local-expanding define-syntax in drracket seg faults with new expander

[Stephen Chang]

The following code produces a seg fault with the new expander (today's
build) in the scenario described below:

file1.rkt:

#lang racket
(require (for-syntax syntax/parse))

(define-syntax (define-constructor stx)
(syntax-parse stx
[(_ X:id)
#'(define-syntax X
(syntax-parser
[(_ x (... ...)) #'(void x (... ...))]))]))

(define-constructor C)

(define-syntax define-alias
(syntax-parser
[(_ alias:id e)
#:with e+ (local-expand #'e 'expression null)
#'(define-syntax alias
(syntax-parser [x:id #'e+]))]))

(define-alias D (C 1))

file2.rkt:
(require "file1.rkt")

A few notes:
- running file2.rkt in DrRacket with the new expander crashes drracket
with the error:

SIGSEGV MAPERR si_code 1 fault on addr (nil)
Aborted

- running file2.rkt with racket (new expander) produces no error
- running file2.rkt produced no error with the old expander
- running file1.rkt with either racket or drracket, new expander or
old expander, produces no error

I'm not completely caught up with the new expander threads so
apologies in advance if someone has already raised this issue.

[Matthew Flatt]

I've just pushed a repair for this bug. Thanks for the report!

> --
> You received this message because you are subscribed to the Google Groups "Racket Developers" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to racket-dev+...@googlegroups.com.
> To post to this group, send email to racke...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/racket-dev/CAFfiA1LDBwHox-1iP9UdDQOKSLMU-TbhUVtbsd-TLrNNO0eB_g%40mail.gmail.com.
> For more options, visit https://groups.google.com/d/optout.
>

[Stephen Chang]

The previous example now works. Thanks.

But here is a variation that still seg faults:

file1.rkt:

#lang racket
(require (for-syntax syntax/parse))

(provide (rename-out [my-define define]))
(define-syntax my-define
(syntax-parser
[(_ x:id e)
;; undef is unbound identifier
#:with y (local-expand #'undef 'expression null)
#'(void)]))

; (my-define X 1) produces: "undef: unbound identifier in module in: undef"

file2.rkt

#lang racket
(require "file1.rkt")
(define X 1)

Running file2.rkt in drracket with the new expander crashes drracket with:

SIGSEGV MAPERR si_code 1 fault on addr 0x20
Aborted

[Matthew Flatt]

Repair pushed. The problem was related to the bytecode format, so I
had to bump the version number.

Thanks!

[Stephen Chang]

> Repair pushed. The problem was related to the bytecode format, so I
> had to bump the version number.

Ah ok. This makes sense because I occasionally could not reproduce the
crash unless I forced a recompile. Sorry, I should have mentioned this
initially. Thanks.

[Stephen Chang]

Here is another seg faulting example.

example3a.rkt:

#lang racket
(require (for-syntax syntax/parse))

(provide test)
(define-syntax test
(syntax-parser
[_
#:with lam+
(local-expand
#'(λ (z) (let-syntax ([z (make-rename-transformer #'z)]) z))
'expression null)
#'(define-syntax x #'lam+)]))
(test)

example3b.rkt:

#lang racket
(require "example3a.rkt")

To reproduce:
1) Open example3b.rkt in drracket and run. It may or may not crash
with a seg fault.
2) Delete compiled directory and run again.

Observations:
- A crash typically occurs after repeating the steps 2-3 times but
sometimes I have to run-and-compile 5 or more times.
- The provide appears to be necessary to cause the crash, even though
it's unused in the second file.
- I'm not sure which parts of the macro are actually needed to create
the crash but the local-expand and define-syntax seem to be involved.

[Matthew Flatt]

Thanks! Repair pushed.

[Stephen Chang]

Thanks! Is there a quick explanation for why it was only crashing
intermittently?

> To view this discussion on the web visit https://groups.google.com/d/msgid/racket-dev/20150723145046.4AD8D6501DD%40mail-svr1.cs.utah.edu.

[Matthew Flatt]

I'm not sure, but he problem was related to a kind of GC on binding information when marshaling. It probably crashed only for some traversal orders, and maybe something like `eq?` hashing affected the order.

> To view this discussion on the web visit https://groups.google.com/d/msgid/racket-dev/CAFfiA1%2BR5qgeLU1ibArgoJLe8GQysm6MgY1-%3DtEVA9Mrfb2prQ%40mail.gmail.com.