Callgrind parser: Fixed associating file names with symbols and detection of root nodes

[glatosinski]

This PR introduces several small improvements to updating the current file name for the Callgrind parser:

• First of all, when called function (cfn) is within an inlined code block, it should use file names provided by fi or fe keys.
• Secondly, when fn key is present, the parser should set the current file name to the value provided in the previous fl key (such reset is also applied in KCacheGrind: link to code.
• Thirdly, the file associated with fn should be stored separately so that it is not overwritten later by new fl calls.

Those small modifications improve detection of root nodes. Before the PR, the same function could have several files associated with it - when fi/fe are not taken into account, the file name for the current callee is taken from the last fl or cfi call, which is later added to childMap in this.childrenTotalWeights instead of the actual function with its file name. After this, only one of the two representations of the same symbol get deleted from rootNodes.

Without the improvement, we can observe detached call stacks for such "orphaned" functions. With the improvement, the functions have properly provided names, which leads to better generation of the call graph.

Some examples before and after applying the commit from the PR (the marked calls are for the same function):

• Example 1 (callgrind file)

  • Before:
    
  • After:
    

• Example 2

  • Before:
    

  • After:
    

• Example 3

  • Before:
    
  • After:
    

This fixes #465

[coveralls]

coverage: 43.652% (+0.03%) from 43.619%
when pulling 20bc6ce on antmicro:glat/callgrind-filename-parsing-fix
into 0121cf9 on jlfwong:main.

[jlfwong]

Hi @glatosinski!

Thanks for contributing this, and sorry for the long response time. Can you please add a minimal test that demonstrates the failure mode this PR fixes?

Thanks!

[glatosinski]

Hi, ok, I will prepare a minimal test

[alexmk92]

Is this ever going to be merged, its very useful - happy to pull it into my fork in the meantime; cheers!

[jlfwong]

Hi @alexmk92

This is just blocked on adding minimal test cases that demonstrate why this change is necessary