Authoritatively forward declare core types #312
Conversation
From now on, anyone can just include `"au/fwd.hh"`, and get authoritatively correct forward declarations. These will be implicitly tested automatically by adding an `#include` of this new file at the top of every file that has a definition for which we now provide a forward declaration: if we get a forward declaration wrong, we'll get a compiler warning. We also include `fwd.hh` in a few places where it lets us get rid of some manual forward declarations. The rep-named aliases move upstream into `fwd.hh`, because they'll be directly useful for forward declaration use cases, and because you can't "forward declare" an alias: you just have to move the definition upstream. Helps #232. To resolve it completely, we will need to provide a forward declaration file for each individual unit file.
chiphogg
added
the
release notes: ✨ lib (enhancement)
| @@ -17,6 +17,8 @@ | |||
| #include <chrono> | |||
| #include <type_traits> | |||
|
|
|||
| #include "au/fwd.hh" | |||
There was a problem hiding this comment.
I tried to include the forward declaration file in every file whose types are forward declared. This should give us better test coverage for cases where the forward declaration goes out of sync with the true definition, such as if we forward declare as class Zero; but define as struct Zero {...};.
That said, I tried creating exactly this error condition, and I wasn't able to get a warning to come up from //au:zero_test alone. I do get errors from some other test in //...:all, but I get these regardless of whether this file includes the fwd.hh file or not.
All in all: I think it's a good best practice, and I'm going to keep it for that reason, although I'm mystified about why I can't seem to trigger the warnings as easily as I would expect.
From now on, anyone can just include
"au/fwd.hh", and getauthoritatively correct forward declarations. These will be implicitly
tested automatically by adding an
#includeof this new file at the topof every file that has a definition for which we now provide a forward
declaration: if we get a forward declaration wrong, we'll get a compiler
warning.
We also include
fwd.hhin a few places where it lets us get rid ofsome manual forward declarations.
The rep-named aliases move upstream into
fwd.hh, because they'll bedirectly useful for forward declaration use cases, and because you can't
"forward declare" an alias: you just have to move the definition
upstream.
Helps #232. To resolve it completely, we will need to provide a forward
declaration file for each individual unit file.