Don't change file encoding unnecessarily.

[grokys]

When comparing files in a PR. we're currently taking a look at the file on disk and converting the encoding of the LHS of the comparison to match the on-disk file. However, if the file doesn't exist locally, we were returning Encoding.Default for this value, which was causing the unicode errors seen in #1908.

This PR changes to behavior to return null if the file doesn't exist locally, and if a null encoding is passed to PullRequestService.ExtractToTempFile then just write the contents out without any encoding.

Related

• Checked out PR file diffs sometimes complain about different encodings #1010: Checked out PR file diffs sometimes complain about different encodings

Fixes #1908.

[codecov]

Codecov Report

Merging #2009 into master will increase coverage by 0.01%.
The diff coverage is 33.33%.

@@            Coverage Diff             @@
##           master    #2009      +/-   ##
==========================================
+ Coverage   39.34%   39.36%   +0.01%     
==========================================
  Files         410      410              
  Lines       17573    17577       +4     
  Branches     2425     2427       +2     
==========================================
+ Hits         6914     6919       +5     
- Misses      10112    10114       +2     
+ Partials      547      544       -3

Impacted Files	Coverage Δ
src/GitHub.App/Services/PullRequestService.cs	35.29% <33.33%> (-0.24%)	⬇️
...nlineReviews/Services/PullRequestSessionService.cs	17.1% <0%> (+0.4%)	⬆️
...c/GitHub.InlineReviews/Tags/InlineCommentTagger.cs	80.53% <0%> (+0.88%)	⬆️
...nlineReviews/Services/PullRequestSessionManager.cs	83.75% <0%> (+1.01%)	⬆️

[meaghanlewis]

This LGTM 👍

[jcansdale]

Good catch. 👍

I believe we can simplify this by returning new UTF8Encoding(false). That way GetEncoding is responsible for returning UTF8Encoding either with or without preamble.

[jcansdale]

I believe can we can simply replace this with the following and avoid having to spacial case null.

Suggested change

	return null;
	return new UTF8Encoding(false);

This makes the GetEncoding responsible for returning Encoding.UTF8 if the target file has a preamble or new UTF8Encoding(false) if the target file doesn't exist or have a preamble. UTF8 without preamble is Git's default text encoding.

Here is what I see if we return new UTF8Encoding(false):

It's possible we could always use new UTF8Encoding(false) and avoid this issue completely. The question is what a diff looks like between UTF8 files with and without preamble. I didn't know new UTF8Encoding(false) was a thing until today.

[jcansdale]

Re: wondering if we could always use new UTF8Encoding(false). Alas, this doesn't work. I tried the following:

File.WriteAllText("foo.txt", "foobar", new UTF8Encoding(true));
File.WriteAllText("bar.txt", "foobar", new UTF8Encoding(false));

When I diff them together, I see:

I think returning return new UTF8Encoding(false) when there is no preamble or file is probably the cleanest solution.

Nonsense

[jcansdale]

This has been tested and works. 👍

It would be nice if we could simplify it at some point, because it isn't very obvious what's going on (or why it's necessary). We could use new UTF8Encoding(false) encoding, which is UTF8 a encoder that doesn't write a preamble (this is similar to git's default).