[Lldb-commits] [PATCH] D40537: Simplify UUID::IsValid()

[Greg Clayton via lldb-commits]

> On Nov 28, 2017, at 10:24 AM, Zachary Turner <zturner at google.com> wrote:
> 
> 
> 
> On Tue, Nov 28, 2017 at 10:18 AM Greg Clayton <clayborg at gmail.com <mailto:clayborg at gmail.com>> wrote:
>> On Nov 27, 2017, at 10:11 PM, Zachary Turner <zturner at google.com <mailto:zturner at google.com>> wrote:
>> 
>> As an aside, I don't really like this class.  For example, You can currently assign a UUID[16] to a UUID[20].  That doesn't make a lot of sense to me.
> 
> What about an invalid UUID[0] being assigned with a valid UUID[16] or UUID[20]? Why doesn't this make sense? I don't follow.
> 
> Nothing is invalid, I just think it's better and expresses the intent more clearly if you can only assign between UUIDs of the same size.  For example, If the UUID class were templated on size, then there would not even be such thing as a UUID[0] or a "universally invalid UUID".  There would be an "invalid 16-byte UUID" and an "invalid 20-byte UUID", and those would be different things.
>  
> 
>> 
>> As a future cleanup, I think this class should probably be a template such as UUID<N>, and then internally it can store a std::array<uint8_t, N>.  And we can static_assert that N is of a known size if we desire.
> 
> UUID values are objects contained as members inside of other objects. They all default to start with no preconceived notion of what the UUID should be. IMHO the UUID class is just fine and needs to be able to represent any UUID, from empty uninitialized ones, and be able to be assigned and changed at will.
> 
> 
> Is there ever a use case for changing the number of bytes in a UUID?  If you're working with 16-byte UUIDs, does it ever actually happen that now you have a 20-byte UUID?  Can you imagine a use case currently where an N-byte UUID is being compared against an M-byte UUID in a real-world scenario?  If the answer is no, then it may as well be enforced by the compiler. 

The ObjectFile class has a "UUID m_uuid;" member that any object file can fill in. Right now mach-o files have 16 byte UUIDs. ELF files can have 20 bytes UUIDs (build ID) or 4 byte UUIDs (debug info CRC if no build ID is around, and these are current represented as 20 byte UUIDs with just the first 4 bytes filled in. So no, we can't enforce this using the compiler. I don't see a need to change way from a byte buffer that has the max number of bytes needed for any currently supported UUID (20 right now). 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-commits/attachments/20171128/0137d37a/attachment.html>