This adds the ability to redirect URLs with `#` fragments. This is
useful when section headers get renamed or moved to other pages.
This works both for deleted pages and existing pages.
The implementation requires the use of JavaScript in order to manipulate
the location. (Ideally this would be handled on the server side.)
This also makes it so that deleted page redirects preserve the fragment
ID. Previously if you had a deleted page redirect, and the user went to
something like `page.html#foo`, it would redirect to `bar.html` without
the fragment. I think preserving the fragment is probably a better
behavior. If the new page doesn't have the fragment ID, then no harm is
really done. This is technically an open redirect, but I don't think
that there is too much danger with preserving a fragment ID?
The reason the ACE editor was failing to load the rust syntax
highlighting is because the syntax highlighting was being created
*after* the editor was created. If the editor is created first, then ACE
tries to load `ace/mode/rust`. Since it isn't already defined, it tried
to compute the URL and load it manually. However, since the URLs now
have a hash in it (via https://github.com/rust-lang/mdBook/pull/1368),
it was unable to load.
The solution here is to make sure `ace/mode/rust` is defined before
creating the editors. Then ACE knows that it can just load the module
directly instead of trying to fetch it from the server.
Fixes https://github.com/rust-lang/mdBook/issues/2700
Although there isn't a direct equivalent in the new testsuite, I felt
like these weren't adding any specific coverage that the existing tests
don't already exercise. If it does turn up that there is specific
coverage missing, then I would prefer to have tests which exercise the
specific thing of interest rather than have these kinds of non-specific
tests.
I don't think these exercise anything in particular that aren't
necessarily covered by other things. If this ends up exposing a lack
of coverage somewhere, I would prefer to add more focused test for
specific things.
After some testing I notice that this test is failing randomly because
the `true` program is exiting before mdbook is able to transmit the
JSON, and it fails with a broken pipe.
This will be replaced with backends_receive_render_context_via_stdin,
which does essentially the same thing, but does suffer from the same
problem.
