We forgot to update the BabelStream case study when we changed the coverage schema. Running the example as it was gave an error due to schema validation.
This commit is a best-effort attempt to convert the old coverage file into the new format. It is "best-effort" because:
The old format did not include filenames, but the new format requires them. Since the case study already uses anonymized platforms and languages, the simplest solution here was to generate names like "file1" and "file2".
The old format stored lines as regions containing the number of "real" lines of code in the range (start, end), rather than storing the specific lines of code that were used. However, since all the data was generated by the same version of Code Base Investigator parsing the same files, all the regions line up. The simplest solution therefore was to pretend that the region represented a contiguous run of lines. The information about the number of comments and amount of whitespace in a region is destroyed by this transformation, but it isn't necessary to compute divergence.
Related issues
Closes #50.
Proposed changes
Replace outdated coverage.csv with one using coverage strings that conform to 0.3.0 schema.
We forgot to update the BabelStream case study when we changed the coverage schema. Running the example as it was gave an error due to schema validation.
This commit is a best-effort attempt to convert the old coverage file into the new format. It is "best-effort" because:
The old format did not include filenames, but the new format requires them. Since the case study already uses anonymized platforms and languages, the simplest solution here was to generate names like "file1" and "file2".
The old format stored lines as regions containing the number of "real" lines of code in the range (start, end), rather than storing the specific lines of code that were used. However, since all the data was generated by the same version of Code Base Investigator parsing the same files, all the regions line up. The simplest solution therefore was to pretend that the region represented a contiguous run of lines. The information about the number of comments and amount of whitespace in a region is destroyed by this transformation, but it isn't necessary to compute divergence.
Related issues
Closes #50.
Proposed changes