Should templates with multiple instantiations contribute to summaries in gcov?

["Jørgen Kvalsvik" <j@lambda.is>]

I have this program:

#include <cstdio>

template<class T>
T add (T x, T y) {
     if (x > y)
         return x + y;
     else
         return x;
}

template<class T>
T sub (T x, T y) {
     if (x > y)
         return x - y;
     else
         return x;
}

int main() {
     int i1 = 10;
     int i2 = 12;

     double d1 = 10.0;
     double d2 = 12.0;

     auto x = add(i1, i2);
     auto y = add(d1, d2);
     auto z = sub(i2, i1);

     printf ("%d %f %d\n", x, y, z);
}

add() is instantiated for int, double, and sub() is only instantiated 
for int.

$ gcc -v
gcc version 12.0.1 20220426 (experimental) (GCC)
$ gcc --coverage demo.cc -o demo && ./demo
$ gcov -mfb demo
Function 'int sub<int>(int, int)'
Lines executed:75.00% of 4

Function 'double add<double>(double, double)'
No executable lines

Function 'int add<int>(int, int)'
No executable lines

Function 'main'
Lines executed:100.00% of 10

File 'demo.cc'
Lines executed:88.89% of 18
Branches executed:100.00% of 2
Taken at least once:50.00% of 2
Calls executed:100.00% of 4
Creating 'demo.cc.gcov'

Lines executed:88.89% of 18

So it reports lines and branches accurately for templates with a single 
instantiation, but nothing when there are multiple. When you look at the 
.gcov file however the branches are reported for all the instantiations:

double add<double>(double, double):
function double add<double>(double, double) called 1 returned 100% 
blocks executed 75%
         1:    4:T add (T x, T y) {
         1:    5:    if (x > y)
branch  0 taken 0% (fallthrough)
branch  1 taken 100%
     #####:    6:        return x + y;
         -:    7:    else
         1:    8:        return x;
         -:    9:}

Skimming gcov.cc I found accumulate_line_info. By applying this patch 
the file summary includes all instantiations:

diff --git a/gcc/gcov.cc b/gcc/gcov.cc
index 04bbc774eec..c0ee9d1cd3f 100644
--- a/gcc/gcov.cc
+++ b/gcc/gcov.cc
@@ -2952,7 +2952,7 @@ accumulate_line_counts (source_info *src)
            it2 != fn->lines.end (); it2++)
           {
             line_info *line = &(*it2);
-           accumulate_line_info (line, src, false);
+           accumulate_line_info (line, src, true);
           }
      }

File 'demo.cc'
Lines executed:84.62% of 26
Branches executed:100.00% of 6
Taken at least once:50.00% of 6
Calls executed:100.00% of 4
Creating 'demo.cc.gcov'

It doesn't fix the function summaries however:

Function 'int add<int>(int, int)'
No executable lines

Looking through the changelog it looks like this was introduced to work 
with destructors/C++ clones 
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=48463

Is there a non-obvious downside to look into groups in order to have 
templates count towards file- and function summaries? Any subtle 
interactions at play? Is this a bug?

Thanks,
Jørgen