Compiler fails with “The value NIL is not of type SB-C::NODE” on certain combinations of FLET, INLINE, APPLY

[Matthias Benkard]

The following function definition makes the compiler fail with the error “The value NIL is not of type SB-C::NODE.”

(defun f (x &rest args)
  (flet ((g (&rest args)
           (and x args)))
    (declare (inline g))
    #'g
    (apply #'g args)))

This happens both on SBCL 1.1.8 on Darwin (AMD64) and on 1.0.57.0.debian on GNU/Linux (AMD64).

*features* of 1.1.8 on Darwin:

(:SWANK :ASDF2 :ASDF :ASDF-UNICODE :ALIEN-CALLBACKS :ANSI-CL :ASH-RIGHT-VOPS
 :BSD :C-STACK-IS-CONTROL-STACK :COMMON-LISP :COMPARE-AND-SWAP-VOPS
 :COMPLEX-FLOAT-VOPS :CYCLE-COUNTER :DARWIN :DARWIN9-OR-BETTER :FLOAT-EQL-VOPS
 :GENCGC :IEEE-FLOATING-POINT :INLINE-CONSTANTS :INODE64 :LINKAGE-TABLE
 :LITTLE-ENDIAN :MACH-EXCEPTION-HANDLER :MACH-O :MEMORY-BARRIER-VOPS
 :MULTIPLY-HIGH-VOPS :OS-PROVIDES-BLKSIZE-T :OS-PROVIDES-DLADDR
 :OS-PROVIDES-DLOPEN :OS-PROVIDES-PUTWC :OS-PROVIDES-SUSECONDS-T
 :PACKAGE-LOCAL-NICKNAMES :RAW-INSTANCE-INIT-VOPS :SB-AFTER-XC-CORE
 :SB-CORE-COMPRESSION :SB-DOC :SB-EVAL :SB-LDB :SB-PACKAGE-LOCKS :SB-SIMD-PACK
 :SB-SOURCE-LOCATIONS :SB-TEST :SB-THREAD :SB-UNICODE :SB-XREF-FOR-INTERNALS
 :SBCL :STACK-ALLOCATABLE-CLOSURES :STACK-ALLOCATABLE-FIXED-OBJECTS
 :STACK-ALLOCATABLE-LISTS :STACK-ALLOCATABLE-VECTORS
 :STACK-GROWS-DOWNWARD-NOT-UPWARD :UD2-BREAKPOINTS :UNIX
 :UNWIND-TO-FRAME-AND-CALL-VOP :X86-64)

*features* of 1.0.57.0.debian:

(:X86-64 :UNIX :ELF :LINUX :LARGEFILE :GENCGC :STACK-GROWS-DOWNWARD-NOT-UPWARD
 :C-STACK-IS-CONTROL-STACK :LINKAGE-TABLE :COMPARE-AND-SWAP-VOPS
 :UNWIND-TO-FRAME-AND-CALL-VOP :RAW-INSTANCE-INIT-VOPS
 :STACK-ALLOCATABLE-CLOSURES :STACK-ALLOCATABLE-VECTORS
 :STACK-ALLOCATABLE-LISTS :STACK-ALLOCATABLE-FIXED-OBJECTS :ALIEN-CALLBACKS
 :CYCLE-COUNTER :COMPLEX-FLOAT-VOPS :FLOAT-EQL-VOPS :INLINE-CONSTANTS
 :MEMORY-BARRIER-VOPS :MULTIPLY-HIGH-VOPS :LITTLE-ENDIAN :OS-PROVIDES-DLOPEN
 :OS-PROVIDES-DLADDR :OS-PROVIDES-PUTWC :OS-PROVIDES-BLKSIZE-T
 :OS-PROVIDES-SUSECONDS-T :OS-PROVIDES-GETPROTOBY-R :OS-PROVIDES-POLL
 :IEEE-FLOATING-POINT :SB-SOURCE-LOCATIONS :SB-EVAL :SB-UNICODE
 :SB-PACKAGE-LOCKS :SB-LDB :SB-TEST :SB-DOC :SBCL :COMMON-LISP :ANSI-CL
 :SB-CORE-COMPRESSION :SB-THREAD :SB-FUTEX)

[Paul Khuong (pvk)]

Looks like we have an XEP for an &optional entry point that's not marked as being allocated a closure anywhere.

Backtrace:

The value NIL is not of type SB-C::NODE.
   [Condition of type TYPE-ERROR]

Restarts:
 0: [RETRY] Retry SLIME REPL evaluation request.
 1: [*ABORT] Return to SLIME's top level.
 2: [ABORT] Abort thread (#<THREAD "new-repl-thread" RUNNING {100887BC53}>)

Backtrace:
  0: (SB-C::NODE-HOME-LAMBDA NIL) [tl,external]
  1: ((LABELS SB-C::FLOOD :IN SB-C::CLOSE-OVER) #<SB-C::PHYSENV :LAMBDA #<SB-C::CLAMBDA :%SOURCE-NAME SB-C::.ANONYMOUS. :%DEBUG-NAME (SB-C::XEP (SB-C::LOCAL-INLINE (FLET G :IN F))) :KIND :EXTERNAL :TYPE #<..
  2: ((LABELS SB-C::FLOOD :IN SB-C::CLOSE-OVER) #<SB-C::PHYSENV :LAMBDA #<SB-C::CLAMBDA :%SOURCE-NAME SB-C::.ANONYMOUS. :%DEBUG-NAME (SB-C::&MORE-PROCESSOR (SB-C::LOCAL-INLINE (FLET G :IN F))) :KIND :OPTIO..
  3: ((LABELS SB-C::FLOOD :IN SB-C::CLOSE-OVER) #<SB-C::PHYSENV :LAMBDA #<SB-C::CLAMBDA :%SOURCE-NAME SB-C::.ANONYMOUS. :%DEBUG-NAME (SB-C::LOCAL-INLINE (FLET G :IN F)) :KIND :OPTIONAL :TYPE #<SB-KERNEL:BU..
  4: (SB-C::CLOSE-OVER #<SB-C::LAMBDA-VAR :%SOURCE-NAME X {100650AAE3}> #<SB-C::PHYSENV :LAMBDA #<SB-C::CLAMBDA :%SOURCE-NAME SB-C::.ANONYMOUS. :%DEBUG-NAME (SB-C::LOCAL-INLINE (FLET G :IN F)) :KIND :OPTIO..
  5: (SB-C::%ADD-LAMBDA-VARS-TO-CLOSURES #<SB-C::CLAMBDA :%SOURCE-NAME F :%DEBUG-NAME NIL :KIND :OPTIONAL :TYPE #<SB-KERNEL:BUILT-IN-CLASSOID FUNCTION (read-only)> :WHERE-FROM :DEFINED :VARS (X ARGS #:REST..
  6: (SB-C::ADD-LAMBDA-VARS-AND-LET-VARS-TO-CLOSURES #<SB-C::CLAMBDA :%SOURCE-NAME F :%DEBUG-NAME NIL :KIND :OPTIONAL :TYPE #<SB-KERNEL:BUILT-IN-CLASSOID FUNCTION (read-only)> :WHERE-FROM :DEFINED :VARS (X..
  7: (SB-C::PHYSENV-ANALYZE #<SB-C:COMPONENT :NAME (FLET G :IN F) {1006542363}>)
  8: (SB-C::COMPILE-COMPONENT #<SB-C:COMPONENT :NAME (FLET G :IN F) {1006542363}>)
  9: (SB-C::%COMPILE (SB-INT:NAMED-LAMBDA F (X &REST ARGS) (BLOCK F (FLET ((G (&REST ARGS) (AND X ARGS))) (DECLARE (INLINE G)) (FUNCTION G) (APPLY (FUNCTION G) ARGS)))) #<SB-C::CORE-OBJECT > :NAME NIL :PAT..

[Paul Khuong (pvk)]

The attached patch looks semi-sane. It seems to me the issue with inlining is that we create a hacked-up copy of the CLAMBDA (without going through full IR1 conversion of a FUNCTION form that'll %allocate-closure), but end up leaving the copy in a :FULL mv-call, and then tons of invariants are lost. The patch is incredibly kludgey and detects when a multiple value combination would not be converted into a MV-LET, and stops before the inlined copy is patched in.

Thoughts?

(This seems to be a common pattern for inlining-related bugs.)

[Paul F. Dietz (paul-dietz)]

I think the random tester was popping up bugs like this some years ago. Is this is duplicate of some existing bug?

[Paul Khuong (pvk)]

Git blame shows a similar issue may have been fixed in 0.8.4.11 (https://github.com/pkhuong/sbcl/commit/11f02398a1a9ccbde847c82fd233e8378e45c29c). That's a different problem.

A text search on launchpad didn't reveal anything promising.

[Stas Boukarev (stassats)]

Reduced test case
(defun f (x)
  (flet ((g (a b)
           (declare (ignore a b))
           x))
    (declare (inline g))
    (g 1 2)
    (multiple-value-call #'g 1 2)))

[Paul F. Dietz (paul-f-dietz)]

The reduced test case does not fail any more, but the original one does.

[Stas Boukarev (stassats)]

None of the tests fail anymore.