rustdoc: Fix source-links for files with absolute-paths

[mitaa]

fixes #26995

r? @alexcrichton

[alexcrichton]

Perhaps this could be a unix-only rustdoc test which has a module with the path as /dev/null? That would avoid run-make and you could still check for the module page, right?

[mitaa]

Didn't think of that, but it seems like rustdoc doesn't generate files for empty public modules.

That did make another bug clear though, namely that absolute paths not contained in the root directory are still broken just like before.
(I'm gonna try to put these in a src/__root__ directory; for example /dev/null -> src/__root__/dev/null.html)

[alexcrichton]

Hm ok, I just try to avoid run-make like the plague as makefiles are basically impossible to get right on both windows and unix...

[alexcrichton]

@bors: r+ 78ca2e256ed86906f5d295a129f3d51e124dbbeb

[mitaa]

I pushed an update so that this also works for absolute paths not contained in the root directory.

That would unfortunately be even worse to test, so I didn't include one. (works locally though)

[alexcrichton]

This seems kinda sketchy, perhaps the iterator below could use components on Path to only look at the Component::Normal parts?

[mitaa]

(updated)

I've preserved the Component::ParentDir -> "up" renaming, so this does not look only at Component::Normal. (maybe this can be dropped?)

[alexcrichton]

@bors: r+ 4037a7da77e529a421c4fe32e910f26abacae1af

[bors]

⌛ Testing commit 4037a7d with merge 4159981...

[bors]

💔 Test failed - auto-win-gnu-32-nopt-t

[mitaa]

I think I've made your /dev/null suggestion work, removing the run-make test.

This has required commit mitaa@27ca250, because .is_file() returns false for /dev/null.
(the utility of this aside from this test seems questionable, but I hope that's okay..)
(also the same check for extern macros has been used in the same file before here, when emitting source files)

[alexcrichton]

@bors: r+ cf76fcf