Different error message vanilla vs CS (again)

[sorawee]

What version of Racket are you using?
7.8.0.6--2020-07-31 [cs]

What program did you run?

#lang racket

(define x y)
(define y 1)

What should have happened?

y: undefined;
 cannot reference an identifier before its definition

in DrRacket, and

y: undefined;
 cannot reference an identifier before its definition
  in module: "/path/to/test.rkt"

in Racket.

If you got an error message, please include it here.

../../../racket/racket/share/pkgs/drracket/drracket/private/stack-checkpoint.rkt:115:30: define-values: skipped variable definition;
 cannot continue without defining variable
  variable: x
  in module: "/path/to/test.rkt"

in DrRacket when "Enforce constant definitions (enables some inlining)" is off.

../../../racket/racket/share/pkgs/drracket/drracket/private/stack-checkpoint.rkt:115:30: y: undefined;
 cannot reference an identifier before its definition

in DrRacket when "Enforce constant definitions (enables some inlining)" is on.

y: undefined;
 cannot reference an identifier before its definition
  module: "/path/to/test.rkt"

in Racket.

Please include any other relevant details

1. Notice that git/racket/racket/share/pkgs/drracket/drracket/private/stack-checkpoint.rkt:115:30 doesn't exist in DrRacket BC.
2. In DrRacket BC, whether "Enforce constant definitions (enables some inlining)" is on or off, the error is the same. In DrRacket CS, when "Enforce constant definitions (enables some inlining)" is off, the error is different, blaming the definition of x instead.
3. In Racket CS, the error message uses "module: ..." whereas in Racket BC, the error message uses "in module:"

[sorawee]

Oh, interesting. For "../../racket/racket/share/pkgs/drracket/drracket/private/stack-checkpoint.rkt:115:30" to appear, "No debugging or profiling" must also be chosen.

[sorawee]

Another problem:

In Racket BC:

> (values (values 1 2))
; result arity mismatch;
;  expected number of values not received
;   expected: 1
;   received: 2

In Racket CS

> (values (values 1 2))
; returned two values to single value return context (no newline)

[sorawee]

Similarly:

In Racket BC:

> (define-values (x y) 1)
; define-values: result arity mismatch;
;  expected number of values not received
;   expected: 2
;   received: 1
;   in: definition of x ...

In Racket CS:

> (define-values (x y) 1)
; define-values: incorrect number of values from rhs 1 (no newline)

[sorawee]

Another problem (sorry for spamming...):

Racket BC:

> (let-values ([(a b) (values 1 2 3)]) 1)
; result arity mismatch;
;  expected number of values not received
;   expected: 2
;   received: 3
;   in: local-binding form

Racket CS:

> (let-values ([(a b) (values 1 2 3)]) 1)
; result arity mismatch;
;  expected number of values not received
;   expected: 2
;   received: 3
;   at: local-binding form

Notice in vs at.

[gus-massa]

The last one looks easy to solve.

racket/racket/src/cs/rumble/error.ss

Line 431 in 454a586

"\n at: local-binding form"

Perhaps you want to write the PR? (Assumin @mflatt doesn't disagree.)

[capfredf]

Similarly:

In Racket BC:

> (define-values (x y) 1)
; define-values: result arity mismatch;
;  expected number of values not received
;   expected: 2
;   received: 1
;   in: definition of x ...

In Racket CS:

> (define-values (x y) 1)
; define-values: incorrect number of values from rhs 1 (no newline)

This difference doesn't appear if we put the definition in a program and run it.

I am guessing this is because at REPL, raise-definition-result-arity-error was not called. I have a plan to fix this issue but I have some trouble figuring out where raise-definition-result-arity-error should be used. ref

@mflatt, can you give me some pointers on this issue?

[sorawee]

Hmm. It's not only that the error message is wrong. Programs that should error might not error at all!

#lang racket

y
(define y 1)

produces a value 'undefined5 when "Enforce constant definitions (enables some inlining)" is off.

We can even use this value in computations:

#lang racket

(eq? y y)
(define y 1)

produces #t.

[sorawee]

Just got hit by this again.

@mflatt I saw you changed several tests to accept error messages like "returned two values to single value return context" in addition to the one raised by raise-result-arity-error. Do I understand correctly that the difference is considered acceptable, and the bug is a WONTFIX?

[mflatt]

No, I just haven't gotten back to this.

[sorawee]

Here's another problem (which might be related to the above reported problem).

Consider an example in the doc of prop:exn:srclocs.

In DrRacket BC, DrRacket can highlight voldemort regardless of the setting of errortrace. The error message is:

oh dear don't say his name

In DrRacket CS, DrRacket can highlight voldemort only when errortrace is enabled, with the same error message mentioned above. However, when errortrace is disabled, voldemort is not highlighted, and the error message is:

 unsaved editor:42:17: oh dear don't say his name

[sorawee]

The remaining issues are also reported elsewhere (see above references), so let me close this.

[sorawee]

Let me also reopen this issue. There is one remaining issue left, which I will restate here:

#lang racket

(/ 1 0)

results in:

/: division by zero

in DrRacketBC, but results in:

projects/racket/extra-pkgs/drracket/drracket/drracket/private/stack-checkpoint.rkt:115:30: /: division by zero

in DrRacketCS.

Note that errortrace must be set to off.

I previously thought this bug would be fixed by the fix of racket/drracket#455, but they seem to be unrelated after all.

[sorawee]

The remaining issues are a combination of several bugs in DrRacket.

• The voldemort problem is specific to only unsaved editor and background expansion, and it's fixed by racket/drracket@7c30a88 and racket/gui@04e8164. (Racket BC on my machine appears to run without places, so the background expansion kicks in very late, causing the bug to occur more rarely).
• The (/ 1 0) problem is due to incorrect stack cutting, which is fixed in stack-checkpoint: evaluate ccm at the correct context drracket#459. (The problem occurs in Racket BC too, but it's "hidden" because continuation-mark-set->context tends to give more information in Racket BC, which happens to hide the problem).

These are all fixed, so I will close the issue.