stack overflow is unreliable

[extemporaneousb]

Simple example below shows a segmentation fault for a recursive function call.

thanks, jim

           _

_ _ ()_ |
() | () () | A fresh approach to technical computing
_ _ | | __ _ |
| | | | | | |/ ` | | Version 0.0.0-prerelease
| | || | | | (| | | Commit c84deda (2012-02-28 19:39:42)
/ |_'|||__'| |
|__/ |

julia> f(n) = if (n == 0) 0 else f(n-1) + n
end

julia> f(10)
55

julia> f(10000000000)
Segmentation fault: 11

[StefanKarpinski]

I just get a stack overflow:

julia> f(n) = if (n == 0) 0 else f(n-1) + n end

julia> f(10)
55

julia> f(10000000000)
error: stack overflow

Can you do two things: 1) send system details, 2) try a newer julia version.

[StefanKarpinski]

Ah, interesting. I get different errors on different runs. Sometimes I get a stack overflow, other times segfault, still others illegal hardware instructions. There's definitely something messy going on here.

[StefanKarpinski]

Basically, I think this comes down to stack overflow detection being somewhat unstable.

[JeffBezanson]

Leaving an asynch signal handler is not required to work.
On Mar 2, 2012 10:02 AM, "Stefan Karpinski" <
[email protected]>
wrote:

Basically, I think this comes down to stack overflow detection being
somewhat unstable.

---

Reply to this email directly or view it on GitHub:
#503 (comment)

[leviboyles]

I think I'm running into the same issue with object printing. I have a doubly-linked tree structure and if I forget the ";" in the julia shell when I construct it, I will get hanging/segfault/hardware instruction errors.

[staticfloat]

Just for fun, I was trying this out, and I can't seem to get it to segfault at all on OSX or Ubuntu. Could this have fixed itself with all the signal handling improvements made recently? I also noticed that while Ubuntu barrages me with output:

julia> f(1000000)
error: stack overflow
 in f at none:1
...
 in f at none:1

julia>

OS X is comparatively stingy with its output:

julia> f(1000000)
error: stack overflow

julia>

Is this because OSX doesn't have libunwind?

[vtjnash]

@JeffBezanson is there actually some goal of providing a patch for this, or can it just be closed as wontfix? (I can't get this to segfault at all on Mac64 or Ubuntu32, even with dozens of repeat executions)

as a sidenote, OSX now has libunwind, and prints the full, long backtrace

[vtjnash]

this potentially had been improved by 72793ed. see also #1485