Spurious missing lines reported if template has CRLF line endings

Summary

If the Django template HTML file has CRLF line endings (\r\n) then, depending on line and file lengths, some random lines may be reported as 'missing' even though they are covered. In my experience it happens more often on short lines later in the file, e.g. some closing HTML tags near the end.

Changing the line endings to LF-only (\n) in an affected file avoids the problem (reports 100%) - the bug is definitely because of the CRLF in the HTML template file, see 'Cause' section below. But first a minimal repro...

Repro

• Running on Windows 10, Python 3.8.5 64-bit (though I expect this reproduces on linux & any recent python)
• Download and unzip issue75_minimal_repro.zip, create a venv & install `requirements.txt
• Run these commands inside the venv:

coverage run --source=./mysite/ mysite/manage.py test mysite
coverage report --include=*.html --show-missing

• Output is as follows (rather than 100% as expected):

Name                                Stmts   Miss  Cover   Missing
-----------------------------------------------------------------
mysite\myapp\templates\index.html      13      1    92%   13
-----------------------------------------------------------------
TOTAL                                  13      1    92%

where line 13 of index.html simply contains </html>

• Change the line-endings of minimal_repro\mysite\myapp\templates\index.html to be LF-only
• Run the above 2 commands again - this time the result is 100% as expected

Cause

I believe the problem is caused by this plugin comparing character offsets (from start-of-file) into the template source string, loaded from 2 different bits of code which do it slightly differnetly:

1. django_coverage_plugin.plugin.read_template_source specifies binary mode: open(filename, "rb") so this plugin is loading the HTML source including 2 characters \r\n at the end of each line if the file is CRLF
2. django.template.loaders.filesystem.Loader.get_contents doesn't specify binary mode, so python 3 universal newlines behaviour results in the loaded HTML source having only 1 character \n at the end of each line, even though the file on disk has CRLF line endings

The comparison mismatch occurs because:

• get_line_map calculates things based on (1) above (read_template_source) so includes \r
• get_line_number looks up in that line map but based on s_start & s_end which come from position which comes from Django which has read the template source without the \r as per (2) above

Therefore, depending on the length of each line, as it progresses down the file, offsets start to land on the wrong line as the off-by-one errors stack up.

Solution

I guess in this plugin we need to do one of these:

• (a) modify read_template_source to match Django's behaviour of getting universal newlines auto-fixed-up
• (b) or is it possible to import & use Django's template loader directly so as to avoid future possible mismatch?

If (a) then it would be nice - if possible - to add a sanity check that the length of the string returned by read_template_source is the same as the length of the template that Django has loaded, if we can access that.

Start with django_coverage_plugin.plugin.read_template_source, then trace get_line_map and get_line_number to compare offsets for CRLF and Django's universal-newline content. Run the commands from the minimal reproduction and verify that the affected template reports full coverage with CRLF line endings, including the previously missing closing-tag line.