Build failure with submodules

[sgeard]

The attached code fails to build:

  simon@localhost autodiff (master)]$ fpm build
   + mkdir -p build/dependencies
   + mkdir -p build/gfortran_2A42023B310FA28D/autodiff
   + gfortran -c ././src/avd.f90 -Wall -Wextra -Wimplicit-interface -fPIC -fmax-errors=1 -g -fcheck=bounds -fcheck=array-temps -fbacktrace -fcoarray=single -J build/gfortran_2A42023B310FA28D/autodiff -I build/gfortran_2A42023B310FA28D/autodiff -o build/gfortran_2A42023B310FA28D/autodiff/src_avd.f90.o
   + gfortran -c app/av_utest.f90 -Wall -Wextra -Wimplicit-interface -fPIC -fmax-errors=1 -g -fcheck=bounds -fcheck=array-temps -fbacktrace -fcoarray=single -J build/gfortran_2A42023B310FA28D/autodiff -I build/gfortran_2A42023B310FA28D/autodiff -o build/gfortran_2A42023B310FA28D/autodiff/app_av_utest.f90.o
  app/av_utest.f90:19:12:
  
     19 |         u = 0.5*sin(2*a)
        |            1
  Error: Unexpected derived-type entities in binary intrinsic numeric operator ‘*’ at (1)
  compilation terminated due to -fmax-errors=1.
   Command failed
  ERROR STOP 
  
  Error termination. Backtrace:
  #0  0x4118f9 in __fpm_environment_MOD_run
          at ././src/fpm_environment.f90:144
  #1  0x476d4a in build_target
          at ././src/fpm_backend.f90:247
  #2  0x476d4a in __fpm_backend_MOD_build_package
          at ././src/fpm_backend.f90:76
  #3  0x41966b in __fpm_MOD_cmd_build
          at ./src/fpm.f90:270
  #4  0x402422 in MAIN__
          at app/main.f90:26
  #5  0x402422 in main
          at app/main.f90:9

This is a working project which builds and runs fine with makefiles. It looks as if the problem could be due to the fact that the submodule needs to be compiled after the module. As an aside I don't think fpm itself should be reporting its own call stack in the final release.

[sgeard]

Seems I can't attach a tgz file :( - I'll try to think of something else

[sgeard]

autodiff.gz

... but it does support .gz - go figure!
Extract with

tar xf autodiff.gz

[awvwgk]

I got it to build with

fpm build --profile debug --flag -fdefault-real-8

or by applying this patch to remove some implied real(4) to real(8) conversions:

diff --git a/app/av_utest.f90 b/app/av_utest.f90
index a5770f0..87935e0 100644
--- a/app/av_utest.f90
+++ b/app/av_utest.f90
@@ -16,10 +16,10 @@ program utest
         ref = T1(0.5*sin(1.0),cos(1.0)) !,-2*sin(1.0d0)) !,-4*cos(1.0d0),8*sin(1.0d0))
         call testResult_av1('T1 => sin(x)*cos(x)',ref,w)
 
-        u = 0.5*sin(2*a)
+        u = 0.5d0*sin(2*a)
         call testResult_av1('T1 => 0.5*sin(2x)',ref,u)
 
-        s = 0.5
+        s = 0.5d0
         w = s**3 + s
         ref = T1(0.5**3+0.5,3*0.5**2 + 1) !,6*0.5) !, 6, 0)
         call testResult_av1('T1 => x**3 + x',ref,w)
@@ -38,10 +38,10 @@ program utest
         ref = T2(0.5*sin(1.0),cos(1.0),-2*sin(1.0)) !,-4*cos(1.0d0),8*sin(1.0d0))
         call testResult_av2('T2 => sin(x)*cos(x)',ref,w)
 
-        u = 0.5*sin(2*a)
+        u = 0.5d0*sin(2*a)
         call testResult_av2('T2 => 0.5*sin(2x)',ref,u)
 
-        s = 0.5
+        s = 0.5d0
         w = s**3 + s
         ref = T2(0.5**3+0.5,3*0.5**2 + 1,6*0.5) !, 6, 0)
         call testResult_av2('T2 => x**3 + x',ref,w)
@@ -61,10 +61,10 @@ program utest
         ref = T3(0.5*sin(1.0),cos(1.0),-2*sin(1.0),-4*cos(1.0)) !,8*sin(1.0d0))
         call testResult_av3('T3 => sin(x)*cos(x)',ref,w)
 
-        u = 0.5*sin(2*a)
+        u = 0.5d0*sin(2*a)
         call testResult_av3('T3 => 0.5*sin(2x)',ref,u)
 
-        s = 0.5
+        s = 0.5d0
         w = s**3 + s
         ref = T3(0.5**3+0.5,3*0.5**2 + 1,6*0.5, 6)
         call testResult_av3('T3 => x**3 + x',ref,w)
@@ -94,10 +94,10 @@ program utest
         ref = T4(0.5*sin(1.0),cos(1.0),-2*sin(1.0),-4*cos(1.0),8*sin(1.0))
         call testResult_av4('T4 => sin(x)*cos(x)',ref,w)
...skipping...
         call testResult_av3('T3 => x**3 + x',ref,w)
@@ -94,10 +94,10 @@ program utest
         ref = T4(0.5*sin(1.0),cos(1.0),-2*sin(1.0),-4*cos(1.0),8*sin(1.0))
         call testResult_av4('T4 => sin(x)*cos(x)',ref,w)
 
-        u = 0.5*sin(2*a)
+        u = 0.5d0*sin(2*a)
         call testResult_av4('T4 => 0.5*sin(2x)',ref,u)
 
-        s = 0.5
+        s = 0.5d0
         w = s**3 + s
         ref = T4(0.5**3+0.5,3*0.5**2 + 1,6*0.5, 6, 0)
         call testResult_av4('T4 => x**3 + x',ref,w)
@@ -142,11 +142,11 @@ program utest
         ref = T4(t*y,1+log(2.0)+log(t),1/t,-1/t**2,2/t**3)
         call testResult_av4('T4 => x*log(2x)',ref,w)
 
-        u = 0.5*sinh(2*a)
+        u = 0.5d0*sinh(2*a)
         ref = T4(0.5*sinh(1.0),cosh(1.0),2*sinh(1.0),4*cosh(1.0),8*sinh(1.0))
         call testResult_av4('T4 => 0.5*sinh(2x)',ref,u)
 
-        u = 0.5*cosh(2*a)
+        u = 0.5d0*cosh(2*a)
         ref = T4(0.5*cosh(1.0),sinh(1.0),2*cosh(1.0),4*sinh(1.0),8*cosh(1.0))
         call testResult_av4('T4 => 0.5*cosh(2x)',ref,u)
 

[sgeard]

Nice catch. The compiler options I use in my makefiles for gfortran/debug builds:

-fPIC -cpp -std=f2018 -fimplicit-none -fdefault-real-8 -ffree-line-length-200 -Wall -Wextra -Jobj_gfortran_debug -D_DEBUG -W -ggdb -fbounds-check -ffpe-trap=denormal,invalid

and for ifort/debug builds:

-fPIC -fpp -r8 -module obj_intel_debug -D_DEBUG -g -check bounds -warn all -debug-parameters used -traceback

I never want real*4 so this gets round having to specify d0 everywhere. Can I put the compiler options in the .toml file?

[awvwgk]

Not yet, unfortunately, but we are on it. See #350 and #112 for the discussion so far.

[awvwgk]

The next best option will be the response file support John is adding with #364. This will make specifying custom arguments easier until we get the compile profiles into the package manifest.

[ivan-pi]

I never want real*4 so this gets round having to specify d0 everywhere.

That's a nice trick, even if achieved in a non-standard way using compiler flags.

[LKedward]

I never want real*4 so this gets round having to specify d0 everywhere.

That's a nice trick, even if achieved in a non-standard way using compiler flags.

I'm not so much a fan of relying on compiler flags for correctness in compilation or execution; it just causes issues when testing other compilers/build systems and makes your code difficult to use by someone else not familiar with it. Considering the small size of Sebastian's patch in this instance, the custom compiler flag doesn't seem beneficial IMO.

Since this is to do with compiler flags I'll close this issue.