There are quite a number of layout tests taken from http://www.hixie.ch/tests/.
The server (www.hixie.ch) emits 'Content-Type: text/html; charset=UTF-8', but actual html documents do not have 'charset meta' element.
When layout tests are run without a http server, obviously there's no HTTP header to look at.
As a result, all of them were interpreted as the default encoding (in case of Layout tests runner, it's windows-1252/iso-8859-1). So, all the expected results were wrong !
For instance, http://www.hixie.ch/tests/adhoc/css/box/inline/bidi/002.html
has a Hebrew character, but the current expected file has two Latin characters in its place.
It's found while I was reviewing https://codereview.chromium.org/1721373002/.
The CL introduces a UTF-8-only detector when there's no charset declaration. And, it leads to a lot of rebaselined results (correct results).
I'll search for any non-ASCII files without meta charset declaration and add 'charset' to the document unless tests are specifically made to test the encoding determination mechanism.
Comment 1 by js...@chromium.org
, Mar 26 2016Summary: A number of layout tests need 'meta charset=UTF-8' or UTF-8 BOM to be added. (was: A number of layout tests need 'meta charset=UTF-8' to be added. )