parse_gcov_file: handle empty/no lnum

[blueyed]

Without this patch it would fail on the line b'xembed_window_array_take:\n',
trying to cast an empty string into an int.

        -:   46:
    #####:   47:DO_ARRAY(xembed_window_t, xembed_window, DO_NOTHING)
------------------
xembed_window_array_take:
    #####:   47:DO_ARRAY(xembed_window_t, xembed_window, DO_NOTHING)
------------------
xembed_window_array_splice:
    #####:   47:DO_ARRAY(xembed_window_t, xembed_window, DO_NOTHING)
------------------
xembed_window_array_grow:
    #####:   47:DO_ARRAY(xembed_window_t, xembed_window, DO_NOTHING)
------------------
        -:   48:

Using gcov (GCC) 8.2.1 20181127.

[blueyed]

Does not appear to be a problem with "gcov (Ubuntu 5.4.0-6ubuntu1~16.04.11) 5.4.0 20160609" (used on Travis with Ubuntu Xenial).

[blueyed]

According to #116 it can also happen with the older gcov version, so might be triggered by something else then.

I am using gcov -pb myself, both locally and on Travis, but have only seen it locally.

[LAHumphreys]

@blueyed I'm not sure if I'm missing something - but I think we still have an issue in parsing this new gcov output?

Originally I saw the issue as nonsense coverage data on a personal repo - but I can reproduce the issue by running the regressions tests I created (#147) using gcov version >= 8. Whilst the parser doesn't crash it gets confused and ends up with more "coverage" lines than actual lines in the file. As a result, all of the coverage is assigned to the wrong lines.

>> Checking /home/travis/build/Grauniad/cpp-coveralls/test-src/template_includes using GCOV (CXX:g++-8, GCOV:gcov-8)...FAIL
   File 'common_header.h': expected 10 lines, but actually has 16
   Expected:
        {
            "source_files": [{
                    "name": "A.cpp",
                    "coverage": [null, null, null, null, 1, null, 1, null]
                }, {
                    "name": "B.cpp",
                    "coverage": [null, null, null, null, 1, null, 1, null]
                }, {
                    "name": "common_header.h",
                    "coverage": [null, null, 1, 1, null, null, 1, 1, null, null]
                }]
        }
    Actual Output:
        {"service_name": "travis-ci", "source_files": [{"source_digest": "9132796bebbb415868c33ca6ad462317", "coverage": [null, null, null, null, 1, null, 1, null], "name": "B.cpp"}, {"source_digest": "4399c8184507be95b3a0d71bd45d8415", "coverage": [null, null, null, null, 1, null, 1, null], "name": "A.cpp"}, {"source_digest": "7f5ea168524b2dd779034709dd29b4fd", "coverage": [null, null, 1, 1, null, 0, 0, null, 1, 1, null, null, 1, 1, null, null], "name": "common_header.h"}], "git": {"branch": "master", "remotes": [], "head": {"committer_email": "[email protected]", "committer_name": "Luke.Humphreys", "author_email": "[email protected]", "author_name": "Luke.Humphreys", "id": "7598296a5f15da37fcde6a8917be33c8ac47d3f8", "message": "Person utilities"}}, "service_job_id": "603626911"}

I believe the issue is that although this fix ensures the parser extracts the correct line number - we don't actually use it. Instead the coverage number is simply appended to a list (line ~280)? The result is that when coveralls proceses the file the gcov files produced by the test:

lhumphreys@LAPTOP-9BKDFPPT:~/dev/cpp-coveralls/test-src/template_includes$ cat A.gcno##common_header.h.gcov
        -:    0:Source:common_header.h
        -:    0:Graph:A.gcno
        -:    0:Data:A.gcda
        -:    0:Runs:1
        -:    1:template<class T>
        -:    2:struct Util {
       1*:    3:    T Method_A(const T& i) {
       1*:    4:        return 2*i;
        -:    5:    }
------------------
_ZN4UtilIcE8Method_AERKc:
    #####:    3:    T Method_A(const T& i) {
    #####:    4:        return 2*i;
        -:    5:    }
------------------
_ZN4UtilIiE8Method_AERKi:
        1:    3:    T Method_A(const T& i) {
        1:    4:        return 2*i;
        -:    5:    }
------------------
        -:    6:
    #####:    7:    T Method_B(const T& i) {
    #####:    8:        return 3*i;
        -:    9:    }
        -:   10:};

lhumphreys@LAPTOP-9BKDFPPT:~/dev/cpp-coveralls/test-src/template_includes$ cat B.gcno##common_header.h.gcov
        -:    0:Source:common_header.h
        -:    0:Graph:B.gcno
        -:    0:Data:B.gcda
        -:    0:Runs:1
        -:    1:template<class T>
        -:    2:struct Util {
    #####:    3:    T Method_A(const T& i) {
    #####:    4:        return 2*i;
        -:    5:    }
------------------
_ZN4UtilIcE8Method_AERKc:
    #####:    3:    T Method_A(const T& i) {
    #####:    4:        return 2*i;
        -:    5:    }
------------------
_ZN4UtilIiE8Method_AERKi:
    #####:    3:    T Method_A(const T& i) {
    #####:    4:        return 2*i;
        -:    5:    }
------------------
        -:    6:
       1*:    7:    T Method_B(const T& i) {
       1*:    8:        return 3*i;
        -:    9:    }
        -:   10:};

we end up with nonsense data:

lhumphreys@LAPTOP-9BKDFPPT:~/dev/cpp-coveralls/test-src/template_includes$ ../../test-utils/dump_output.py output.json common_header.h
WARNING: Coverage reports 16 lines, but file has 10
1             :  template<class T>
2             :  struct Util {
3            1:      T Method_A(const T& i) {
4            1:          return 2*i;
5             :      }
6         MISS:
7         MISS:      T Method_B(const T& i) {
8             :          return 3*i;
9            1:      }
10           1:  };
11            :  /*EOF*/
12            :  /*EOF*/
13           1:  /*EOF*/
14           1:  /*EOF*/
15            :  /*EOF*/
16            :  /*EOF*/