[COMMITTED] ada: Small adjustment to special resolution of membership test
Checks
Commit Message
From: Eric Botcazou <ebotcazou@adacore.com>
It's needed because, in GNAT, universal_integer does not cover all the
values of all the supported integer types.
gcc/ada/
* sem_res.adb (Resolve_Membership_Op): Adjust latest change.
Tested on x86_64-pc-linux-gnu, committed on master.
---
gcc/ada/sem_res.adb | 18 +++++++++++-------
1 file changed, 11 insertions(+), 7 deletions(-)
@@ -10105,11 +10105,11 @@ package body Sem_Res is
then
T := Etype (R);
- -- If the left operand is of a universal numeric type and the right
- -- operand is not, we do not resolve the operands to the tested type
- -- but to the universal type instead. If not conforming to the letter,
- -- it's conforming to the spirit of the specification of membership
- -- tests, which are typically used to guard a specific operation and
+ -- If the type of the left operand is universal_integer and that of the
+ -- right operand is smaller, then we do not resolve the operands to the
+ -- tested type but to universal_integer instead. If not conforming to
+ -- the letter, it's conforming to the spirit of the specification of
+ -- membership tests, which are typically used to guard an operation and
-- ought not to fail a check in doing so. Without this, in the case of
-- type Small_Length is range 1 .. 16;
@@ -10122,10 +10122,14 @@ package body Sem_Res is
-- the function Is_Small_String would fail a range check for strings
-- larger than 127 characters.
+ -- The test on the size is required in GNAT because universal_integer
+ -- does not cover all the values of all the supported integer types,
+ -- for example the large values of Long_Long_Long_Unsigned.
+
elsif not Is_Overloaded (L)
- and then Is_Universal_Numeric_Type (Etype (L))
+ and then Etype (L) = Universal_Integer
and then (Is_Overloaded (R)
- or else not Is_Universal_Numeric_Type (Etype (R)))
+ or else RM_Size (Etype (R)) < RM_Size (Universal_Integer))
then
T := Etype (L);