WIP: Make tests and examples compile with Rust 1.10 #104
Conversation
error_chain uses attributes on statements, stablized in Rust 1.13. Factor them out of functions in the non-backtrace codepath, making it possible to build a crate that uses error_chain with Rust 1.10. The backtrace feature still requires Rust 1.13 for attributes on statements, due to the impl_extract_backtrace macro, which needs to emit a function with one block for each linked error type, some of which may have attributes on them. And any code using compile-time conditionals in attributes on variants within an error_chain! invocation (which includes many of the tests and examples) will produce an error due to rust-lang/rust#22250 (fixed in Rust 1.11). But a crate using error_chain with default-features = false and avoiding such compile-time conditionals will now compile.
| let state = error_chain::State { | ||
| next_error: None, | ||
| }; | ||
| println!(" State.next_error: {}", size_of_val(&state.next_error)); | ||
| } | ||
| size_of_state(); |
There was a problem hiding this comment.
Could you change the code to use State::default()? That way, only the second println would be compiled conditionally.
There was a problem hiding this comment.
Sure. But this isn't intended to merge, anyway, since it doesn't work completely. #103 works, this just archives the attempt to make tests and examples work.
| @@ -262,6 +262,8 @@ fn links() { | |||
| Test(test::Error, test::ErrorKind); | |||
| } | |||
| } | |||
|
|
|||
| {} | |||
There was a problem hiding this comment.
Documented in the commit message: this works around a Rust 1.10 bug (fixed in 1.11 via rust-lang/rust#34436 ), which causes Rust to think that an item-generating macro invoked at the end of a function can only generate expressions instead (and then producing piles of errors when the macro attempts to generate items instead of expressions).
|
|
||
| error_chain! { | ||
| foreign_links { | ||
| Fmt(fmt::Error); | ||
| Var(env::VarError); |
There was a problem hiding this comment.
Documented in the commit message: in Rust 1.10, fmt::Error doesn't impl Error.
|
Sorry, I only checked the diff ^^ |
- Avoid using attributes on statements; use them on items instead. - Change a test to use std::env::VarError instead of std::fmt::Error, because the latter didn't impl Error until Rust 1.11. - Add an empty block after invocations of `error_chain!` inside functions, to work around rust-lang/rust#34436 (fixed in Rust 1.11). - Replace a single instance of `?` with `try!`. This still fails to work with 1.10 due to rust-lang/rust#22250 . Because of that issue, an invocation of `#[derive(Debug)]` inside a macro doesn't take cfg(...) attributes into account, and fails to compile due to referencing enum variants that don't exist. Posted for reference only.
joshtriplett
force-pushed
the
rust-1.10-tests-examples
branch
from
c52440b to
eb9ba6e
Compare
|
So, do you want to also merge this PR? |
|
No, it doesn't seem worthwhile if it only partially works (and if Travis can't test it to make sure it keeps working). Someone who wants to try to make this work can use this PR as a starting point, but for now, closing and leaving for reference. |
Avoid using attributes on statements; use them on items instead.
Change a test to use std::env::VarError instead of std::fmt::Error,
because the latter didn't impl Error until Rust 1.11.
Add an empty block after invocations of
error_chain!insidefunctions, to work around Allow statement-generating braced macro invocations at the end of blocks rust-lang/rust#34436
(fixed in Rust 1.11).
Replace a single instance of
?withtry!.This still fails to work with 1.10 due to
rust-lang/rust#22250 . Because of that issue,
an invocation of
#[derive(Debug)]inside a macro doesn't take cfg(...)attributes into account, and fails to compile due to referencing enum
variants that don't exist. Posted for reference only.