Robzz
Docset index contains relocated symbols that only result in a redirection
Describe the bug
When a crate re-exports symbols from a private module, cargo doc still produces html files for these re-exported symbols in the private module, and these files simply redirect to the public location, and omits the private location from the JS search index. Unfortunately cargo-docset is still indexing these private locations.
To Reproduce
$ cargo init --lib --name doctest
$ cat >src/lib.rs <<EOF
mod foo {
/// Yes this is Bar.
pub struct Bar;
}
pub use foo::Bar;
EOF
$ cargo docset
$ sqlite3 target/docset/doctest.docset/Contents/Resources/docSet.dsidx 'SELECT * FROM searchIndex;'
1|doctest|Package|doctest/index.html
2|doctest::Bar|Struct|doctest/struct.Bar.html
3|doctest::foo::Bar|Struct|doctest/foo/struct.Bar.html
Expected behavior The index should look like
1|doctest|Package|doctest/index.html
2|doctest::Bar|Struct|doctest/struct.Bar.html
Screenshots
Desktop (please complete the following information):
- OS: macOS 11.2.3 (20D91)
- Documentation browser [e.g. dash, zeal]: Dash
- Version: cargo-docset 0.2.1
Additional context These redirection files look like
<!DOCTYPE html>
<html lang="en">
<head>
<meta http-equiv="refresh" content="0;URL=../../doctest/struct.Bar.html">
</head>
<body>
<p>Redirecting to <a href="../../doctest/struct.Bar.html">../../doctest/struct.Bar.html</a>...</p>
<script>location.replace("../../doctest/struct.Bar.html" + location.search + location.hash);</script>
</body>
</html>
It shouldn't be hard to detect the redirection and ignore the symbol.
I also wonder how hard it would be to extract the JSON search index from target/doc/search-index.js and use that to build the docset index? That seems like a more long-term effort though.
Thanks for this report and the other ones, I'll look into this and make a new release.
I finally looked into this and it appears redirection pages now include <title>Redirection</title> in the head section, which is probably the easiest way to identify them. I implemented this in #45 which seems to work and I haven't noticed any missing entries as a consequence. Docset generation does seem a bit slower as a result of reading the beginning of every HTML page, but it's still fast enough that I didn't feel the need to measure it.
I'll make a point release in a few days if I don't notice anything wrong by then.