repository_ctx is unable to get another repository_rule's attributes

[jin]

I'm currently trying to implement a couple of repository_rule rules in Skylark.

My maven_rules.bzl file looks like this:

def maven_jar_impl(repo_ctx):
   ...

maven_jar_attrs = { 
  server: attr.label(), # refers to a maven_server target
  ... 
}

maven_jar = repository_rule(
  maven_jar_impl, 
  attrs=maven_jar_attrs, 
  ...
)

def maven_server_impl(repo_ctx):
  ...

maven_server_attr = { 
  url=attr.string(),
  ...
}

maven_server = repository_rule(
  maven_server_impl,
  attrs=maven_server_attr,
  ...
)

How do you retrieve the url attribute of the referenced maven_server target within maven_jar_impl, hopefully via repo_ctx? I've looked around and found native.existing_rules() and native.existing_rule(name), but they do not work in the analysis phase.

[damienmg]

There is no way to do that currently.

[damienmg]

Oh sorry, it should be possible. native.existing_rule should just work but only at the macros level, so you can wrap maven_jar as a macro that calls the real maven_jar passing along the url to the actual maven_jar

[damienmg]

I will leave it as a documentation issue.

[kchodorow]

you can wrap maven_jar as a macro that calls the real maven_jar passing along the url to the actual maven_jar

What?

[jin]

I got this to work:

In WORKSPACE

load(":maven_rules.bzl", "maven_jar_macro")

# this has to come before maven_jar_macro declarations
maven_server( 
    name = "maven_uk_server",
    url = "http://uk.maven.org/maven2"
)

maven_jar_macro(
    name = "com_google_guava_guava",
    artifact = "com.google.guava:guava:18.0",
    sha1 = "cce0823396aa693798f8882e64213b1772032b09",
    repository = "http://repo.exist.com/maven2",
    server = "maven_uk_server"
)

maven_rules.bzl

def maven_jar_macro(name, artifact, sha1, server, repository, visibility=None):
  print(native.existing_rule(server)["url"]) # this prints "http://uk.maven.org/maven2"
  maven_jar( # maven_jar is the actual repository_rule
      name = name,
      visibility = visibility,
      artifact = artifact,
      sha1 = sha1,
      repository = repository
  )

However, the maven_server rule now needs to come before the maven_jar_macro rule for it to be included in native.existing_rules. This makes the WORKSPACE file rules positional.

[jin]

@damienmg Why is native.existing_rules available in the macros level but not during WORKSPACE analysis? Wouldn't the rules and macros be parsed into existing_rules dictionary prior to WORKSPACE analysis?

[damienmg]

something along this should work:

maven_server = repository_rule(...)
_maven_jar = repository_rule(... attrs = { "server_url": attr.string(),  ... } ...)

def maven_jar(server=None, **kwargs):
  if server:
    server_rule = native.existing_rule(server)
    server_url = server_rule["url"]
  else:
    server_url = "repo.maven.apache.org"
  _maven_jar(server = server_url, **kwargs)

[damienmg]

Ok cross-post. So seems like @jin understood what I meant :)

This is how skylark is made. It was not necessary to access them from other rule before. We could probably add it to repository_ctx. This need to be thought through though.

[jin]

It still feels a little fragile and not obvious that the maven_server declaration needs to come before maven_jar.

What are the implications of allowing native.existing_rules() to be called during the analysis phase?

[damienmg]

There is no real action we can take on this one, closing.