Support adding paths to include key

[mathstuf]

Currently, include is a complete list. It would be nice to have a way to just specify additional includes rather than having to list all paths when I just want one extra file to trigger a rebuild (namely a custom derive reading files).

[mathstuf]

Hmm. For some reason, when setting include myself to a list of paths, it still isn't triggering a rebuild when changing a listed input file.

[lights0123]

That's not what include does—it's just an opposite of exclude. There's this feature in build scripts, but that doesn't necessarily trigger a rebuild of the whole crate.

[epage]

@mathstuf Could you give a reproduction case? It sounds like you are talking about package.include which is just about what files to include when running cargo package or cargo publish and doesn't have anything to do with rebuilding.

[mathstuf]

We ended up doing this:

https://gitlab.kitware.com/utils/ghostflow-director/-/blob/e1c6c0b828baa8a6a573006a9378fb9dc09b89f6/build.rs#L46-50

because using include or rerun-if-changed= (seems to?) disable the rerun-if-changed= for the paths it does so by default for.

See also:

• Make cargo rerun if input files change graphql-rust/graphql-client#117 (oh, linked above already; I chased links back to here)
• Dependencies to query file and schema file are not tracked by the compiler when deriving GraphQLQuery graphql-rust/graphql-client#215
• https://users.rust-lang.org/t/opening-files-in-proc-macros/11387/6

[epage]

because using include or rerun-if-changed=

What do you mean by include?

[epage]

btw #4587 is the issue for dealing with built-in heuristics.

[mathstuf]

What do you mean by include?

Apparently package.include, per a link from the URLO thread I linked: https://doc.rust-lang.org/cargo/reference/manifest.html#the-exclude-and-include-fields

[weihanglo]

From my understanding, this is essentially a duplicate of #4587, as they all talk about keeping the default file change detection when adding a new rerun-if-changed instruction. The main difference is this one has brought the other solution — make package.include additive. While this could fix the original issue, it complicates the meaning of package.include. It mainly serves for cargo package and has a side effect on default rebuild detection coincidentally.

I think we should merge this into #4587 so we won't lose any bit of different possible solutions. Closing and thank you