[cmake-developers] Questions about coding conventions

[Daniel Pfeifer]

On Sun, Jun 12, 2016 at 4:26 PM, Ben Boeckel <ben.boeckel at kitware.com> wrote:
> On Fri, Jun 10, 2016 at 15:30:05 +0200, Daniel Pfeifer wrote:
>> Passing and returning strings: We have `const char*`, `std::string`,
>> and `std::string const&`. Unifying this can affect performance.
>> Storing `const char*` in a `std::string` creates a (potentially
>> unneded) copy (assuming it is not null).
>
> I went through and changed all `const char*` paramters to `std::string
> const&` where the pointer was given to a string within the function
> (unless it was an error path or such). This allows eliding another
> allocation.

I am aware of that. In case of passing `foo.c_str()`, this moved the
creation of the unnecessary copy from inside the function to the call
site, resulting in more allocations rather than less, assuming the
function is called more than once. It is important to run clang-tidy
after each such refactoring to remove the now unnecessary c_str()
calls.

>> `const char*` can distinguish
>> invalid from empty. Do we actually need this distinction?
>
> It's used all over the place.

Can you show an example? To be clear: We are looking for a function,
that has a code path for `str == NULL` and a *different* codepath for
`str[0] = '\0'`.