This allows us to set the encoding of an IO passed without reading it
into memory.
This is useful if we want to stream files into Gitaly. Like we do when
uploading a new file to the repository.
Saved Wiki content goes through the GitalyClient::WikiService, which calls
StringIO#set_encoding on the input stream. The problem is that this call
mutates the encoding of the given string object to ASCII-88BIT, which
causes problems for models expecting the data to still be in UTF-8.
Freezing the input disables this behavior:
https://github.com/ruby/ruby/blob/v2_4_4/ext/stringio/stringio.c#L1583Closes#50590
If we set `ENCODING_CONFIDENCE_THRESHOLD` to 40,
this test case would not pass. If we raise to 50,
this would pass.
Note that if in the future rugged didn't return
corrupt data, this would be less relevant. But still
icu recommend the threshold to be 50, we should just
stick with 50.
GitLab Bot