# HG changeset patch
# User Paul Boddie <paul@boddie.org.uk>
# Date 1256490675 -3600
# Node ID f8de3c09afd2d3af102a82b6d6b1deabb73e87d3
# Parent  5c26eb35251e1107f357ea2930fc8a25967d1fd2
Added notes on type constraints deduced from attribute usage.
Added a link to a discussion of module-level initialisation restrictions.

diff -r 5c26eb35251e -r f8de3c09afd2 docs/rationale.txt
--- a/docs/rationale.txt	Sun Oct 11 02:40:20 2009 +0200
+++ b/docs/rationale.txt	Sun Oct 25 18:11:15 2009 +0100
@@ -146,3 +146,48 @@
     * Module and class attributes, locals
   * The exception:
     * Instance attributes
+
+Wild idea: isn't AttributeError just evidence of a bug in the program?
+
+  * Note down all attributes used by a particular name during its lifetime
+  * Upon recording name usage, record the location of usage
+  * Use the set of attributes to constrain the range of types that can be used
+    without an AttributeError being raised
+  * Convert all name usage to specific attribute usage (on the range of types
+    found) for all attribute accesses
+
+Attributes on locals
+
+  * Control-flow can make attribute tracking awkward:
+
+    obj.x           # obj must have x
+    if ...:         # (branch)
+        obj = ...   #   obj reset
+        obj.attr    #   obj must have attr
+    else:           # (branch)
+        obj.name    #   obj must have x, name
+                    # (merge)
+                    # obj must have <nothing>
+
+Attributes on locals with loops
+
+  * Loops complicate matters still further:
+
+    obj.x           # obj must have x
+    while ...:      # (branch)
+        obj.y       #   obj must have x, y
+        obj = ...   #   obj reset
+        obj.z       #   obj must have z
+                    # (re-branch)
+                    #   obj must have z, y (obj.y)
+                    #   obj reset (obj = ...)
+                    #   obj must have z, y (obj.z)
+                    # (merge)
+                    # obj must have <nothing>
+
+Attributes on attributes
+
+  * Classes and modules should preserve single interpretations of attributes
+  * Attributes on such attributes should accumulate on the targets
+  * Tracking attributes on such targets is only useful for unknown targets
+  * Instances cannot be relied upon to refer to a coherent set of targets
diff -r 5c26eb35251e -r f8de3c09afd2 docs/related.txt
--- a/docs/related.txt	Sun Oct 11 02:40:20 2009 +0200
+++ b/docs/related.txt	Sun Oct 25 18:11:15 2009 +0100
@@ -11,3 +11,7 @@
 CapPython limits attribute access in order to facilitate code verification:
 
 http://mail.python.org/pipermail/python-dev/2008-September/082475.html
+
+Advocacy for limited module-level initialisation:
+
+http://plope.com/Members/chrism/import_time_side_effects