[Issue 11721] regression: std.utf.toUTFindex throws error instead of exception

[d-bugmail at puremagic.com]

https://d.puremagic.com/issues/show_bug.cgi?id=11721



--- Comment #1 from Kenji Hara <k.hara.pg at gmail.com> 2013-12-14 08:38:56 PST ---
Until 2.063, std.utf.toUTFindex was non-template function. In each release,
Phobos library is built with -release switch, so the asserts in non-template
code is always removed.
So, std.utf.toUTFindex had called release version of std.utf.stride, and it did
not throw AssertError "Past the end of the UTF-8 sequence".
But, stride function is @safe function, so array indexing was always checked
boundaries, and could throw RangeError.

In 2.064, std.utf.toUTFindex was changed to template function.

https://github.com/D-Programming-Language/phobos/commit/87a2ee7f4d4e4959cf4db9a1710beac67c00129c#diff-27410a0be96392a06647e61c73131b64L882

So, if you compile your code without -release, the assertion is properly
checked and it throws AssertError.
If you compile the code with -release, the whole execution still succeeds with
2.064 and git-head.

The main point is, today most of Phobos funcitons have no specifications
against erroneous inputs. In this case, if you give invalid index for
std.utf.stride via std.utf.toUTFindex, what happens is
"implementation-dependent".
(Note: std.utf.toUTFindex is marked as @safe, so it won't occur "undefined
behavior", but it is not relevant with this issue.)

The most better way I think is, changing your unittest code to
assertThrown!Error(a.slice(2,8));

-- 
Configure issuemail: https://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------