> On Apr 11, 2019, at 5:29 AM, Ben Boeckel <ben.boeckel_at_[hidden]> wrote:
>
>> On Wed, Apr 10, 2019 at 21:36:25 +0000, Gabriel Dos Reis wrote:
>> Right, but wouldn't the build system would have set up a step to build
>> the BMI for '<foo.h>' because foo.cpp has a dependency on the BMI of
>> <foo.h>? Why is the information required for this to work any
>> different from if you had used #include of a header that contained an
>> import-declaration?
>
> Yes, I have to set up the rule to build the BMI, but if nothing imports
> the header, I don't want to build the BMI. Imagine that I import one
> single Boost header. Should I have to wait for the build system to build
> a BMI for every single (public) Boost header? I hope not.

Right - and I am not suggesting that.

>
>> I guess I am trying to understand why we need to generalize this to
>> header units...
>
> It's required for creating an efficient build (one where only what needs
> built is built).

I am for efficient build. But, I guess part of the issue I am having here is how the build system treats dependencies on headers. Let me mull over this a bit more.

Part of my worry is that we end up with a very elaborate scheme.

>
> --Ben