Optional CARGO_MANIFEST_DIR

[Ahmed-Ali]

In askama_shared/src/lib.rs#L36 this will cause the Template macro to fail if the CARGO_MANIFEST_DIR doesn't exist;
This works well with the default rustc tool chain; however, this also means if for whatever reason CARGO_MANIFEST_DIR doesn't exist, the crate won't work. (proc-macro panic)

Any chance this can be optional instead?

[djc]

What do you think it should do if the manifest dit is not available?

[Ahmed-Ali]

Looking at the code now, doesn't seem like it is reasonable to be able to have optional "Root" path; but maybe we can provide it through a template parameter or something; currently we can provide path relative to the manifest root dir, but maybe alternatively we can provide an absolute path?

which I guess will be way less portable -_-

[djc]

What is your use case anyway? What build system are you using if not Cargo? It doesn't seem too big of a deal to just set CARGO_MANIFEST_DIR in your environment in some other way.

[Ahmed-Ali]

We are using BUCK; there is away to set the env variables in there; but it felt bit hacky; yet I can't think for better solution; the absolute path doesn't seem like an improvement

[djc]

If there is some other environment variable in a Buck context that we can work off of, I'd be open to having that as a fallback.

[Ahmed-Ali]

doesn't seem like we have an equivalent really, for some reason. The only way that worked just fine, is to create a filegroup buck target, set its location is the value of CARGO_MANIFEST_DIR and place the templates there. Yet the challenge here is any other project uses askama (directly or transitively, might be required to do some additional work arounds).

[Ahmed-Ali]

few folks suggested to use include_str instead. Also another approach might be to avoid invoking this code path if one uses source instead of path in the macro; with some document mentioning that path depends on CARGO_MANIFEST_DIR

[djc]

Now I'm wondering if include_str!() works in attribute values...

[vallentin]

I'm not entirely up to date on this, but as far as I recall using e.g. include_str!() in an attribute is possible, but it is unstable and requires enabling the extended_key_value_attributes feature. See rust-lang/rust#78835 and rust-lang/rust#78837.

[sunshowers]

You can use include_str in attributes on stable through a declarative macro, I believe. It's a bit of a hack but it works well. See https://github.com/facebookincubator/cargo-guppy/blob/main/tools/determinator/src/rules.rs#L159-L187 for an example, and https://docs.rs/determinator/0.4.0/determinator/rules/struct.DeterminatorRules.html#associatedconstant.DEFAULT_RULES_TOML as the result.