rfc, reserve-throw-fail: initial version.

Author: Centril

text/0000-reserve-throw-fail.md

@@ -0,0 +1,113 @@
+- Feature Name: `reserve_throw_fail`
+- Start Date: 2018-05-14
+- RFC PR:
+- Rust Issue:
+
+# Summary
+[summary]: #summary
+
+The keywords `throw` and `fail` are reserved in edition 2018 and beyond.
+
+# Motivation
+[motivation]: #motivation
+
+[RFC 2426]: https://github.com/rust-lang/rfcs/pull/2426
+
+The motivation for reserving `fail` and `throw` are so that we have the option
+to later use them for some `fail expr` or `throw expr`-like construct such as
+proposed in [RFC 2426].
+
+Since edition 2018 is approaching, we are under time constraints to get the
+keyword reserved even if the details of [RFC 2426] or similar proposals have
+not been fully fleshed out.
+
+The reason we are reserving two keywords is so that we can delay the choice
+between them since there isn't consensus for which one to pick.
+
+# Guide-level explanation
+[guide-level-explanation]: #guide-level-explanation
+
+The words `fail` and `throw` are reserved as keywords in edition 2018.
+This means that code in edition 2018 can't use it directly as an identifier.
+However, you can always use raw identifiers as in `r#fail` if you need
+to refer to `fail`, used in a crate from 2015, from an edition 2018 crate.
+
+# Reference-level explanation
+[reference-level-explanation]: #reference-level-explanation
+
+[list of keywords]: https://doc.rust-lang.org/book/second-edition/appendix-01-keywords.html
+
+The words `fail` and `throw` are reserved as keywords in edition 2018 and
+added to the [list of keywords].
+
+# Drawbacks
+[drawbacks]: #drawbacks
+
+## It might not end up being used
+
+Simply put, we *might* not end up using any of the keywords.
+We can also be certain that only one of the keywords will be used in the end.
+
+## Edition breakage
+
+Some code will break when transitioning from edition 2015 to 2018.
+Most of this will be easily fixable with `rustfix`.
+However, transitioning between editions will add some churn.
+
+### For `throw`
+
+We analyse the extent of the breakage and find that `throw`:
+
++ is not used as an identifier in the standard library.
++ is used as the name of a [crate](https://crates.io/crates/throw).
+  This crate has zero reverse dependencies.
++ is found 3+ times by [sourcegraph](https://sourcegraph.com/search?q=repogroup:crates+case:yes++\b((let|const|type|)\s%2Bthrow\s%2B%3D|(fn|impl|mod|struct|enum|union|trait)\s%2Bthrow)\b+max:400).
+  The extent of breakage is minimal.
+
+### For `fail`
+
+We analyse the extent of the breakage and find that `fail`:
+
++ is not used as an identifier in the standard library.
++ is used as the name of a [crate](https://crates.io/crates/fail).
+  This one does have 5 reverse dependencies.
+  However, they are all written by the same author.
+  
++ is found 20+ times by [sourcegraph](https://sourcegraph.com/search?q=repogroup:crates+case:yes++%5Cb%28%28let%7Cconst%7Ctype%7C%29%5Cs%2Bfail%5Cs%2B%3D%7C%28fn%7Cimpl%7Cmod%7Cstruct%7Cenum%7Cunion%7Ctrait%29%5Cs%2Bfail%29%5Cb+max:400).
+  The extent of breakage is fairly minimal.
+
+# Rationale and alternatives
+[alternatives]: #alternatives
+
+A more frugal option to reserving two words would be to reserve one word.
+However, as mentioned before, there is no consensus for which word that would be.
+
+Another option is to simply not reserve anything, which would limit our options
+for the future. However, we feel confident that we should keep this option open
+to us right now. Not doing so would mean that we couldn't use the words `fail`
+or `throw` as keywords for another 3 years or so.
+
+[keyword policy]: https://paper.dropbox.com/doc/Keyword-policy-SmIMziXBzoQOEQmRgjJPm
+[permalink]: https://gist.github.com/Centril/4c82c19b3cb02cc565622a37d1591785
+
+The keywords also can't be contextual since `fail {}` would clash with a struct
+named `fail`. See [RFC 2426] for a longer discussion. 
+Furthermore, a recent [keyword policy] ([permalink]), adopted by the language
+team, decided that moving forward, keywords for new features in new editions
+should be real keywords instead of being contextual. The main motivation
+for this was to optimize for maintenance (and reduce technical debt).
+
+With respect to the choice of keyword, it is also discussed in [RFC 2426].
+We pick `fail` as the non-exceptional alternative and `throw` as the most
+popular exceptional alternative. Since `raise` is exceptional terminology but
+not as frequently used as `throw`, we will not reserve it.
+
+# Prior art
+[prior-art]: #prior-art
+
+For usage of a `throw` like construct, see [RFC 2426](https://github.com/Centril/rfcs/blob/rfc/throw-expr/text/0000-throw-expr.md#prior-art)'s prior art.
+
+# Unresolved questions
+[unresolved]: #unresolved-questions
+
+None.