wip: moar builders

Author: rickeylev

.bazelversion

@@ -1 +1 @@
-8.x
+8.0.1

api_notes.md

@@ -0,0 +1,90 @@
+
+
+## For rule.cfg
+
+optional vs rule-union vs cfg-union?
+
+* Optional: feels verbose. Requires extra get() calls.
+* Optional: seems harder to detect value
+* Rule-union: which API feels verbose.
+* Cfg-Union: seems nicest? More underlying impl work though.
+
+```
+# optional
+# Rule.cfg is type Optional[TransitionBuilder | ConfigNone | ConfigTarget]
+
+r = RuleBuilder()
+cfg = r.cfg.get()
+if <cfg is TransitionBuilder>:
+  cfg.inputs.append(...)
+elif <cfg is config.none>:
+  ...
+elif <cfg is config.target>:
+  ...
+else: error()
+
+# rule union
+# Rule has {get,set}{cfg,cfg_none,cfg_target} functions
+# which() tells which is set.
+# Setting one clears the others
+
+r = RuleBuilder()
+which = r.cfg_which()
+if which == "cfg":
+  r.cfg().inputs.append(...)
+elif which == "cfg_none":
+  ...
+elif which == "cfg_target":
+  ...
+else: error
+
+# cfg union (1)
+# Rule.cfg is type RuleCfgBuilder
+# RuleConfigBuilder has {get,set}{implementation,none,target}
+# Setting one clears the others
+
+r = RuleBuilder()
+
+if r.cfg.implementation():
+  r.cfg.inputs.append(...)
+elif r.cfg.none():
+  ...
+elif r.cfg.target():
+  ...
+else:
+  error
+
+# cfg-union (2)
+# Make implementation attribute polymorphic
+impl = r.cfg.implementation()
+if impl == "none":
+  ...
+elif impl == "target":
+  ...
+else: # function
+  r.cfg.inputs.append(...)
+
+# cfg-union (3)
+# impl attr is an Optional
+impl = r.cfg.implementation.get()
+... r.cfg.implementation.set(...) ...
+```
+
+## Copies copies everywhere
+
+To have a nicer API, the builders should provide mutable lists/dicts/etc.
+
+But, when they accept a user input, they can't tell if the value is mutable or
+not. So they have to make copies. Most of the time, the values probably _will_
+be mutable (why use a builder if its not mutable?). But its an easy mistake to
+overlook that a list is referring to e.g. some global instead of a local var.
+
+So, we could defensively copy, or just document that a mutable input is
+expected, and behavior is undefined otherwise.
+
+Alternatively, add a function to py_internal to detect immutability, and it'll
+eventually be available in some bazel release.
+
+## Collections of of complex objects
+
+Should these be exposed as the raw collection, or a wrapper? e.g.