Path resolution semantics should be hygenic #53007
Comments
|
cc @aturon |
bors
added a commit
that referenced
this issue
Nov 25, 2018
[beta] resolve: Implement edition hygiene for imports and absolute paths The changes in behavior of imports and absolute paths are the most significant breaking changes of 2018 edition. However, these changes are not covered by edition hygiene, so macros defined by 2015 edition crates expanded in 2018 edition crates are still interpreted in the 2018 edition way when they contain imports or absolute paths. This means the promise of seamless integration of crates built with different editions, including use of macros, doesn't hold fully. This PR fixes that and implements edition hygiene for imports and absolute paths, so they behave according to the edition in which they were written, even in macros. ### Detailed rules (mostly taken from #50376) #### Preface We keep edition data per-span in the compiler. This means each span knows its edition. Each identifier has a span, so each identifier knows its edition. #### Absolute paths Explicitly written absolute paths `::ident::...` are desugared into something like `{{root}}::ident::...` in the compiler, where `{{root}}` is also treated as an identifier. `{{root}}` inherits its span/hygienic-context from the token `::`. If the span is from 2015 edition, then `{{root}}` is interpreted as the current crate root (`crate::...` with same hygienic context). If the span is from 2018 edition, then `{{root}}` is interpreted as "crate universe" (`extern::...`). #### Imports To resolve an import `use ident::...` we need to resolve `ident` first. The idea in this PR is that `ident` fully knows how to resolve itself. If `ident`'s span is from 2015 edition, then the identifier is resolved in the current crate root (effectively `crate::ident::...` where `crate` has the same hygienic context as `ident`). If `ident`'s span is from 2018 edition, then the identifier is resolved in the current scope, without prepending anything (well, at least with uniform paths). There's one corner case in which there's no identifier - prefix-less glob import `use *;`. In this case the span is inherited from the token `*`. `use *;` && `is_2015(span(*))` -> `use crate::*;` && `span(crate) == span(*)`. `use *;` && `is_2018(span(*))` -> error. --- Why beta: - Compatibility of 2015 edition crates with 2018 edition crates, including macros, is one of the main promises of the edition initiative. - This is technically a breaking change for 2018 edition crates or crates depending on 2018 edition crates. - ~This PR is based on #55884 which hasn't landed on nightly yet :)~ No longer true (#56042). Previous attempt #50999 Closes #50376 Closes #52848 Closes #53007 r? @ghost
|
Closed by #56053 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Macros written in 2018 edition that are expanded within 2015 edition crates, should still get to use 2018 path resolution.
That is, if the macro
foo!expands to::random::prelude::random()this should work when invoked in 2015 even ifrandomwas not previously mentioned viaextern crateThe text was updated successfully, but these errors were encountered: