https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99624
Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |WAITING
--- Comment #1 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
> While building the ada compiler the address sanitizer detecst an
> heap-buffer-overflow in namet.adb line 157:
>
> Index : constant Int := Name_Entries.Table (Id).Name_Chars_Index;
>
> because Id=-399990000
The table is declared like this though:
package Name_Entries is new Table.Table (
Table_Component_Type => Name_Entry,
Table_Index_Type => Valid_Name_Id'Base,
Table_Low_Bound => First_Name_Id,
Table_Initial => Alloc.Names_Initial,
Table_Increment => Alloc.Names_Increment,
Table_Name => "Name_Entries");
with:
First_Name_Id : constant Name_Id := Names_Low_Bound + 2;
-- Subscript of first entry in names table
subtype Valid_Name_Id is Name_Id range First_Name_Id .. Name_Id'Last;
-- All but No_Name and Error_Name
and:
Names_Low_Bound : constant := -399999999;
so Id = -399990000 is well within the allowed range (9997 0-based index).
Are you sure that Id is not equal to Names_Low_Bound, which would be the -2
0-based index and, therefore, -32 bytes since the size of Name_Entry is 16?
On the other hand, this would mean that:
pragma Assert (Is_Valid_Name (Id));
would have triggered because the compiler is supposed to be configured with
assertions enabled on the mainline, so I'm quite at a loss here.
It looks like the address sanitizer is miscompiling the Ada compiler?