Partial fix of Issue 8384 - std.conv.to should allow conversion betwe… #4199
Conversation
|
| @@ -875,11 +875,19 @@ T toImpl(T, S)(S value) | |||
| ()@trusted{ memcpy(result.ptr, value.ptr, value.length); }(); | |||
| return cast(T) result; | |||
| } | |||
| else static if (isPointer!S && is(S : const(char)*)) | |||
| else static if (isPointer!S && (is(S : const(char)*) || is(S : const(wchar)*) || is(S : const(dchar)*))) | |||
There was a problem hiding this comment.
the rest of the constraint should use std.traits.isSomeChar
There was a problem hiding this comment.
isSomeChar checks if it's a char of some sort, not a char* of some sort.
There was a problem hiding this comment.
isSomeChar!(ElementEncodingType!(S))
| @@ -29929,7 +29929,7 @@ else version(Windows) | |||
| } | |||
|
|
|||
|
|
|||
| this(string name, TIME_ZONE_INFORMATION tzInfo) @safe immutable pure | |||
| this(string name, TIME_ZONE_INFORMATION tzInfo) @trusted immutable pure | |||
There was a problem hiding this comment.
I'm more than a little unsure about this. The conversion on line 29934 is definitely wrong as is (conversion from wchar* to string does not work, and fails in ugly ways at runtime). The question is, should this function be @trusted? @System?
Should it even be using this conversion? Am I right to address the issue here or should that be a separate pull?
Regarding to the safety of using to!string(wchar*), MSDN claims in one place that DaylightName will always be zero-terminated. If this is true, this should be safe. However, there are many versions of the info page on TIME_ZONE_INFORMATION, and not all of them include this information. The constructor is private though, and I haven't found a case where it could be called with invalid data.
There was a problem hiding this comment.
I believe that the consensus at this point is that if the @safety depends on the argument being correct, then it should be marked as @trusted. The reality of the matter in this case is that either will work. The constructor is private, and the function calling it is already @trusted. So, just mark it as @trusted.
And if the Windows API gives a wchar* that isn't zero-terminated, then it's just plain broken, since it's a C API. But actually, in this case, it doesn't even matter, because the TIME_ZONE_INFORMATION is constructed by std.datetime in the caller, and it's explicitly zero-terminated.
…en any pair of string/wstring/dstring/char*/wchar*/dchar*
|
LGTM |
|
Auto-merge toggled on |
…en any pair of string/wstring/dstring/char_/wchar_/dchar*
std.conv.to currently fails uglily with 'object.Error@(0): array cast misalignment.' when casting from a char* to any wider string type, and does not support any kind of conversion from e.g. a wchar* to wchar[]. This should fix both those issues.
Note that Issue 8384 also mentions conversions from {X}char[] to {Y}char*. I have not tackled that problem, as std.conv.to does not currently support any kind of conversion in that direction.