SIGSEGV on closure / stack heavy program

Version: v7.4.0
Platform: Linux remba2 4.4.0-59-generic Adding more specific README URLs to io.js, keeping node in too #80-Ubuntu SMP Fri Jan 6 17:47:47 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux
Subsystem:

$ node --stack-size=100000 ./async-fizzbuzz.js
1
2
Fizz
4
Buzz
[…]
Fizz
88
89
FizzBuzz
fish: “node --stack-size=100000 ./asyn…” terminated by signal SIGSEGV (Address boundary error)

This reproduces for me with d8 on V8’s master for me. Could you open an issue at https://bugs.chromium.org/p/v8/issues and link that here?

v8 bugreport.

FWIW it might be better to include the code from the gist inline (but collapsed) in case the gist goes away.

@nodejs/v8 Looks like the v8 issue was closed, but this seems like a valid bug report to me… can one of you take a look?

Works for me with master on OS X, with or without --stack-size=100000.

@bnoordhuis

Same as @bnoordhuis for me, on Fedora 25.

FWIW it might be better to include the code from the gist inline (but collapsed) in case the gist goes away.

EDIT: This is the fixed version, see below.

async-fizzbuzz.js

// Async FizzBuzz -- implemented using only callbacks
//
// node --stack-size=100000 ./async-fizzbuzz.js
//

// Wrap process.stdout.write
const print = (message, callback) =>{
	process.stdout.write(message, callback)
}

// Scott numeral constructors
const Z = (callback)=>{callback((a,b)=>{a()})}
const S = (n, callback)=>{callback((a,b)=>{b(n)})}

// Numeral constants
const n0 = (a,b)=>{a()}
const n1 = (a,b)=>{b(n0)}
const n2 = (a,b)=>{b(n1)}
const n3 = (a,b)=>{b(n2)}
const n4 = (a,b)=>{b(n3)}
const n5 = (a,b)=>{b(n4)}
const n6 = (a,b)=>{b(n5)}
const n7 = (a,b)=>{b(n6)}
const n8 = (a,b)=>{b(n7)}
const n9 = (a,b)=>{b(n8)}
const n10 = (a,b)=>{b(n9)}

const if_zero = (n, then, else_)=>{n(
	()=>{then()},
	(r)=>{else_()}
)}

const if_equal = (m, n, then, else_)=>{m(
	()=>{if_zero(n, then, else_)},
	(mr)=>{n(
		else_,
		(nr)=>{if_equal(mr, nr, then, else_)}
	)}
)}

const add = (m, n, callback)=>{m(
	()=>{callback(n)},
	(mr)=>{S(n, (sn)=>{add(mr, sn, callback)})}
)}

const sub = (m, n, callback, error)=>{n(
	()=>{callback(m)},
	(nr)=>{m(
		error,
		(mr)=>{sub(mr, nr, callback, error)}
	)}
)}

const mul = (m, n, callback)=>{m(
	()=>{Z(callback)},
	(mr)=>{mul(mr, n, (r)=>{add(r, n, callback)})}
)}

const divmod = (m, n, callback)=>{
	sub(m, n,
		(mr)=>{divmod(mr, n, (q, r)=>{
			S(q, (sq)=>{callback(sq, r)})
		})}, 
		()=>{callback(n0, m)})
}

const divides = (m, n, then, else_)=>{
	divmod(m, n, (q, r)=>{if_equal(r, n0, then, else_)})
}

// Decimal output

const print_decimal = (n, callback, error)=>{
	if_equal(n, n0, ()=>{print("0", callback)}, ()=>{
	if_equal(n, n1, ()=>{print("1", callback)}, ()=>{
	if_equal(n, n2, ()=>{print("2", callback)}, ()=>{
	if_equal(n, n3, ()=>{print("3", callback)}, ()=>{
	if_equal(n, n4, ()=>{print("4", callback)}, ()=>{
	if_equal(n, n5, ()=>{print("5", callback)}, ()=>{
	if_equal(n, n6, ()=>{print("6", callback)}, ()=>{
	if_equal(n, n7, ()=>{print("7", callback)}, ()=>{
	if_equal(n, n8, ()=>{print("8", callback)}, ()=>{
	if_equal(n, n9, ()=>{print("9", callback)}, error
	)})})})})})})})})}) // LISP flashbacks
}

const print_number = (n, callback)=>{
	divmod(n, n10, (q, r)=>{
		if_zero(q, ()=>{
			print_decimal(r, callback)
		}, ()=>{
			print_number(q, ()=>{
				print_decimal(r, callback)
			})
		})
	})
}

// Loops

// for needs more callbacks.
// So far, we have only used anonymous functions. It would
// be a shame to stop that practice here. Therefore, I present
// you a fixed-point combinator in pure CPS:
const for_ = (start, end, func, callback)=>{
	((rec, n, end, func, callback)=>{
		rec(rec, n, end, func, callback)
	})((rec, n, end, func, callback)=>{
		func(n, ()=>{
			if_equal(n, end, callback, ()=>{
				S(n, (sn)=>{
					rec(rec, sn, end, func, callback)
				})
			})
		})
	}, start, end, func, callback)
}

// We are coming to the main show

const fizzbuzz_number = (n, callback)=>{mul(n3, n5, (n15)=>{
	divides(n, n15, ()=>{print("FizzBuzz\n", callback)}, ()=>{
	divides(n, n3, ()=>{print("Fizz\n", callback)}, ()=>{
	divides(n, n5, ()=>{print("Buzz\n", callback)}, ()=>{
	print_number(n, ()=>{print("\n", callback)})
	})})})
})}

const fizzbuzz = (callback)=>{mul(n10, n10, (n100)=>{
	for_(n1, n100, fizzbuzz_number, callback)
})}

fizzbuzz(()=>{})

@gibfahn

@gibfahn The gist was updated which causes the bug to no longer be triggered. The snippet you posted is the updated version and works for me. I suspect @targos and @bnoordhuis also ran this version. (It's not hard to make that one crash again, this is described in the Gist comments.) I apologize for the confusion.

Here's a hard link to the original breaking version and here's the code of the breaking version:

Edit: Sorry, I can't figure out the GitHub-Markdown syntax for collapsing.

Details

// Async FizzBuzz — implemented using *only* callbacks
//
// node --stack-size=100000 ./FizzBuzz.js
//

// We will need print function with a callback
var output = "";
const print = (message, callback)=>{
	output += message;
	callback();
}
const flush = (callback)=>{
	console.log(output);
	output = "";
	callback()
}

// Scott numeral constructors
const Z = (callback)=>{callback((a,b)=>{a()})}
const S = (n, callback)=>{callback((a,b)=>{b(n)})}

// Numeral constants
const n0 = (a,b)=>{a()}
const n1 = (a,b)=>{b(n0)}
const n2 = (a,b)=>{b(n1)}
const n3 = (a,b)=>{b(n2)}
const n4 = (a,b)=>{b(n3)}
const n5 = (a,b)=>{b(n4)}
const n6 = (a,b)=>{b(n5)}
const n7 = (a,b)=>{b(n6)}
const n8 = (a,b)=>{b(n7)}
const n9 = (a,b)=>{b(n8)}
const n10 = (a,b)=>{b(n9)}

const if_zero = (n, then, else_)=>{n(
	()=>{then()},
	(r)=>{else_()}
)}

const if_equal = (m, n, then, else_)=>{m(
	()=>{if_zero(n, then, else_)},
	(mr)=>{n(
		else_,
		(nr)=>{if_equal(mr, nr, then, else_)}
	)}
)}

const add = (m, n, callback)=>{m(
	()=>{callback(n)},
	(mr)=>{S(n, (sn)=>{add(mr, sn, callback)})}
)}

const sub = (m, n, callback, error)=>{n(
	()=>{callback(m)},
	(nr)=>{m(
		error,
		(mr)=>{sub(mr, nr, callback, error)}
	)}
)}

const mul = (m, n, callback)=>{m(
	()=>{Z(callback)},
	(mr)=>{mul(mr, n, (r)=>{add(r, n, callback)})}
)}

const divmod = (m, n, callback)=>{
	sub(m, n,
		(mr)=>{divmod(mr, n, (q, r)=>{
			S(q, (sq)=>{callback(sq, r)})
		})}, 
		()=>{callback(n0, m)})
}

const divides = (m, n, then, else_)=>{
	divmod(m, n, (q, r)=>{if_equal(r, n0, then, else_)})
}

// Decimal output

const print_decimal = (n, callback, error)=>{
	if_equal(n, n0, ()=>{print("0", callback)}, ()=>{
	if_equal(n, n1, ()=>{print("1", callback)}, ()=>{
	if_equal(n, n2, ()=>{print("2", callback)}, ()=>{
	if_equal(n, n3, ()=>{print("3", callback)}, ()=>{
	if_equal(n, n4, ()=>{print("4", callback)}, ()=>{
	if_equal(n, n5, ()=>{print("5", callback)}, ()=>{
	if_equal(n, n6, ()=>{print("6", callback)}, ()=>{
	if_equal(n, n7, ()=>{print("7", callback)}, ()=>{
	if_equal(n, n8, ()=>{print("8", callback)}, ()=>{
	if_equal(n, n9, ()=>{print("9", callback)}, error
	)})})})})})})})})}) // LISP flashbacks
}

const print_number = (n, callback)=>{
	divmod(n, n10, (q, r)=>{
		if_zero(q, ()=>{
			print_decimal(r, callback)
		}, ()=>{
			print_number(q, ()=>{
				print_decimal(r, callback)
			})
		})
	})
}

// Loops


// for needs more callbacks.
// We need recursion, so anonymous functions are not enough here
// TODO: Can this be done with anonymous functions?
// https://stackoverflow.com/questions/35918279/define-fix-point-combinator-in-continuation-passing-style#
const for_ = (start, end, func, callback)=>{
	(function rec(n) {
		func(n, ()=>{
			if_equal(n, end, callback, ()=>{
				S(n, (sn)=>{rec(sn)})
			})
		})
	})(start)
}

// We are coming to the main show

const fizzbuzz_number = (n, callback)=>{mul(n3, n5, (n15)=>{
	divides(n, n15, ()=>{print("FizzBuzz", callback)}, ()=>{
	divides(n, n3, ()=>{print("Fizz", callback)}, ()=>{
	divides(n, n5, ()=>{print("Buzz", callback)}, ()=>{
	print_number(n, callback)
	})})})
})}

const fizzbuzz_number_newline = (n, callback)=>{
	fizzbuzz_number(n, ()=>{flush(callback)})
}

const fizzbuzz = (callback)=>{mul(n10, n10, (n100)=>{
	for_(n1, n100, fizzbuzz_number_newline, callback)
})}

fizzbuzz(()=>{})

I can reproduce now but it's not a bug, it's a misunderstanding about what --stack-size=... does. It tells V8 to assume - not create - the stack is that big. The help message for the option mentions that.

Your test informs it the stack is approx. 100 MB big when in reality it's probably only around 8 MB. Set ulimit -s 10000 first and it should work. :-)

@bnoordhuis

@bnoordhuis Confirmed. ulimit -s 10000; node --stack-size=100000 ./async-fizzbuzz.js works!

The stack-size option is not in the man page or in the --help. I got it from this stackoverflow post where it is introduced as increasing the available stack.

After some digging I found that node --v8-options has --stack_size (note the underscore) and the cryptic/ambiguous "default size of stack region v8 is allowed to use", which can be interpreted as you say.

Perhaps node should use getrlimitor setrlimit for the stacksize and not allow foot-shooting this way?

It's not out the question but it opens a can of worms. Resizing the stack is complicated on Windows and needs some decisions in general: should node exit with an error when resizing fails or do we silently fix up --stack_size to a reasonable value? If so, what constitutes 'a reasonable value' and do we print a warning?

And so on - there is probably more that I haven't thought of. You are welcome to have a go at it, though.

Navigation Menu

Repository navigation

SIGSEGV on closure / stack heavy program #11091

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

SIGSEGV on closure / stack heavy program #11091

Description

Activity

addaleax commented on Jan 31, 2017

recmo commented on Jan 31, 2017

mscdex commented on Feb 1, 2017

addaleax commented on Feb 16, 2017

bnoordhuis commented on Feb 16, 2017

targos commented on Feb 18, 2017

gibfahn commented on Feb 19, 2017

recmo commented on Feb 19, 2017

bnoordhuis commented on Feb 20, 2017

recmo commented on Feb 23, 2017

bnoordhuis commented on Feb 28, 2017

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions