Clean build directory before preprocessors run instead of after.

[XAMPPRocky]

The change shown below introduced a regression in the https://forge.rust-lang.org where the our precessor can no longer generate HTML redirects that we need to maintain compatibility. (rust-lang/rust-forge#336). I would propose to perform the clean step before preprocessors are run.

mdBook/src/renderer/html_handlebars/hbs_renderer.rs

Lines 315 to 318 in 6c4c344

	if destination.exists() {
	utils::fs::remove_dir_content(destination)
	.with_context(|| "Unable to remove stale HTML output")?;
	}

[ehuss]

I think this is a duplicate of #1087. It was intentionally changed, and there is some discussion there of possible solutions. It would be nice to have someone would make a proposal. I've only skimmed it, but it looks like there's an interesting discussion of allowing preprocessors to add "resources", which I think would also solve #1222.

[XAMPPRocky]

Well I think it should be reverted until there is a solution, otherwise I'll have to pin or maintain a fork of mdbook. It's not a great experience to intentionally break in a patch release.

[ehuss]

I appreciate that it is frustrating when something breaks, but it has been almost 10 months since the change, it would also cause breakage to revert it. I think a more constructive discussion on #1087 would be more fruitful, and provide solutions for everyone. I suspect it is not too difficult to come up with a design for preprocessors to indicate additional resources they want to include. I encourage you (or anyone) to work with @Michael-F-Bryan to come up with some solution, I'd be happy to review.

Alternatively, if it is just redirect support, I would be happy to have a first-class solution for that (#430). I know it has come up many times, and would be very valuable. I'd by happy to discuss that, as well.

Closing as a duplicate of #1087.

[XAMPPRocky]

I appreciate that it is frustrating when something breaks, but it has been almost 10 months since the change, it would also cause breakage to revert it.

My problem is not the time scale, it's the fact that this project has no respect for semver, this isn't the first time patch releases have broken forge.

[Dylan-DPC-zz]

Breaking compatibility accidentally doesn't mean we are not respecting semver

[XAMPPRocky]

I think this is a duplicate of #1087. It was intentionally changed

[Michael-F-Bryan]

@XAMPPRocky, how are you generating the redirects? You mention before/after preprocessors, so it sounds like you've written your own Preprocessor to insert redirects?

The intent behind a Preprocessor is to modify the Book after it has been loaded into memory, and return the modified Book so it can be rendered by a particular backend (e.g. the HTML renderer).

Given that design, when a renderer chooses to clear its build directory shouldn't matter, because Preprocessors shouldn't know it exists and the directory is "owned" by the renderer.

---

Alternatively, if it is just redirect support, I would be happy to have a first-class solution for that (#430). I know it has come up many times, and would be very valuable. I'd by happy to discuss that, as well.

I think the best solution would be to incorporate redirects directly into the HTML renderer (as discussed in #430) instead of having to write your own preprocessor. @XAMPPRocky, is that something you would like us to look into implementing?

I'll be free tomorrow and the day after so should have plenty of time to implement HTML redirects as a first-class feature.

[XAMPPRocky]

I've linked how forge generates redirects in its pre-processor below. I'm not interested in collaborating on a solution, breaking my code has already cost me a good chunk of time and stress dealing with people complaining that forge no longer works.

https://github.com/rust-lang/rust-forge/blob/master/blacksmith/src/lib.rs#L131