The search index is changed due to the addition of an extra chapter, which means you have to regenerate the searchindex_fixture.json file by:

• Setting GENERATE_FIXTURE to true on line 469
• Running cargo test
• Setting GENERATE_FIXTURE back to false

Unfortunately you can't just copy it from a normal book build, as it's not pretty printed.

The current code basically concatenates all the files into one big html file and does it as it's rendering all the files. I.e, it renders a file, outputs it to the location it's at, then appends the html into print.html.

I was going to use another project I did (https://github.com/cetra3/mdcollate) to generate a correct print.md file.

With mdcollate, I was going to:

• Traverse from SUMMARY.md
• Generate a single print.md file
• Use this as print.html

Unfortunately there would need to be a few enhancements to mdcollate to support this. Namely, dealing with windows file systems (backslashes!) & converting it into a library.

which means you have to regenerate the searchindex_fixture.json file by:

Oh wow, I just missed this, thanks :(

The current code basically concatenates all the files into one big html file and does it as it's rendering all the files. I.e, it renders a file, outputs it to the location it's at, then appends the html into print.html.

Yep, that's the root cause here. The strategy I'm attempting to take is to check if links are relative, and then re-base them. We'll see...

Okay! While this hasn't built here, things pass locally, as well as testing in the main Rust repository (cc rust-lang/rust#57495)

Can someone review this?

I don't know the mdbook source code well enough to provide a review, but I have kicked the tires on this PR and it seems to work for me.

I tossed together a quick script to convert mdbook 0.1 to 0.2: https://gist.github.com/ehuss/b2a647eb69aa8f57bc4fb397656dfac1 I ran it against the reference and I found some curious issues (unrelated to this PR) that maybe someone can answer:

• Previously you could have a link like #foo and it would be relative to the same page. Is it possible to do that in mdbook 0.2? Or does it have to include the page name like page.html#foo?
• The .md to .html translation doesn't seem to work for links with # fragments that have md in the middle. For example, foo.html#phantomdata gets rewritten as foo.html#phant.html. Am I doing something wrong, or is this just a bug?

@ehuss oh that's super cool!

I have to re-read the code, but I think those are both bugs; I'd like to support them, at least. Maybe I can fix those issues quickly and make them a part of this PR as well. I'll investigate later today and report back.

@ehuss I've pushed a commit to fix the second issue, nice catch!

The first issue, #foo, is hard though: when we make everything one page, we can't guarantee that these are all accurate. Imagine you have two different pages that contain #foo headings and anchors... fixing that problem is hard. Does linkcheck fail on these links, do you know?

The print.html page just ends up with a few links that point to the wrong section. The headers get unique anchor names (#foo #foo-1 #foo-2), but the links will all point to #foo. I believe this has always been the case (see #542).

The issue I was specifically referring to is if a fragment-only link can elide the page name or not. It's not a big deal, just a minor convenience. The script I posted inserts the page name, so it's an easy workaround.

I plan on merging this tomorrow and cutting a new release.

Thanks folks!