mystery: Devel::Cover showing missing statement coverage on a line that is being run

[pryrt]

Given the SSCCE below, Devel::Cover v1.40 shows missing "statement" coverage for a line of code that is being run:

#!perl
use 5.012; # strict, //
use warnings;

{
    my $i = 1;
    my $j = 2;
    sub get {
        $i + $j;
    }
}
get();

and running with

perl -MDevel::Cover x.pl 
cover

why does it say only 90.9% coverage?

----- ------ ------ ------ ------ ------ ------ ------
File    stmt   bran   cond    sub    pod   time  total
----- ------ ------ ------ ------ ------ ------ ------
x.pl    90.9    n/a    n/a  100.0    n/a  100.0   92.8
Total   90.9    n/a    n/a  100.0    n/a  100.0   92.8
----- ------ ------ ------ ------ ------ ------ ------

The coverage report shows the following table:

line	stmt	bran	cond	sub	pod	time	code
1							#!perl
2	1 1			1		0 0	use 5.012; # strict, //
3	1 1 1			1		0 0 0	use warnings;
4
5							{
6	1 0					0 0	my $i = 1;
7	1					0	my $j = 2;
8							sub get {
9	1			1			$i + $j;
10							}
11							}
12	1 1					74157 0	get();

... but I cannot see why line 6 has two rows in the table, or why it says that the second row for that line was never run; there's only one statement, as far as I can tell. Removing the  = 1 or changing the order of the declarations doesn't change the fact that the first row in that block claims to never run.

If I change to do { ... }; instead of just a block, then line 6 cleans up and only shows one row in the table, but now the line 4 do { shows it was run 0 times:

line	stmt	bran	cond	sub	pod	time	code
1							#!perl
2	1 1			1		5875 0	use 5.012; # strict, //
3	1 1 1			1		0 0 0	use warnings;
4
5	0					0	do {
6	1					0	my $i = 1;
7	1					0	my $j = 2;
8							sub get {
9	1			1			$i + $j;
10							}
11	1					58427	};
12	1					0	get();

If I convert the block to a sub init { ... } and then call it with init(), then the sub init { line shows as uncovered statement, just like the do { version.

I can work around it by getting rid of the block, so eliminating the coverage-flag isn't a problem (but I want the block in my real code to encapsulate the my variables to be private just for my get function, so I'm not going to do that). Or I can cheat with # uncoverable statement count:2. But my curiosity has been piqued as to why I should need a workaround at all.

[haukex]

I can confirm the same results, and it looks like a bug to me. The minimal test case I was able to reduce it to is { sub x {} } x() (a print can be inserted into the sub to prove it actually does run). The issue trackers contain a few bugs for similar kinds of issues where coverage is not reported correctly, so you may want to open a new bug report.

[pryrt]

Thanks for the confirmation. Created issue 324.