Strict mode for requiring full line coverage (in gcov)?

[MichaelChirico]

For some years, we've slowly been adding newlines to code to improve our code coverage due to {covr} reporting full coverage of partially-covered lines, e.g.

-if (verbose) Rprintf("reached.\n");
+if (verbose)
+  Rprintf("reached.\n");

Now I am learning a bit more about gcov internals:

Rdatatable/data.table#6950 (comment)

It seems gcov indeed recognizes this line is only partially covered if the test suite always uses verbose=0 by reporting 1* "partially covered":

https://gcc.gnu.org/onlinedocs/gcc/Invoking-Gcov.html

Can this somehow be reported to the user? Is there any option to require lines be fully-covered in order for codecov to consider them "covered" in a 0/1 sense?

[MichaelChirico]

Looks like as of today, we only check for "uncovered":

covr/R/compiled.R

Lines 32 to 40 in d8d33b4

	matches$coverage[matches$coverage == "#####"] <- 0 # nolint
	# gcov lines which have parse error, so make untracked
	matches$coverage[matches$coverage == "====="] <- "-"
	coverage_lines <- matches$line != "0" & matches$coverage != "-"
	matches <- matches[coverage_lines, ]
	values <- as.numeric(matches$coverage)

[MichaelChirico]

To start with we can add maybe("*") here to account for these "unexecuted blocks" for partial coverage:

covr/R/compiled.R

Line 19 in d8d33b4

capture(name = "coverage", some_of(digit, "-", "#", "=")),

With that, I can get something like this in the matches data.frame:

#     coverage line
# 102      20*   98

[aitap]

How about gcov --json-format? The feature is old enough to be in Debian Stable. It may prove easier to parse than the normal text output.

[MichaelChirico]

I had the same thought but it feels like a much bigger project to ensure everything keeps WAI: #609

Happily {jsonlite} is already Imports:

covr/DESCRIPTION

Line 51 in d8d33b4

jsonlite,