Repository navigation
missing edition note is silenced by any flag #164116
Description
rust-lang/compiler-team#1019 states
Implementation-wise, the changes are minimal if we go with a plain unsilenceable note: #158102
The warning seems to be silenced however, as soon as we pass any commandline arg, such --crate-type lib or -g for example.
Is this the intended behaviour?
cc #160243
Activity
hanna-kruppe commentedon Oct 10, 2026
This is apparently intentional: #158102 (comment)
But I agree that the warning is significantly less useful if it only fires in this narrow case.
cc @estebank
estebank commentedon Oct 10, 2026
The people that need the alert are the ones that are not using any flags, the benefit gets incrementally less the more "competent" the user is.
We can tweak the logic to trigger more, but trigerring less seemed like the safest thing to stabilize.
hanna-kruppe commentedon Oct 10, 2026
FWIW I've been in favor of the MCP because I also often invoke
rustcmyself and forget about the edition defaulting to 2015, and I usually remember that I have to add other flags for whatever I'm trying to do.matthiaskrgr commentedon Oct 10, 2026
I get the idea and it seems sensible imo.
The "always print a message when no args are passed" approach broke parts of my fuzzing machinery which previously did not pass "--edition=2015" explicitly and would assume "if rustc exits with no output AND bad exit code, we have some kind of bug". But that was quickly fixed by adding the 2015 edition explicitly
This was probably discussed already but I feel like it would be a bit smoother and consistent if this was an on-by-default lint like
implicit_edition_fallbackwhich could be turned off via the old and reliable. Then the "styling" of the message would also fit in with all the other warnings that we have. My first initial reaction when I saw the message was "huh did someone forget to remove a debug print?" 😅 🙈But if you say "I don't wanna invest an hour turning this very niche thing into a lint" I also understand that.
estebank commentedon Oct 10, 2026
@matthiaskrgr my concern was more about not implying that it is a mistake to use 2015, but at the same time we're trying to eliminate bare notes, so eprintln seemed like the reasonable solution.