Do not refresh the names of implicit arguments.

Try just going with the user-given names, and not worrying about
what happens with repeated names or anonymous implicits.
(Support for anonymous implicits is due to herbelin in #11098.)

This PR should not change behaviour in the absence of repeated names.
Since repeated names are already a poorly handled corner case, I would
recommend changing binder names to avoid overlap in the case of a
change in behavior.

Since anonymous implicits and implicits with repeated names can already
happen, I think this is unlikely to cause too many new problems,
though it might exacerbate existing ones. However, I already had to fix
one newly possible anomaly, so I can't be too confident.

The most common change in external developments was that an argument
no longer gets `0` appended to it, causing the `Arguments` command
to complain about renaming.
To fix this and keep the old name, one can simply use the `rename` flag
as suggested, or switch to the new, un-suffixed name.

Closes #6785
Closes #12001

Another step towards checking the standard library with `-mangle-names`.

1 files changed, 1 insertions, 1 deletions

diff --git a/test-suite/success/Typeclasses.v b/test-suite/success/Typeclasses.v
index 563651cfa5..7acaa92b89 100644
--- a/test-suite/success/Typeclasses.v
+++ b/test-suite/success/Typeclasses.v
@@ -190,7 +190,7 @@ Record Monad {m : Type -> Type} := {
 
 Print Visibility.
 Print unit.
-Arguments unit {m m0 α}.
+Arguments unit {m _ α}.
 Arguments Monad : clear implicits.
 Notation "'return' t" := (unit t).