inconsistency between manifest-based and directory-based runfiles implementation about rules_go HOT 2 CLOSED

Just re-read the library interface doc:

Runfiles strategies:
Manifest-based: reads the runfiles manifest file to look up runfiles.
Directory-based: appends the runfile's path to the runfiles root. The client is responsible for checking that the resulting path exists.

I guess this is correct according to spec then.

Looking at the Java implementation, it uses the manifest-based approach only on Windows [ref]. In the same spirit, i wonder if it would be better to default to directory-based strategy first when both .runfiles_manifest and .runfiles directories exist?

from rules_go.

The .runfiles directory always exists, but it may be empty or contain only the binary itself. That makes it difficult to decide whether it is safe to use when the manifest exists. Basing the decision on the OS is also difficult, as we may be in a host != exec build and Windows can optionally use runfiles directories as well.

Given how tricky all of this is, I would prefer to change this only if there are concrete performance benefits and we understand all the implications.

I may work on the spec in the future and agree that this inconsistency is a major pain point.

from rules_go.