Fix source filename handling when using --use-gcov-files

[mayeut]

gcov documentation states that gcov should be run with the current directory the same as that when you invoked the compiler.
It makes sense to handle source filename resolution using this information.

Fixes #145

[mayeut]

@latk, if you get some time, will you please review this one. It's rebased on latest master (232fb11).

[latk]

I've looked at this PR a couple of times and still don't understand it. There's a quite complicated test case for a change of three lines of code. While I'm sure this represents a good improvement, I'm not seeing the big picture here.

Could you give me an explanation on what the problem is, and how the test case represents this problem? It would be extremely helpful if you can walk me through the various file paths and what gcovr will do with them under your change. We can then later edit this explanation into a README for the test case.

Though not strictly required, I prefer to merge a change only if I understand what I'm doing :D

[mayeut]

The new test nested2-use-existing is built exactly the same way as nested2 test. The only difference is that gcov is ran outside of gcovr following gcov documentation (gcov should be run with the current directory the same as that when you invoked the compiler.) rather than letting gcovr manage gcov invocation.

First a little sample of what happens without the patch.
Output for nested2-use-existing:

------------------------------------------------------------------------------
                           GCC Code Coverage Report
Directory: subdir
------------------------------------------------------------------------------
File                                       Lines    Exec  Cover   Missing
------------------------------------------------------------------------------
B/main.cpp                                     9       9   100%   
------------------------------------------------------------------------------
TOTAL                                          9       9   100%
------------------------------------------------------------------------------

Output for nested2:

------------------------------------------------------------------------------
                           GCC Code Coverage Report
Directory: subdir
------------------------------------------------------------------------------
File                                       Lines    Exec  Cover   Missing
------------------------------------------------------------------------------
A/C/D/file6.cpp                                4       3    75%   4
A/C/file5.cpp                                  4       3    75%   4
A/file1.cpp                                    4       3    75%   4
A/file2.cpp                                    7       4    57%   8,10-11
A/file3.cpp                                    9       4    44%   8,10-12,14
A/file4.cpp                                    4       3    75%   6
A/file7.cpp                                    2       0     0%   1,3
B/main.cpp                                     9       9   100%   
------------------------------------------------------------------------------
TOTAL                                         43      29    67%
------------------------------------------------------------------------------

As we can see, we're missing quite a bit of coverage information.
This comes from the way gcovr currently tries to resolve the source filename when using existing gcov files.

If we have a look at the verbose output for A/file1.cpp:

Finding source file corresponding to a gcov data file
  currdir      /Users/Matt/Dev/gcovr/gcovr/tests/nested2-use-existing
  gcov_fname   /Users/Matt/Dev/gcovr/gcovr/tests/nested2-use-existing/subdir/A/file1.cpp.gcov
               ['        -', '    0', 'Source', 'file1.cpp\n']
  source_fname None
  root         /Users/Matt/Dev/gcovr/gcovr/tests/nested2-use-existing/subdir
  fname        /Users/Matt/Dev/gcovr/gcovr/tests/nested2-use-existing/file1.cpp
Parsing coverage data for file /Users/Matt/Dev/gcovr/gcovr/tests/nested2-use-existing/file1.cpp
  Filtering coverage data for file /Users/Matt/Dev/gcovr/gcovr/tests/nested2-use-existing/file1.cpp

We can see that gcovr is building the source filename from its current directory and the filename found in the gcov file. This results in /Users/Matt/Dev/gcovr/gcovr/tests/nested2-use-existing/file1.cpp which is not the path of the file which gets filtered out.

Instead, gcovr shall build the source filename from the directory in which the gcov file resides in and, as it already does, the filename found in the gcov file.

If we do that, here's the verbose output for that same file:

Finding source file corresponding to a gcov data file
  currdir      /Users/Matt/Dev/gcovr/gcovr/tests/nested2-use-existing
  gcov_fname   /Users/Matt/Dev/gcovr/gcovr/tests/nested2-use-existing/subdir/A/file1.cpp.gcov
               ['        -', '    0', 'Source', 'file1.cpp\n']
  source_fname None
  root         /Users/Matt/Dev/gcovr/gcovr/tests/nested2-use-existing/subdir
  fname        /Users/Matt/Dev/gcovr/gcovr/tests/nested2-use-existing/subdir/A/file1.cpp
Parsing coverage data for file /Users/Matt/Dev/gcovr/gcovr/tests/nested2-use-existing/subdir/A/file1.cpp

This time, the source filename is resolved correctly.
Given the number of tests using existing gcov files, I added a test not to modify the previous behavior of gcovr but, if sticking to gcov documentation, it shouldn't be necessary at all. In fact, only the last line of the patch could be kept in the surrounding if source_fname is None: test.

[latk]

Thank you very much for the clear and thorough explanation. I understand now. This is a great improvement and seems like a very reasonable approach for resolving the path.

[latk]

Can we assign the path to some variable to avoid code duplication? E.g.

possible_gcov_path = os.path.join(...)
if os.path.exists(possible_gcov_path):
  fname = aliases.unalias_path(possible_gcov_path)

[mayeut]

Done