LPBSTR parameters should retain metadata about their being BSTRs

[AArnott]

The QueryPathOfRegTypeLib function is documented as having a last parameter that is an [out] LPBSTR. But in the metadata, we just get [out] uint16** as the parameter type. This blocks projections from providing friendly methods (e.g. returning a string).
Can the metadata be fixed as has been done with other APIs that accept BSTR for example, so that projections can offer better APIs?

[kennykerr]

Agreed. Note that the docs are wrong:

The caller allocates the BSTR that is passed in, and must free it after use.

Should be something like:

The callee allocates the BSTR and the caller must free it after use.

[AArnott]

Whoa, that's quite the doc mistake. How do you know it's a doc bug?
I was surprised that if the caller was supposed to allocate the string that the docs didn't indicate how large the buffer must be or that the length wasn't passed in as an argument.

[kennykerr]

How do you know it's a doc bug?

I looked up the implementation because it sounded wrong. 😏

[AArnott]

I sent feedback on that doc. Do you suppose they should use SysFreeString?

[kennykerr]

Yes, the caller should free with SysFreeString.

[mikebattista]

Per feedback from @kennykerr, remapping all LPBSTR occurrences to BSTR* would be better than introducing PBSTR.