missing default method for gray-streams:stream-finish-output

[Anton Vodonosov]

Test case:

(ql:quickload :flexi-streams)

(let* ((binary-stream (flexi-streams:make-in-memory-output-stream)))
  (finish-output binary-stream))

The value #<FLEXI-STREAMS::VECTOR-OUTPUT-STREAM {1E566F3}> is not of type #<STANDARD-CLASS FLEXI-STREAMS:FLEXI-OUTPUT-STREAM {7D84CE}>.
   [Condition of type SIMPLE-TYPE-ERROR]

Backtrace:
  0: (#<FUNCTION {1C9E6C0}> #<SIMPLE-TYPE-ERROR {FC82A6}> #<FUNCTION {1C9E6C0}>)
  1: (APPLY #<FUNCTION {1C9E6C0}> (#<SIMPLE-TYPE-ERROR {FC82A6}> #<FUNCTION {1C9E6C0}>))
  2: (SYSTEM::RUN-HOOK SYSTEM::*INVOKE-DEBUGGER-HOOK* #<SIMPLE-TYPE-ERROR {FC82A6}> #<FUNCTION {1C9E6C0}>)
  3: (INVOKE-DEBUGGER #<SIMPLE-TYPE-ERROR {FC82A6}>)
  4: (GRAY-STREAMS:STREAM-FINISH-OUTPUT #<FLEXI-STREAMS::VECTOR-OUTPUT-STREAM {1E566F3}>)
  5: (SYSTEM::%FINISH-OUTPUT #<FLEXI-STREAMS::VECTOR-OUTPUT-STREAM {1E566F3}>)
  6: (FINISH-OUTPUT #<FLEXI-STREAMS::VECTOR-OUTPUT-STREAM {1E566F3}>)

This happens because there is no default method for stream-finish-output.
According to the Gray proposal, default method should exist and do nothing.

The fix:

Index: src/org/armedbear/lisp/gray-streams.lisp
===================================================================
--- src/org/armedbear/lisp/gray-streams.lisp    (revision 14464)
+++ src/org/armedbear/lisp/gray-streams.lisp    (working copy)
@@ -314,6 +314,10 @@
   (declare (ignore stream))
   nil)

+(defmethod stream-finish-output (stream)
+  (declare (ignore stream))
+  nil)
+
 (defmethod stream-clear-output (stream)
   (declare (ignore stream))
   nil)

Also, ABCL handles the absence of default method not very correctly.
It should signal "no applicable method". What happens here is that flexi-streams overrides
stream-finish-output for one of its classes - FLEXI-STREAMS:FLEXI-OUTPUT-STREAM.
So, I think ABCL sees that there is only one variant of method,
and tries to apply this method, despite we pass object of another class.
And here typecast happen.

[Mark Evenson]

With patch the ansi-compiled tests fail:

The value &REST is not of type FIXNUM.
   [Condition of type TYPE-ERROR]
Backtrace:
  0: (#<FUNCTION {2CBEE72C}> #<TYPE-ERROR {5A80EE6E}> #<FUNCTION {2CBEE72C}>)
  1: (APPLY #<FUNCTION {2CBEE72C}> (#<TYPE-ERROR {5A80EE6E}> #<FUNCTION {2CBEE72C}>))
  2: (SYSTEM::RUN-HOOK SYSTEM::*INVOKE-DEBUGGER-HOOK* #<TYPE-ERROR {5A80EE6E}> #<FUNCTION {2CBEE72C}>)
  3: (INVOKE-DEBUGGER #<TYPE-ERROR {5A80EE6E}>)
  4: org.armedbear.lisp.Lisp.error(Lisp.java:382)
  5: org.armedbear.lisp.Lisp.type_error(Lisp.java:435)
  6: org.armedbear.lisp.Fixnum.getValue(Fixnum.java:293)
  7: org.armedbear.lisp.LispCharacter$6.execute(LispCharacter.java:350)
  8: org.arme

[Anton Vodonosov]

That's very strange. How to can I run this test?

[Anton Vodonosov]

Ok, I found how to run the tests:

ant -Dabcl.build.incremental=true test.ansi.compiled

[Anton Vodonosov]

For me, with or without the patch, ant test.ansi.compiled gives the same result:

12 out of 21707 total tests failed:
(CALL-NEXT-METHOD.ERROR.1
CALL-NEXT-METHOD.ERROR.2
MAKE-CONDITION.3
MAKE-CONDITION.4
SXHASH.8
MAP.48
TYPE-OF.1
TYPE-OF.4
MAKE-CONCATENATED-STREAM.30
PRINT.RANDOM-STATE.1
PPRINT-LOGICAL-BLOCK.17
TRACE.8)

Tested on SVN revision 14464

[Mark Evenson]

After redoing my compilation/test environment from scratch, my error has disappeared.

Committed as part of r14605