[C++20] [Modules] Record BMI Hash which is consistent if the interfaces doesn't change

[ChuanqiXu9]

The motivating example comes from #70569.

For example,

// a.cppm
export module a;
export int a() {
    return 43;
}

// use.cc
import a;
int use() {
    return a();
}

After we change the implementation of a() from return 43; to return 44;, we can avoid recompiling use.cc to use.o since the interface doesn't change.
This is pretty helpful to improve the user's experience by avoiding unnecessary recompilations as much as possible.

[llvmbot]

@llvm/issue-subscribers-clang-modules

Author: Chuanqi Xu (ChuanqiXu9)

The motivating example comes from https://github.com//issues/70569.

For example,

// a.cppm
export module a;
export int a() {
    return 43;
}

// use.cc
import a;
int use() {
    return a();
}

After we change the implementation of a() from return 43; to return 44;, we can avoid recompiling use.cc to use.o since the interface doesn't change.
This is pretty helpful to improve the user's experience by avoiding unnecessary recompilations as much as possible.

[dwblaikie]

(might be worth linking to the discourse thread from here so we don't duplicate the discussion)

[ChuanqiXu9]

Good idea. We've discussed the ideas here: https://discourse.llvm.org/t/rfc-c-20-modules-introduce-thin-bmi-and-decls-hash/74755

[ChuanqiXu9]

This is somehow covered by #86912 with -fexperimental-modules-reduced-bmi.

Now for the example with reduced BMI:

//--- A.cppm
export module A;
export int funcA() {
    return 43;
}

//--- A.v1.cppm
export module A;

export int funcA() {
    return 43;
}

//--- B.cppm
export module B;
import A;

export int funcB() {
    return funcA();
}

//--- C.cppm
export module C;
import B;

...

Now although B will almost always get changed if module A changes, but the BMI of B won't change if it doesn't affect the interface. Then we can avoid the recompilation of C.

Note that the original idea that avoid changing the BMI of A is considered not good now since we think the source location is significant.

Then I'd like to close the issue since we may not record and read a BMI hash but generate the same BMI directly.