[asdf-devel] Possible problem with check-compile-results

[Faré]

That sounds like a problem with Babel and/or a library it uses — and
indeed reminds me of a previous bug in Alexandria — is your Alexandria
up to date? It's using a symbol as a format string, which happens to
work on some implementations at some (un)safety levels.


—♯ƒ • François-René ÐVB Rideau •Reflection&Cybernethics• http://fare.tunes.org
Don't expect coherence from a loosely-defined group of people. Even
individuals have trouble with coherence. Argue issues. Change incentives.


On Fri, Aug 23, 2013 at 8:40 PM, Robert Goldman <rpgoldman at sift.net> wrote:
> I would be grateful if someone could confirm this for me.
>
> I see this in quickloading BABEL on a relatively recent SBCL
> 1.1.7.112.master.104-d62278d
>
> Babel crashes on compiling:
>
> caught ERROR:
> ;   (during macroexpansion of (DEFINE-UCS :UCS-2LE ...))
> ;   #:~A-CODE-POINT-COUNTER fell through ETYPECASE expression.
> ;   Wanted one of (STRING SIMPLE-STRING).
>
> ; in: DEFINE-UCS :UCS-2BE
> ;     (BABEL-ENCODINGS::DEFINE-UCS :UCS-2BE 2 :BE 65536)
>
> After that, some logic in UIOP'S CHECK-LISP-COMPILE-RESULTS goes
> pear-shaped.
>
> I believe the problem starts here:
>
> (with-upgradability ()
>   (defmacro define-convenience-action-methods
>       (function formals &key if-no-operation if-no-component
> operation-initargs)
>     (let* ((rest (gensym "REST"))
>            (found (gensym "FOUND"))
>            (keyp (equal (last formals) '(&key)))
>            (formals-no-key (if keyp (butlast formals) formals))
>            (len (length formals-no-key))
>            (operation 'operation)
>            (component 'component)
>            (opix (position operation formals))
>            (coix (position component formals))
>            (prefix (subseq formals 0 opix))
>            (suffix (subseq formals (1+ coix) len))
>            (more-args (when keyp `(&rest ,rest &key &allow-other-keys))))
>       (assert (and (integerp opix) (integerp coix) (= coix (1+ opix))))
>       (flet ((next-method (o c)
>                (if keyp
>                    `(apply ',function , at prefix ,o ,c , at suffix ,rest)
>                    `(,function , at prefix ,o ,c , at suffix))))
>         `(progn
>            (defmethod ,function (, at prefix (,operation symbol) component
> , at suffix , at more-args)
>              (if ,operation
>                  ,(next-method
>                    (if operation-initargs ;backward-compatibility with
> ASDF1's operate. Yuck.
>                        `(apply 'make-operation ,operation
> :original-initargs ,rest ,rest)
>                        `(make-operation ,operation))
>                    `(or (find-component () ,component) ,if-no-component))
>                  ,if-no-operation))
>            (defmethod ,function (, at prefix (,operation operation)
> ,component , at suffix , at more-args)
>              (if (typep ,component 'component)
>                  (error "No defined method for ~S on
> ~/asdf-action:format-action/"
>                         ',function (cons ,operation ,component))
>                  (if-let (,found (find-component () ,component))
>                     ,(next-method operation found)
>                     ,if-no-component))))))))
>
>
> The ERROR created here, with its embedded funcall to
> ASDF-ACTION:FORMAT-ACTION seems to do something bad when the format
> string is repackaged in UIOP/LISP-BUILD:CHECK-LISP-COMPILE-RESULTS:
>
>     (unless output
>       (error 'compile-file-error :context-format context-format
> :context-arguments context-arguments))
>
> possibly in this REPORT-FUNCTION:
>
>     (:report (lambda (c s)
>                (format s (compatfmt "~@<~A~@[ while ~?~]~@:>")
>                        (or (compile-condition-description c) (type-of c))
>                        (compile-condition-context-format c)
>                        (compile-condition-context-arguments c)))))
>
> I am not particularly adept in the SBCL debugger wrapped by SLDB,
> especially where one of these
> error-wrapped-in-error-wrapped-in-ASDF-wrapped-in-quicklisp contexts, so
> I would be grateful if someone would  try to replicate this (try a clean
> QL install of BABEL).
>
> Note that the error comes with SBCL's built-in ASDF, not bleeding-edge ASDF.
>
> Thanks!
> r
>
> --
> Robert P. Goldman
> Principal Scientist
> Smart Information Flow Technologies (d/b/a SIFT, LLC)
>
> 211 N. First St., Suite 300
> Minneapolis, MN 55401
>
> Voice:    (612) 326-3934
> Email:    rpgoldman at SIFT.net
>
>