PSETF.47 ANSI regression

[Mark Evenson]

According to a bisect caused by ​http://trac.common-lisp.net/armedbear/changeset/14035, but this looks wrong as r14035 merely adds an additional REQUIRE to format.lisp.

[rschlatte]

(setf mask-bit) is undefined until mask-bit is called. Maybe outoloading is to blame?

[ehuelsmann]

Yes. Autoloading *is* to blame. The reasons are documented in ​setf.lisp line 61 through 67 and ​setf.lisp lines 93 through 98:

We used to assume we should resolve the symbol-function's autoloader. However, since DEFSETF and DEFINE-SETF-EXPANDER forms aren't required to be in the same file as the actual function, this doesn't work in the general case. Also: resolving every function for which a SETF form is encountered is undesirable because it can/will cause internal functions to be resolved before they're actually required. (This is the old strategy, btw, which caused circular dependencies in autoloaded code.)

[Mark Evenson]

(In [14369]) Implementation of autoloader for SETF generalized references.

Fixes #296. Fixes #266. Fixes #228.

For forms which set the symbol properties of SETF-EXPANDER or
SETF-FUNCTION to function definitions, places stub of type
AutoloadGeneralizedReference? to be resolved when first invoked.

[Mark Evenson]

(In [14378]) Backport r14369 | mevenson | 2013-02-13 20:01:20 +0100 (Wed, 13 Feb 2013) | 7 lines

Implementation of autoloader for SETF generalized references.

Fixes #296. Fixes #266. Fixes #228.

For forms which set the symbol properties of SETF-EXPANDER or
SETF-FUNCTION to function definitions, places stub of type
AutoloadGeneralizedReference? to be resolved when first invoked.

Does NOT include changes to asdf.