Hello again,

As often happens, describing the problem in detail prompted my brain to come up with another idea. I changed my output translation configuration to:

  `(:output-translations
    ;; (ABSOLUTE-COMPONENT-DESIGNATOR ABSOLUTE-COMPONENT-DESIGNATOR)
    (,(let* ((device (pathname-device *application-jar-path*))
    (path (merge-pathnames "**/*.*" *application-jar-path*)))
  (make-pathname :defaults path :device device)))
    :ignore-inherited-configuration
    :disable-cache)

And now I don't get the "access denied" error, but it seems that ASDF:INITIALIZE-OUTPUT-TRANSLATIONS never returns...

On 14 December 2015 at 22:39, John Pallister <john@synchromesh.com> wrote:
Hello list,

I'm trying to configure ASDF to load my application while running within the Google App Engine development server (Jetty 6.1). My Ant build script uses ASDF-JAR:PACKAGE to produce a single JAR file containing my application and all dependencies, and copies this to the war/WEB-INF/lib/ directory so that it's in the classpath.

At runtime (in a separate bootstrap.lisp file), following the example of ASDF-JAR:ADD-TO-ASDF[1], I search the application JAR file (obtained by grovelling JAVA:DUMP-CLASSPATH) for the .ASD files and push them all onto ASDF:*CENTRAL-REGISTRY*. So far so good. (Having read the ASDF manual I now know I should either build a .cl-source-registry.cache file into the JAR or compute an ASDF source registry configuration and include that.)

When ASDF starts up it tries to read various configuration files from the filesystem, for both the source registry and for its output translations. This isn't allowed by App Engine. On #lisp, Attila Lendvai suggested that giving ASDF a source registry configuration of '(:SOURCE-REGISTRY :IGNORE-INHERITED-CONFIGURATION) should suppress this behaviour, and it does seem to work.

Now I'm struggling to get the output translations to work. Based on ADD-TO-ASDF again I'm using an output translations configuration of:

`(:output-translations
    (,(merge-pathnames "/**/*.*" *application-jar-path*)) ; => #P"/**/*.*"
    :ignore-inherited-configuration
    :disable-cache)

My *APPLICATION-JAR-PATH* is #P"jar:file:/Volumes/SoftRAID/Users/john/src/synchromesh/forumfeedme/war/WEB-INF/lib/forumfeedme-lisp-0.1.0.jar!/", i.e. a pathname with the JAR path as the PATHNAME-DEVICE.

With this I get the error "access denied ("java.io.FilePermission" "/___jar___file___root___" "read")". This matches the default translations seen in asdf.lisp:WRAPPING-OUTPUT-TRANSLATIONS:

  (defun wrapping-output-translations ()
    `(:output-translations
    ;; Some implementations have precompiled ASDF systems,
    ;; so we must disable translations for implementation paths.
      #+(or #|clozure|# ecl mkcl sbcl)
      ,@(let ((h (resolve-symlinks* (lisp-implementation-directory))))
          (when h `(((,h ,*wild-path*) ()))))
      #+mkcl (,(translate-logical-pathname "CONTRIB:") ())
      ;; All-import, here is where we want user stuff to be:
      :inherit-configuration
      ;; These are for convenience, and can be overridden by the user:
      #+abcl (#p"/___jar___file___root___/**/*.*" (:user-cache #p"**/*.*"))
      #+abcl (#p"jar:file:/**/*.jar!/**/*.*" (:function translate-jar-pathname))
      ;; We enable the user cache by default, and here is the place we do:
      :enable-user-cache))

Although I don't understand exactly how the "JAR file as device" (JFAD) translations are supposed to work, one thing that struck me as odd while investigating things was this:

CL-USER> (make-pathname :device (list #P"/foo.jar"))
#P"jar:file:/foo.jar!/"
CL-USER> (merge-pathnames "**/*.*" *)
#P"jar:file:/foo.jar!/**/*.*"
CL-USER> (merge-pathnames "/**/*.*" **)
#P"/**/*.*"
CL-USER> (pathname-directory **)
(:RELATIVE :WILD-INFERIORS)
CL-USER> (pathname-directory **)
(:ABSOLUTE :WILD-INFERIORS)

That is, merging a relative path with the (absolute) JFAD path retained the default JAR file "device" (and remains a relative path, which ASDF won't accept), but merging an absolute path reset the device component. I've read through the documentation for MERGE-PATHNAMES in the HyperSpec[2] and AFAICT the device component should be copied across. I'm reasonably confident that if it were, things would work better. But I could be (doubly) wrong.

I am trying to figure this out for myself, but I'm stuck for now, so I'm hoping someone (i.e. Mark) can, on reading this, offer some guidance as to where I should go from here. I haven't (yet) tried posting to the ASDF mailing list as this seems like a fairly ABCL-specific issue.

I guess the fallback is to unpack the application JAR and just load the FASLs directly, but (based again on ADD-TO-ASDF) I'm pretty sure that this should work, and it does seem to be the Right Thing.

Any or all comments and/or advice appreciated,

Cheers,

John :^P