Re: racing srfi-18 threads

Re: racing srfi-18 threads

[Neil Jerram]

Neil Jerram <neil@ossau.uklinux.net> writes:

> Tristan Colgate <tcolgate@gmail.com> writes:
>
>> A similar, though much simpler, test case exhibits the same problem.
>> I'll raise a bug. report and
>> include a log of the issue.
>
> Hi Tristan,
>
> Thanks for reporting this.  FWIW I've checked that it still happens
> (exactly as you've described) with the current Git HEAD.  Not much clue
> yet about the underlying problem or fix, but I'll keep looking.

What happens is that each of the spawned threads throws an exception
before it gets going.  If I run test-broken in the (srfi srfi-18)
module, with the following code added at the end:

(write threads)
(newline)
(for-each (lambda (t)
	    (write (thread->exception t))
	    (newline))
	  threads)
(for-each thread-join! threads)

I see:

===============
build thread
build thread
build thread
build thread
build thread
build thread

(#<thread 3081010064 (8d308a0)> #<thread 3064224656 (8d305c0)> #<thread 3064224656 (8d30450)> #<thread 3064224656 (8d302e0)> #<thread 3064224656 (8d30170)> #<thread 3064224656 (8d30000)>)
((uncaught-exception) wrong-type-arg "with-exception-handler" "Wrong type argument: ~S" (#<program 8e13ac0>) ())
((uncaught-exception) wrong-type-arg "with-exception-handler" "Wrong type argument: ~S" (#<program 8e13a00>) ())
((uncaught-exception) wrong-type-arg "with-exception-handler" "Wrong type argument: ~S" (#<program 8e13940>) ())
((uncaught-exception) wrong-type-arg "with-exception-handler" "Wrong type argument: ~S" (#<program 8e13890>) ())
((uncaught-exception) wrong-type-arg "with-exception-handler" "Wrong type argument: ~S" (#<program 8e13800>) ())
((uncaught-exception) wrong-type-arg "with-exception-handler" "Wrong type argument: ~S" (#<program 8e13740>) ())
===============

This is apparently because of srfi-18.scm's `with-exception-handler'
including

    (check-arg-type thunk? thunk "with-exception-handler")

It seems that when run under the VM, (thunk? thunk) => #f.

scm_thunk_p depends on scm_i_program_arity, which depends on
scm_program_arities, and adding this -

       (write ((@ (system vm program) program-arities) thunk))
       (newline)
       (write (procedure-property thunk 'arity))
       (newline)

- into srfi-18's make-thread code, I get

#f
#f

every time.

So I think this is a good point to report and ask if this makes sense.
Andy / Ludo, can it be correct for (program-arities PROGRAM) to be #f ?
If not, any idea what is the root cause of this?

Another question here is why the thread-join! doesn't cause the uncaught
thread exceptions to be raised on the main thread.  I'll look further
into that.

Thanks,
        Neil

Re: racing srfi-18 threads

[Andy Wingo]

On Mon 16 Nov 2009 23:16, Neil Jerram <neil@ossau.uklinux.net> writes:

> It seems that when run under the VM, (thunk? thunk) => #f.

Ugly. Thanks for debugging this, Neil.

> Andy / Ludo, can it be correct for (program-arities PROGRAM) to be #f
> ?

Well... no, I don't think so. Looks like a bug to me.

> Thanks,

Thank you!

Andy
-- 
http://wingolog.org/

Re: racing srfi-18 threads

[Neil Jerram]

Andy Wingo <wingo@pobox.com> writes:

> On Mon 16 Nov 2009 23:16, Neil Jerram <neil@ossau.uklinux.net> writes:
>
>> It seems that when run under the VM, (thunk? thunk) => #f.
>
> Ugly. Thanks for debugging this, Neil.

I now have it down to this: a program compiled inside the RHS of a
define-syntax form apparently has no meta; whereas the same program
compiled outside a define-syntax form does have meta:

(use-modules (system vm program))

(define-syntax race
  (syntax-rules ()
    ((_ n)
     (begin
       (write (cons 'race (program-meta (lambda () n))))
       (newline)))))

(race 3)

(begin
  (write (cons 'race (program-meta (lambda () n))))
  (newline))

|=

(race . #f)
(race . #<program 944ccf0>)

I'll keep following this through, but if you have any idea where the
root problem is, please say.

       Neil

Re: racing srfi-18 threads

[Neil Jerram]

Neil Jerram <neil@ossau.uklinux.net> writes:

> I now have it down to this: a program compiled inside the RHS of a
> define-syntax form apparently has no meta; whereas the same program
> compiled outside a define-syntax form does have meta:

Apologies for taking so long to follow up on this!

I've been trying to remedy my lack of detailed understanding about how
compilation works, and that has led me to wondering whether and how we
will provide similar debugging facilities for compiled code as we have
in 1.8.x for interpreted code.

One option would be not to take the 1.8.x approach at all (i.e. using
special hooks from the core of the evaluator/VM) but instead rely on
instrumenting code at runtime.  I had a quick play with this and it
seems quite simple and promising.  For example, here's a way of tracing
when a procedure is called, and its return values:

(define-macro (trace proc)
  `(let ((proc ,proc))
     (set! ,proc
	   (lambda args
	     (display ";TRACE: ")
	     (display ',proc)
	     (display " ")
	     (write args)
	     (newline)
	     (call-with-values
		 (lambda ()
		   (apply proc args))
	       (lambda results
		 (for-each (lambda (result)
			     (display ";TRACE: ")
			     (display ',proc)
			     (display " => ")
			     (write result)
			     (newline))
			   results)
		 (apply values results)))))))

Yes, I know I should write that with define-syntax instead. :-)

Then, with that in my .guile:

neil@arudy:~/SW/Guile/git$ meta/uninstalled-env guile
Guile Scheme interpreter 0.5 on Guile 1.9.5
Copyright (C) 2001-2008 Free Software Foundation, Inc.

Enter `,help' for help.
scheme@(guile-user)> (trace (@@ (system base compile) compile-fold))
;;; note: source file /home/neil/SW/Guile/git/module/language/glil/compile-assembly.scm
;;;       newer than compiled /home/neil/SW/Guile/git/module/language/glil/compile-assembly.go
;;; found fresh local cache at /home/neil/SW/Guile/git/cache/guile/ccache/1.9-0.L-LE-4/home/neil/SW/Guile/git/module/language/glil/compile-assembly.scm.go
scheme@(guile-user)> (trace (@@ (language glil compile-assembly) compile-assembly))
;TRACE: (@@ (system base compile) compile-fold) ((#<program compile-tree-il (x e opts)> #<program compile-glil (x e opts)> #<program compile-asm (x e opts)> #<program compile-bytecode (assembly env . opts)> #<program compile-objcode (x e opts)>) (trace (@@ (language glil compile-assembly) compile-assembly)) #<directory (guile-user) 9947b98> (()))
;TRACE: (@@ (system base compile) compile-fold) => #<objcode 9b9f800>
;TRACE: (@@ (system base compile) compile-fold) => #<directory (guile-user) 9947b98>
;TRACE: (@@ (system base compile) compile-fold) => #<directory (guile-user) 9947b98>
scheme@(guile-user)> (+ 3 4 5)
;TRACE: (@@ (system base compile) compile-fold) ((#<program compile-tree-il (x e opts)> #<program compile-glil (x e opts)> #<program compile-asm (x e opts)> #<program compile-bytecode (assembly env . opts)> #<program compile-objcode (x e opts)>) (+ 3 4 5) #<directory (guile-user) 9947b98> (()))
;TRACE: (@@ (language glil compile-assembly) compile-assembly) (#<glil (program () (source ((breakpoint . #f) (line . 2) (column . 0) (filename . #f))) (std-prelude 0 0 #f) (label :LCASE113) (const 3) (const 4) (const 5) (call add 2) (source ((breakpoint . #f) (line . 2) (column . 0) (filename . #f))) (call add 2) (source ((breakpoint . #f) (line . 2) (column . 0) (filename . #f))) (call return 1))>)
;TRACE: (@@ (language glil compile-assembly) compile-assembly) (#<glil (program () (const (() ((0 2 . 0)) ((6 15 0)))) (call return 1))>)
;TRACE: (@@ (language glil compile-assembly) compile-assembly) => (load-program () 25 #f (make-eol) (make-int8:0) (make-int8 2) (make-int8:0) (cons) (cons) (list 0 1) (make-int8 6) (make-int8 15) (make-int8:0) (list 0 3) (list 0 1) (list 0 3) (return))
;TRACE: (@@ (language glil compile-assembly) compile-assembly) => (load-program ((:LCASE113 . 6)) 16 (load-program () 25 #f (make-eol) (make-int8:0) (make-int8 2) (make-int8:0) (cons) (cons) (list 0 1) (make-int8 6) (make-int8 15) (make-int8:0) (list 0 3) (list 0 1) (list 0 3) (return)) (assert-nargs-ee 0 0) (reserve-locals 0 0) (make-int8 3) (make-int8 4) (make-int8 5) (add) (add) (return) (nop))
;TRACE: (@@ (system base compile) compile-fold) => #<objcode 9bd3f40>
;TRACE: (@@ (system base compile) compile-fold) => #<directory (guile-user) 9947b98>
;TRACE: (@@ (system base compile) compile-fold) => #<directory (guile-user) 9947b98>
12
scheme@(guile-user)> 

That seems quite nice and useful.

(SLIB has stuff like this too.  I wonder if it would just work.)

We should be able to do breakpoints like this too, using either the
command line or the GDS debugger - although I'm not sure how much of the
stack inspection facilities will immediately work.  I'll try that next.

I don't see how single-stepping could easily be implemented this way,
though.  I think that may require hooks in the VM.  But then again,
would single stepping through VM operations be useful and comprehensible
anyway?

All thoughts welcome, as ever...

      Neil

Re: racing srfi-18 threads

[Neil Jerram]

Neil Jerram <neil@ossau.uklinux.net> writes:

> [...]  wondering whether and how we
> will provide similar debugging facilities for compiled code as we have
> in 1.8.x for interpreted code.

Some more thoughts here, to try to build a complete new picture.

Things that we can do quite nicely in 1.8.x are:
- stack inspection
  - seeing the frames and what is happening, for context
  - mapping back to source code
  - querying variable values
  - evaluating an expression in a frame's local environment
- breakpoints
- tracing
- single stepping.

We should try to include profiling and code coverage in the picture.

My thoughts for doing this in 2.0 are as follows.

- Single stepping, breakpoints, code coverage and tracing can all be
  done by instrumenting code - either in Scheme, or with lower-level
  language ops that somehow get inserted as code is compiled.

- The more I think about this, the more it seems clearly preferable to
  the 1.8.x evaluator traps model - i.e. where there are ways of
  _marking_ code in some way, and the evaluator or VM calls out to a
  hook when it sees one of these marks.  It's a simpler solution, and...

- It in particular removes the need for most of the complexity that we
  have in 1.8.x's (ice-9 debugging traps).  The complexity is mostly
  about wanting to say 'do THIS when you start executing procedure FOO',
  but the lowlevel traps interface not allowing us to specify either
  'THIS' or 'when you start executing procedure FOO' precisely.  (All we
  can say is 'call a globally-defined function when you start executing
  any of the currently marked procedures'.)

- I'm not clear how this interacts with optimisation...  What
  happens when an optimisation reorders or eliminates code with a
  breakpoint?  How do we present this to the user?  It feels like a
  soluble problem though.

- Instrumentation-based single-stepping would be more like edebug than
  what we have in 1.8.x - i.e. the mode of operation would probably be
  to instrument the whole of the body of a given function.  But I think
  that would be fine (and consistent for our plan for future emacs
  domination :-)).  (In contrast, 1.8.x single-stepping doesn't require
  prior instrumentation, and allows stepping over function boundaries.)

- An unfortunate consequence of psyntax is backtraces being harder to
  read, and to correlate back to source code, because of all the
  alpha-renamed variables.  When paused at a breakpoint, I think this
  also makes it harder for the user to ask what the value of a given
  variable is.  Is there anything we can do about this - such as mapping
  all the variable names back to their names in the original source
  code?

- As one of Andy's eval commits says, we don't have local-eval any more
  and so can't currently do "evaluate in a stack frame".  I suppose the
  most important case here is querying local variable values; is there a
  reasonable solution for that?

Any thoughts on that?

> (SLIB has stuff like this too.  I wonder if it would just work.)

(Currently blocked by SLIB not loading at all in 1.9/2.0:

scheme@(guile-user)> (use-modules (ice-9 slib))
;;; note: autocompilation is enabled, set GUILE_AUTO_COMPILE=0
;;;       or pass the --no-autocompile argument to disable.
;;; compiling /usr/share/slib/guile.init
;;; WARNING: compilation of /usr/share/slib/guile.init failed:
;;; key syntax-error, throw args (sc-expand "~a in ~a" ("unexpected syntax" define) #f)
ERROR: In procedure sc-expand:
ERROR: unexpected syntax in define

)

Regards,
        Neil

Re: racing srfi-18 threads

[Andy Wingo]

Hi Neil,

On Wed 02 Dec 2009 22:46, Neil Jerram <neil@ossau.uklinux.net> writes:

> Neil Jerram <neil@ossau.uklinux.net> writes:
>
>> I now have it down to this: a program compiled inside the RHS of a
>> define-syntax form apparently has no meta; whereas the same program
>> compiled outside a define-syntax form does have meta:
>
> Apologies for taking so long to follow up on this!

Dude, my apologies for not tracking mail... This particular bug was due
to a bad check in compile-assembly.scm:make-meta. Fixed in
8986ff7ae9dcaae79d3ab262c360a6cbbc86c263.

> I've been trying to remedy my lack of detailed understanding about how
> compilation works, and that has led me to wondering whether and how we
> will provide similar debugging facilities for compiled code as we have
> in 1.8.x for interpreted code.

I would hope so! 

> One option would be not to take the 1.8.x approach at all (i.e. using
> special hooks from the core of the evaluator/VM) but instead rely on
> instrumenting code at runtime.  I had a quick play with this and it
> seems quite simple and promising.

That is indeed useful, and robust.

Some thoughts...

1. Something like your trace procedure does seem to be quite useful. 

2. At the same time, we should be able to trace any procedure at runtime
without modifying it -- whether by using a different VM (possible) or by
enabling hooks on the current VM.

3. When it comes time to have native compilation, things get trickier.
Did you see today's LWN article on ftrace? It looks really really sweet.

  http://lwn.net/SubscriberLink/365835/07f149ad48a74856/ -- and do
  subscribe if you're not already, &c &c.

The compiler could easily instrument interesting pieces of code with
NOPs, and a tracer could patch the code at runtime.

Even more easy would be having the compiler produce actual calls to
trace procedures at various points, for serious debugging.

Also there are hardware breakpoints, but that's trickier.

Dunno, my thoughts here are scattered.

> Yes, I know I should write that with define-syntax instead. :-)

Probably yes :)

> We should be able to do breakpoints like this too, using either the
> command line or the GDS debugger - although I'm not sure how much of the
> stack inspection facilities will immediately work.  I'll try that
> next.

There is the break instruction. We have code for inspecting the local
vars of a stack frame -- see program.scm and frame.scm.

> I don't see how single-stepping could easily be implemented this way,
> though.  I think that may require hooks in the VM.  But then again,
> would single stepping through VM operations be useful and comprehensible
> anyway?

Not usually no. But sometimes. More often expression-level stepping
would be nice, or at least stepping function calls.

Cheers,

Andy
-- 
http://wingolog.org/

Debugging infrastructure (was Re: racing srfi-18 threads)

[Neil Jerram]

Andy Wingo <wingo@pobox.com> writes:

> Hi Neil,
>
> On Wed 02 Dec 2009 22:46, Neil Jerram <neil@ossau.uklinux.net> writes:
>
>> Neil Jerram <neil@ossau.uklinux.net> writes:
>>
>>> I now have it down to this: a program compiled inside the RHS of a
>>> define-syntax form apparently has no meta; whereas the same program
>>> compiled outside a define-syntax form does have meta:
>>
>> Apologies for taking so long to follow up on this!
>
> Dude, my apologies for not tracking mail... This particular bug was due
> to a bad check in compile-assembly.scm:make-meta. Fixed in
> 8986ff7ae9dcaae79d3ab262c360a6cbbc86c263.

Thanks!  I think I was on the right track, but still some way away from
this solution.

>> One option would be not to take the 1.8.x approach at all (i.e. using
>> special hooks from the core of the evaluator/VM) but instead rely on
>> instrumenting code at runtime.  I had a quick play with this and it
>> seems quite simple and promising.
>
> That is indeed useful, and robust.
>
> Some thoughts...
>
> 1. Something like your trace procedure does seem to be quite useful. 
>
> 2. At the same time, we should be able to trace any procedure at runtime
> without modifying it -- whether by using a different VM (possible) or by
> enabling hooks on the current VM.

Why?  Does this approach provide specific benefits, compared to
instrumentation?

> 3. When it comes time to have native compilation, things get trickier.
> Did you see today's LWN article on ftrace? It looks really really sweet.

Certainly the idea of having a ring buffer for tracing is good, instead
of outputting the trace directly.

Otherwise I think the main concept here is tracing function entry and
exit.

Mapped into Guile, my feeling/guess is that it is relatively unlikely
that we will need to support this at the C level of the VM.  The macro
that I posted adds instrumentation at the Scheme level, and I agree that
that is probably not optimal.  But we could also implement tracing by
inserting code in one of the compiler passes; that would be similar to
ftrace's use of -pg.  And we could arrange the inserted code to output
to a ring buffer, and so that its output was subject to current runtime
configuration.

>   http://lwn.net/SubscriberLink/365835/07f149ad48a74856/ -- and do
>   subscribe if you're not already, &c &c.

I am already.

> The compiler could easily instrument interesting pieces of code with
> NOPs, and a tracer could patch the code at runtime.
>
> Even more easy would be having the compiler produce actual calls to
> trace procedures at various points, for serious debugging.

Ah, sorry, I didn't read ahead before writing above!  So obviously I
agree.

> Also there are hardware breakpoints, but that's trickier.

So I'd suggest doing the immediately tractable stuff first, and seeing
if a need for hardware breakpoints arises.

>> We should be able to do breakpoints like this too, using either the
>> command line or the GDS debugger - although I'm not sure how much of the
>> stack inspection facilities will immediately work.  I'll try that
>> next.
>
> There is the break instruction.

Interesting.  But does anything currently generate that instruction?

> We have code for inspecting the local
> vars of a stack frame -- see program.scm and frame.scm.

Cool, thanks.

>> I don't see how single-stepping could easily be implemented this way,
>> though.  I think that may require hooks in the VM.  But then again,
>> would single stepping through VM operations be useful and comprehensible
>> anyway?
>
> Not usually no. But sometimes. More often expression-level stepping
> would be nice, or at least stepping function calls.

Agreed.  I wrote a bit more about this in my followup post.

Anyway, how should we proceed for playing with this area?  I should have
a little time over Christmas, and can use that to work on debugging and
the manual.  Do you want/plan to spend time on debugging too, or are you
happy just to discuss and provide input?

Regards,
        Neil

Re: racing srfi-18 threads

[Andy Wingo]

Just pulling things back together here:

On Thu 03 Dec 2009 23:52, Neil Jerram <neil@ossau.uklinux.net> writes:

> Neil Jerram <neil@ossau.uklinux.net> writes:
>
>> [...]  wondering whether and how we
>> will provide similar debugging facilities for compiled code as we have
>> in 1.8.x for interpreted code.
>
> Some more thoughts here, to try to build a complete new picture.
>
> Things that we can do quite nicely in 1.8.x are:
> - stack inspection
>   - seeing the frames and what is happening, for context
>   - mapping back to source code
>   - querying variable values
>   - evaluating an expression in a frame's local environment
> - breakpoints
> - tracing
> - single stepping.
>
> We should try to include profiling and code coverage in the picture.

We can do all this in 1.9/2.0 now, except breakpoints and single
stepping, and evaluating in the context of a frame isn't accessible
right now. I want to fix the breakpoints and stepping situation soonish,
though.

Andy
-- 
http://wingolog.org/