READ-CHAR-NO-HANG hangs on SB-SYS:WAIT-UNTIL-FD-USABLE

[Michał "phoe" Herda]

READ-CHAR-NO-HANGS hung on me with 100% CPU usage after I tried to read from a freshly closed socket.
This socket was freshly closed though, so I suspect some kind of race condition.
If I read the docstrings at https://github.com/sbcl/sbcl/blob/e98378ca004ef6d101b384f6c3130f24b7a1bc1f/src/code/serve-event.lisp#L215 correctly, then the call (SB-SYS:WAIT-UNTIL-FD-USABLE 7 :INPUT NIL NIL) has no timeout (will hang indefinitely). I don't think that's the correct behavior in case of READ-CHAR-NO-HANG that shouldn't hang at all.

I'm on SBCL 1.4.9.

-------------------------------------

Interrupt from Emacs
   [Condition of type SIMPLE-ERROR]

Restarts:
 0: [CONTINUE] Continue from break.
 1: [RETRY] Retry SLIME REPL evaluation request.
 2: [*ABORT] Return to SLIME's top level.
 3: [ABORT] abort thread (#<THREAD "repl-thread" RUNNING {1002EC83E3}>)

Backtrace:
  0: ("bogus stack frame")
      [No Locals]
  1: (SB-SYS:WAIT-UNTIL-FD-USABLE 7 :INPUT NIL NIL)
      Locals:
        SB-IMPL::DIRECTION = :INPUT
        SB-IMPL::FD = 7
        SB-IMPL::SERVE-EVENTS = NIL
        SB-IMPL::SIGNALP = NIL
        SB-IMPL::STOP-SEC = NIL
        SB-IMPL::STOP-USEC = NIL
        TIMEOUT = NIL
        SB-IMPL::TO-SEC = NIL
        SB-IMPL::TO-USEC = NIL
  2: (SB-IMPL::REFILL-INPUT-BUFFER #<SB-SYS:FD-STREAM for "socket 127.0.0.1:44262, peer: 127.0.0.1:38347" {1007E3F583}>)
      Locals:
        COUNT = 0
        ERRNO = 0
        FD = 7
        STREAM = #<SB-SYS:FD-STREAM for "socket 127.0.0.1:44262, peer: 127.0.0.1:38347" {1007E3F583}>
  3: (SB-IMPL::INPUT-CHAR/UTF-8 #<SB-SYS:FD-STREAM for "socket 127.0.0.1:44262, peer: 127.0.0.1:38347" {1007E3F583}> T NIL)
      Locals:
        DECODE-BREAK-REASON = NIL
        EOF-ERROR = T
        EOF-VALUE = NIL
        IBUF = #S(SB-IMPL::BUFFER :SAP #.(SB-SYS:INT-SAP #X7F32EA374000) :LENGTH 4096 :HEAD 0 :TAIL 0)
        SIZE = NIL
        STREAM = #<SB-SYS:FD-STREAM for "socket 127.0.0.1:44262, peer: 127.0.0.1:38347" {1007E3F583}>
      Catch-tags:
        SB-IMPL::EOF-INPUT-CATCHER
  4: (READ-CHAR-NO-HANG #<SB-SYS:FD-STREAM for "socket 127.0.0.1:44262, peer: 127.0.0.1:38347" {1007E3F583}> T NIL #<unused argument>)
      Locals:
        SB-IMPL::%FRC-INDEX% = 512
        SB-IMPL::%FRC-STREAM% = #<SB-SYS:FD-STREAM for "socket 127.0.0.1:44262, peer: 127.0.0.1:38347" {1007E3F583}>
        SB-IMPL::EOF-ERROR-P = T
        SB-IMPL::EOF-VALUE = NIL
        STREAM = #<SB-SYS:FD-STREAM for "socket 127.0.0.1:44262, peer: 127.0.0.1:38347" {1007E3F583}>
  5: (PEEK-CHAR-NO-HANG #<SB-SYS:FD-STREAM for "socket 127.0.0.1:44262, peer: 127.0.0.1:38347" {1007E3F583}> T NIL NIL)
      Locals:
        EOF-ERROR-P = T
        EOF-VALUE = NIL
        INPUT-STREAM = #<SB-SYS:FD-STREAM for "socket 127.0.0.1:44262, peer: 127.0.0.1:38347" {1007E3F583}>
        RECURSIVE-P = NIL
  ...

[Michał "phoe" Herda (phoe-krk)]

When I compile a function in slime (via C-c C-c), the loop is broken and the execution continues. I assume this is because SBCL serves some other events and then it discovers that the FD is (un)usable and it is capable of proceeding with computation.

[Michał "phoe" Herda (phoe-krk)]

I was attempting to read from the same socket in two different threads which caused READ-CHAR-NO-HANG to hang nonetheless. This fixed the issue.