remove a pub publish warning, or make the warnings configurable per-package

[devoncarew]

We encountered this publish warning in our publishing automation (context):

 Package validation found the following potential issue:
  * The name of "lib/builder.dart", "builder", should match the name of the package, "shelf_router_generator".
    This helps users know what library to import.

This is for a package where we do want to use lib/builder.dart for the entry-point instead of lib/shelf_router_generator.dart as the warning would suggest.

Can we either:

• remove this warning from the set we generate (false positives are expensive), or
• make warnings configurable per-package, so a package can intentional suppress them?

The configuration might be done in a section of the pubspec file or in some other durable location. Thanks!

[sigurdm]

Yeah - publication warnings are not very ergonomic.

I am a bit wary of building a way of ignoring specific warnings, as that would involve quite some formalization of the set of warnings. (not saying it is bad, just a big task).

I wonder if we could push this warning into the analyzer as a hint/lint. The analyzer already has the // ignore: mechanism in place, and we run dart analyze as a publication validation step.

The rule is implemented here:

pub/lib/src/validator/name.dart

Line 22 in e1fbda7

if (libraries.length == 1) {

If there is only a single exposed library, we want it to be named the same as the package.

I think this is usually right, and maintains a somewhat nice convention (though personally I would have preferred a fixed name for the main library, but that is too late).

@pq do you think this would work as a hint/lint/warning?

We could also downgrade this warning to an "info", such that it doesn't block the upload, but it might still be seen by enough uploaders that it has some merit.

@jonasfj might have opinions when he is back

[pq]

do you think this would work as a hint/lint/warning?

I think so. We have a file_names lint that enforces lowercase_with_underscores so there's precedent in hanging diagnostics off the first offset in a file.