Re: A stack overflow problem.

[Ludovic Courtès]

Mark H Weaver <address@hidden> skribis:

> Hi Ludovic,
>
>> The problem here is that, when the second overflow occurs, that reserve
>> has already been used, so it just aborts.
>>
>> Commit 70057f3 in ‘stable-2.0’ fixes that (will be in 2.0.10.)
>
> Thanks for fixing this, but there's still a slight problem.  After a VM
> overflow is caught and the user is left in the debugger, typing ",q"
> sometimes causes another stack overflow.
>
> To reproduce this problem, type the exact commands shown in
> <http://bugs.gnu.org/15065>.  Using v2.0.9-118-g70057f3 on either x86_64
> or mips64el (with N32 ABI), this is what I see:
>
> [...]
> scheme@(guile-user)> (integral cube 0.0 1.0 0.01)
> $1 = 0.24998750000000042
> scheme@(guile-user)> (integral cube 0.0 1.0 0.001)
> $2 = 0.249999875000001
> scheme@(guile-user)> (integral cube 0.0 1.0 0.0001)
> <unnamed port>:7:11: In procedure sum:
> <unnamed port>:7:11: Throw to key `vm-error' with args `(vm-run "VM: Stack 
> overflow" ())'.
>
> Entering a new prompt.  Type `,bt' for a backtrace or `,q' to continue.
> scheme@(guile-user) [1]> ,q
> ice-9/boot-9.scm:65:2: In procedure abort-to-prompt:
> ice-9/boot-9.scm:65:2: Throw to key `vm-error' with args `(vm-run "VM: Stack 
> overflow" ())'.
>
> Entering a new prompt.  Type `,bt' for a backtrace or `,q' to continue.
> scheme@(guile-user) [1]> ,q
> scheme@(guile-user)> 

The second overflow occurs because the stack reserve is reinstated
before the handler is invoked, so the handler doesn’t have enough stack
space to run:

--8<---------------cut here---------------start------------->8---
scheme@(guile-user) [1]> ,q
ice-9/boot-9.scm:65:2: In procedure abort-to-prompt:
ice-9/boot-9.scm:65:2: Throw to key `vm-error' with args `(vm-run "VM: Stack 
overflow" ())'.

Entering a new prompt.  Type `,bt' for a backtrace or `,q' to continue.
scheme@(guile-user) [1]> ,bt
In ice-9/boot-9.scm:
   157:177216 (catch quit #<procedure 16adcf0 at system/repl/repl.scm:181:18 
()> #<procedure ff4220 at system/repl/repl.scm:200…> …)
In system/repl/repl.scm:
   191:317215 (#<procedure 16adcf0 at system/repl/repl.scm:181:18 ()>)
In ice-9/boot-9.scm:
   157:177214 (catch #t #<procedure 16bd0c0 at 
system/repl/error-handling.scm:96:6 ()> #<procedure 1105e20 at 
system/repl/error…> …)
In system/vm/trap-state.scm:
    172:47213 (with-default-trap-handler #<procedure debug-trap-handler (frame 
trap-idx trap-name)> #<procedure 1105d60 at syst…> …)
In ice-9/boot-9.scm:
   1650:47212 (%start-stack #t #<procedure 16bd120 at 
system/repl/repl.scm:192:32 ()>)
   1655:97211 (#<procedure 16bd030 ()>)
In system/repl/repl.scm:
    148:47210 (with-stack-and-prompt #<procedure 1105e40>)
In ice-9/boot-9.scm:
   1650:47209 (%start-stack #t #<procedure 1105b20 at 
system/repl/repl.scm:149:33 ()>)
   1655:97208 (#<procedure 16bd000 ()>)
In current input:
   2:40957207 (integral #<procedure cube (x)> 0.0 1.0 1.0e-4)
--8<---------------cut here---------------end--------------->8---

I tried using a making the ‘quit’ handler in repl.scm a pre-unwind
handler but that doesn’t help.  Ideas?

Ludo’.