Re: [PATCH v5 4/5] c++modules: report imported CMI files as dependencies

[Ben Boeckel <ben.boeckel@kitware.com>]

On Thu, Jul 27, 2023 at 18:13:48 -0700, Jason Merrill wrote:
> On 7/23/23 20:26, Ben Boeckel wrote:
> > Sure, *CMake* knows them, but the *build tool* needs to be told
> > (typically `make` or `ninja`) because it is what is actually executing
> > the build graph. The way this is communicated is via `-MF` files and
> > that's what I'm providing in this patch. Note that `ninja` does not
> > allow rules to specify such dependencies for other rules than the one it
> > is reading the file for.
> 
> But since the direct imports need to be rebuilt themselves if the 
> transitive imports change, the build graph should be the same whether or 
> not the transitive imports are repeated?  Either way, if a transitive 
> import changes you need to rebuild the direct import and then the importer.

I suppose I have seen enough bad build systems that don't do everything
correctly that I'm interested in creating "pits of success" rather than
"well, you didn't get thing X 100% correct, so you're screwed here too".

The case that I think is most likely here is that someone has a
"superbuild" with 3 projects A, B, and C where C uses B and B uses A. At
the top-level the superbuild exposes just "make projectA
projectB projectC"-granularity (rather than a combined build graph; they
may use different build systems) and then users go into some projectC
directly and forget to update projectB after updating projectA (known to
all use the same compiler/flags so that CMI sharing is possible). The
build it still broken, but ideally they get notified in some useful way
when rebuilding the TU rather than…whatever ends up catching the problem
incidentally.

> I guess it shouldn't hurt to have the transitive imports in the -MF 
> file, as long as they aren't also in the p1689 file, so I'm not 
> particularly opposed to this change, but I don't see how it makes a 
> practical difference.

Correct. The P1689 shouldn't even know about transitive imports (well,
maybe from header units?) as it just records "I saw an `import`
statement" and should never look up CMI files (indeed, we would need
another scanning step to know what CMI files to create for the P1689
scan if they were necessary…).

--Ben