Uh oh!
There was an error while loading. Please reload this page.
rustdoc: separate test collection from the main "clean"-ing pipeline. - #37890
Conversation
rust-highfive
commented
Nov 20, 2016
(rust_highfive has picked a reviewer for you, use r? to override) |
eddyb
commented
Nov 20, 2016
GuillaumeGomez
commented
Nov 20, 2016
Huge 👍 ! |
eddyb
commented
Nov 20, 2016
r? @nrc cc @rust-lang/tools |
alexcrichton
commented
Nov 20, 2016
Sounds like a good change to me, there's tons of technical debt in rustdoc to be shed at this point |
nrc
commented
Nov 21, 2016
@bors: r+ |
bors
commented
Nov 21, 2016
📌 Commit 60aa513 has been approved by |
bors
commented
Nov 21, 2016
⌛ Testing commit 60aa513 with merge 5940fe3... |
bors
commented
Nov 21, 2016
💔 Test failed - auto-win-msvc-64-cargotest |
GuillaumeGomez
commented
Nov 21, 2016
|
eddyb
commented
Nov 21, 2016
@alexcrichton Should the PR Travis builds include cargotest? |
alexcrichton
commented
Nov 21, 2016
@eddyb yeah once we start gating on travis I'd like to look into PR test coverage |
eddyb
commented
Nov 23, 2016
@bors r=nrc |
bors
commented
Nov 23, 2016
📌 Commit e7264d0 has been approved by |
bors
commented
Nov 23, 2016
⌛ Testing commit e7264d0 with merge d6d8d70... |
bors
commented
Nov 23, 2016
💔 Test failed - auto-win-gnu-32-opt-rustbuild |
084a3c9 to
1c5a529Compareeddyb
commented
Nov 24, 2016
Travis passed but the OSX builds are broken. @bors r=nrc |
bors
commented
Nov 24, 2016
📌 Commit 4be7786 has been approved by |
bors
commented
Nov 24, 2016
rustdoc: separate test collection from the main "clean"-ing pipeline. While reusing the documentation "clean"-ing infrastructure for collecting code examples to test may have seemed appealing at some point, doing the same through a HIR visitor is barely any harder. At the same time, supporting both "regular documentation" and "test collection" modes in `rustdoc::clean` has its cost, requiring any use of a `TyCtxt` to be speculative, and provide some sort of fallback. This simplification is the first step towards bringing rustdoc closer to the compiler, and perhaps even unifying the "local crate" (based on the HIR AST) and "inlinined across crates" (based on crate metadata and typesystem information) implementations of rustdoc. Sadly, not all possible changes to rustdoc will be uncontroversial, so I'm starting small with this patch.
While reusing the documentation "clean"-ing infrastructure for collecting code examples to test may have seemed appealing at some point, doing the same through a HIR visitor is barely any harder.
At the same time, supporting both "regular documentation" and "test collection" modes in
rustdoc::cleanhas its cost, requiring any use of aTyCtxtto be speculative, and provide some sort of fallback.This simplification is the first step towards bringing rustdoc closer to the compiler, and perhaps even unifying the "local crate" (based on the HIR AST) and "inlinined across crates" (based on crate metadata and typesystem information) implementations of rustdoc.
Sadly, not all possible changes to rustdoc will be uncontroversial, so I'm starting small with this patch.