dir\stdlib_lapack_auxiliary.f90.obj
C:\\...\\fortran_lang\\stdlib\\build\\src\\lapack\\stdlib_lapack_auxiliary.f90:26:23: warning: underflow on division [-Wfolding-exception]
                small = one / huge(zero)
                        ^^^^^^^^^^^^^^^^
C:\\...\\fortran_lang\\stdlib\\build\\src\\lapack\\stdlib_lapack_auxiliary.f90:75:23: warning: underflow on division [-Wfolding-exception]
                small = one / huge(zero)

Cascade of errors on LAPACK

There was a very large cascade of errors, ilp and other kind parameters not being found, which seems like it coudn't find them given the hierarchy of files in between the parameters definition and their usage within the submodules. Also errors regarding the some pure procedures interface. These errors seem to be most likely due to flangs current limitations. I tried to mimick this issue with a MWE, but if everything is in a single file, then it manages, so it might be related with the multi-file architecture.

error: Semantic errors in src\lapack\CMakeFiles\fortran_stdlib_lapack.dir\stdlib_lapack_eigv_gen2.f90-pp.f90
C:\\...\\fortran_lang\\stdlib\\build\\src\\lapack\\stdlib_lapack_eigv_gen2.f90:5281:39: error: Parameter 'ilp' not found
             equivalence        ( ci( 1_ilp, 1_ilp ), civ( 1_ilp ) ),( cr(&

C:\\...\\fortran_lang\\stdlib\\build\\src\\lapack\\stdlib_lapack_eigv_svd_drivers2.f90:13562:43: error: No explicit type declared for 'dp'
             rwork( 3_ilp ) = real( n2,KIND=dp)

C:\\...\\fortran_lang\\stdlib\\build\\src\\lapack\\stdlib_lapack_eigv_svd_drivers2.f90:13538:18: warning: Reference to the procedure 'stdlib_zlascl' has an implicit interface that is distinct from another reference: incompatible dummy argument #5: incompatible dummy data object types: REAL(4) vs REAL(8) [-Wincompatible-implicit-interfaces]
                   call stdlib_zlascl( 'G',0_ilp,0_ilp, sva(p), one, m, 1_&
                   ^
C:\\...\\fortran_lang\\stdlib\\build\\src\\lapack\\stdlib_lapack_eigv_svd_drivers2.f90:3784:15: previous reference to 'stdlib_zlascl'  
                call stdlib_zlascl( 'G', 0_ilp, 0_ilp, anrm, smlnum, m, n,&
                ^
C:\\...\\fortran_lang\\stdlib\\build\\src\\lapack\\stdlib_lapack_eigv_svd_drivers2.f90:13538:18: error: Procedure 'stdlib_zlascl' referenced in pure subprogram 'stdlib_zgesvj' must be pure too
                   call stdlib_zlascl( 'G',0_ilp,0_ilp, sva(p), one, m, 1_&

Expected Behaviour

full working build was not expected as flang is still under active development but it does provide useful checks.

Version of stdlib

v0.8.1

Platform and Architecture

Windows

Additional Information

No response

[edvinolo]

The error that you get when building LAPACK is related to how flang was handling the equivalence statement and has been fixed (llvm/llvm-project#170056) in 22.1.0 (available through Homebrew, probably conda as well).

With the following configuration, I was able to build stdlib 0.8.1 using flang-22 on Ubuntu 24.04.03:

cmake -B build -DCMAKE_Fortran_COMPILER=flang -DWITH_XDP=off -DWITH_QP=off

The homebrew version does not support xdp or qp, I think currently you have to build the compiler from source for that.

Warnings

These are the warnings I got when compiling.

Underflow

Same as you saw above

/home/edvin/stdlib-0.8.1/build/src/lapack/stdlib_lapack_auxiliary.f90:26:23: warning: underflow on division [-Wfolding-exception]
                small = one / huge(zero)
                        ^^^^^^^^^^^^^^^^
/home/edvin/stdlib-0.8.1/build/src/lapack/stdlib_lapack_auxiliary.f90:75:23: warning: underflow on division [-Wfolding-exception]
                small = one / huge(zero)
                        ^^^^^^^^^^^^^^^^

Undefined function result

Interestingly, I don't get the other warnings you saw about ambiguous structure constructors.

/home/edvin/stdlib-0.8.1/src/stringlist/stdlib_stringlist_type.f90:152:19: warning: Function result is never defined [-Wundefined-function-result]
      pure function new_stringlist()
                    ^^^^^^^^^^^^^^

Macro redefinition

This one also gfortran warns about.

/home/edvin/stdlib-0.8.1/src/system/stdlib_system_subprocess.c:11:9: warning: '_POSIX_C_SOURCE' macro redefined [-Wmacro-redefined]
   11 | #define _POSIX_C_SOURCE 199309L
      |         ^
/usr/include/features.h:292:10: note: previous definition is here
  292 | # define _POSIX_C_SOURCE        200809L
      |          ^
1 warning generated.

Missing kind suffixes

For some of the tests, flang warns about missing kind suffixes on some real literals.

/home/edvin/stdlib-0.8.1/build/test/linalg/test_linalg_solve_iterative.f90:53:16: warning: Default real literal in REAL(8) context might need a kind suffix, as its rounded value [REAL(4)::9.0899996459484100341796875e-2_4,6.363999843597412109375e-1_4] is inexact [-Wreal-constant-widening]
          xref = [0.0909, 0.6364]
                 ^^^^^^^^^^^^^^^^

/home/edvin/stdlib-0.8.1/build/test/specialfunctions/test_specialfunctions_activations.f90:243:21: warning: Default real literal in REAL(8) context might need a kind suffix, as its rounded value [REAL(4)::-4.5500278472900390625e-2_4,-9.318850934505462646484375e-2_4,-1.4806695282459259033203125e-1_4,-1.6832835972309112548828125e-1_4,-9.1571204364299774169921875e-2_4,1.30650997161865234375e-1_4,4.98338282108306884765625e-1_4,9.63044226169586181640625e-1_4,1.46236705780029296875_4,1.954499721527099609375_4] is inexact [-Wreal-constant-widening]
              y_ref = [-0.0455002784729  , -0.093188509345055, -0.148066952824593,&
                      ^

/home/edvin/stdlib-0.8.1/build/test/specialfunctions/test_specialfunctions_activations.f90:311:21: warning: Default real literal in REAL(8) context might need a kind suffix, as its rounded value [REAL(4)::1.19202919304370880126953125e-1_4,1.7428530752658843994140625e-1_4,2.4766378104686737060546875e-1_4,3.392436504364013671875e-1_4,4.44671928882598876953125e-1_4,5.55328071117401123046875e-1_4,6.607563495635986328125e-1_4,7.52336204051971435546875e-1_4,8.2571470737457275390625e-1_4,8.8079702854156494140625e-1_4] is inexact [-Wreal-constant-widening]
              y_ref = [0.119202919304371, 0.174285307526588, 0.247663781046867,&
                      ^

Executable stack

ld (not flang) warns about an executable stack in some cases. I think it comes from the use of internal procedures and how they are implemented by the compiler, see e.g. the discussion here. I also get the same warnings with gfortran-13, but some of the comments in the discourse thread I linked suggest that later versions of gfortran might handle things differently.

/usr/bin/ld: warning: stdlib_linalg_iterative_solvers_bicgstab.f90.o: requires executable stack (because the .note.GNU-stack section is executable)
/usr/bin/ld: warning: stdlib_linalg_iterative_solvers_bicgstab.f90.o: requires executable stack (because the .note.GNU-stack section is executable)
/usr/bin/ld: warning: CMakeFiles/example_solve_bicgstab_wilkinson.dir/example_solve_bicgstab_wilkinson.f90.o: requires executable stack (because the .note.GNU-stack section is executable)
/usr/bin/ld: warning: stdlib_linalg_iterative_solvers_cg.f90.o: requires executable stack (because the .note.GNU-stack section is executable)
/usr/bin/ld: warning: stdlib_linalg_iterative_solvers_pcg.f90.o: requires executable stack (because the .note.GNU-stack section is executable)
/usr/bin/ld: warning: CMakeFiles/example_solve_custom.dir/example_solve_custom.f90.o: requires executable stack (because the .note.GNU-stack section is executable)

Failed tests

Logger

Note that I only kept some of the output generated by this test.

 running test_adding_log_files
 Able to open "first_log_file.txt" as expected
 Incremented to 1 LOG_UNITS as expected.
 Able to open "second_log_file.txt" as expected
 Incremented to 2 LOG_UNITS as expected.
 Able to open "third_log_file.txt" as expected
 Incremented to 3 LOG_UNITS as expected.
 Unable to OPEN "fourth_log_file.txt" as "NEW" as it already exists, which is a
 n expected result.
Fortran ERROR STOP: Unable to open "fourth_log_file.txt" as "OLD".
<end of output>
Test time =   0.00 sec
----------------------------------------------------------
Test Failed.
"stdlib_logger" end time: Mar 12 00:50 CET
"stdlib_logger" time elapsed: 00:00:00
----------------------------------------------------------

string_to_string

The issue here seems to be that for g0 flang formats some reals without any trailing zeros, i.e. 1. instead of 1.0, while the test assumes that there will be at least one zero after the decimal sign. I'm not familiar enough with the standard to tell if what flang is doing is ok here.

# Testing: string-to_string
  Starting to_string-complex ... (1/9)
       ... to_string-complex [FAILED]
  Message: Default formatter for complex number
Expected: '(1.0' but got '(1.,1.)'
  Starting to_string-integer ... (2/9)
       ... to_string-integer [PASSED]
  Starting to_string-logical ... (3/9)
       ... to_string-logical [PASSED]
  Starting to_string-real ... (4/9)
       ... to_string-real [FAILED]
  Message: Default formatter for real number
Expected: '100.0' but got '100.'
  Starting to_string-limit-i1 ... (5/9)
       ... to_string-limit-i1 [PASSED]
  Starting to_string-limit-i2 ... (6/9)
       ... to_string-limit-i2 [PASSED]
  Starting to_string-limit-i4 ... (7/9)
       ... to_string-limit-i4 [PASSED]
  Starting to_string-limit-i8 ... (8/9)
       ... to_string-limit-i8 [PASSED]
  Starting to_c_char ... (9/9)
       ... to_c_char [PASSED]
2 test(s) failed!
 - Default formatter for integer number
   Result: '100'
 - Formatter for integer number
   Result: '   100'
 - Formatter with zero padding for integer number
   Result: '000100'
 - Multiple formatters for integers
   Result: '   100   1000'
 - Binary formatter for integer number
   Result: '  100010'
 - Octal formatter with zero padding for integer number
   Result: '042'
 - Hexadecimal formatter for integer number
   Result: ' 22'
 - Default formatter for logcal value
   Result: 'T'
 - Formatter for logical value
   Result: ' T'
 - Multiple formatters for logical values
   Result: ' F    T'
 - Invalid formatter for logical value
   Result: '[*]'
Fortran ERROR STOP: code