Problems with pc:relative

[George]

Ben and Minoru and others

I have some code that ran nicely in the old version 9 but doesn't run now on 10.2

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

;; This in v9 would play a 60 note and then random walk either side of that note.

;; Now in v10.2 the note just climbs higher and higher - never comes back.

(*metro* 'set-tempo 80)

;;

(define node1

(lambda (beat pitch dlst)

(println pitch)

(play piano1 pitch (cosr 80 20 8) 2.2 0)

(callback (*metro* (+ beat (* .9 (car dlst))))

'node1

(+ beat (car dlst))

(pc:relative pitch (random '( -1 1 0 )) scale) ;; the -1 results in a 0 !!

(if (null? (cdr dlst))

(make-rhythm 4 2 .9 '(1/3 1/3 1/2 1/2 1/2))

(cdr dlst)))))

(node1 (*metro* 'get-beat 4) 60 '(1/2 1/3 1/3 1/2 1/3 1) )

; Random walk 60, 60, 62, 60, 60, 60, 62, 64, 64, 62, 62, 60, 62, 64, 62, 62, 60, 59, on old version 9

;; But on version 10.2 the pitch just keeps climbing.

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

;; Here are some tests

(println scale) ;; (0 2 4 5 6 8 10 11)

(println (pc:relative 68 (random '(-1 1 0)) scale)) ;; 66 68 70 on v9.0 .... but 68 and 70 only on v10.2

(println (random '( -1 1 0 ))) ;; 0 1 -1 on both v9.0 and v10.2

(pc:relative 64 -1 '(0 2 4 5 7 9 11)) ;; 63 on v9.0 .... but 64 on v10.2

(pc:relative 64 -2 '(0 2 4 5 7 9 11)) ;; 60 on v9.0 .... but 64 on v10.2

(pc:relative 64 1 '(0 2 4 5 7 9 11)) ;; 65 on both v9.0 and v10.2

(pc:relative 64 0 '(0 2 4 5 7 9 11)) ;; 64 on both v9.0 and v10.2

;; I note that pc:relative makes use of real->integer in its define. As do most pc:* defs.

;; https://github.com/digego/extempore/blob/2af178b58a6d003b2de3143f7c26457d1f0d3a98/libs/core/pc_ivl.xtm#L347

(real->integer 4.7) ;; 4 in both v9.0 and v10.2

(real->integer -4.7) ;; -4 in v9.0 but 0 in v10.2

;; Maybe real->integer is the source of problems for other pc:* I haven't yet tested.

;;

Regards

George

[Minoru]

Hi George,

Thanks for your report !
I got the return value like below .... I have 4 extempore versions ....

extempore version          ---->              .7    .89  .94  .10.2
(pc:relative 64 -1 '(0 2 4 5 7 9 11))  ;; 62,  62,  64,   64
(pc:relative 64 -2 '(0 2 4 5 7 9 11))  ;; 60,  60,  64,   64
(pc:relative 64 1 '(0 2 4 5 7 9 11))   ;; 65,  65,  65,   65
(pc:relative 64 0 '(0 2 4 5 7 9 11))   ;; 64,  64,  64,   64

I think .7 and .89 is correct .... anyway, your 63 is typing mistake ?

2026年7月16日木曜日 13:27:29 UTC+9 George:

[Ben Swift]

Hi George, Minoru,

Good sleuthing — real->integer was a bit dodgy. The C code behind it converts the double to an unsigned 64-bit integer, which is undefined behaviour for negative numbers: Intel machines happen to wrap around to the correct negative answer, but Apple Silicon clamps to 0. With the step truncated to 0, pc:relative returns the input pitch unchanged, which is why George's random walk could climb but never come back down.

That also explains the version table: the buggy cast has been in the codebase since 0.8.3, but 0.7/0.8.9 predate native Apple Silicon builds (they ran as Intel code under Rosetta), so it only surfaced from 0.9.x. And yes — 62 is the expected answer for (pc:relative 64 -1 ...), so George's 63 looks like a typo, as Minoru guessed.

The fix (plus regression tests, including real->rational which had the same problem) is on master, and I've cut a v0.10.3 patch release: binaries should be up at https://github.com/digego/extempore/releases shortly.

Cheers,
Ben

>--
>You received this message because you are subscribed to the Google Groups "Extempore" group.
>To unsubscribe from this group and stop receiving emails from it, send an email to extemporelan...@googlegroups.com.
>To view this discussion visit https://groups.google.com/d/msgid/extemporelang/c3ffbfc8-2261-46d7-bd40-61e59a61dd2cn%40googlegroups.com.


--

Cheers,
Ben

[George]

Thanks Minoru & Ben

Actually I don't know which version I have on the older Intel computer. It doesn't have then"./extempore --version" command that later ones have.

It was loaded first in May 2012. OK on the typo - checked again.

Ben I guess that you are confident that the other pc:* things are fixed too?

Regards

George

[Ben Swift]

Reasonably confident that this will fix the issue you described (the new release that is).

Ben

>To view this discussion visit https://groups.google.com/d/msgid/extemporelang/60bd8321-bb1a-450a-b88b-50347e9829e1n%40googlegroups.com.


--

Cheers,
Ben

[Minoru]

Hi Ben and George,

Thanks much Ben, I got it, CPU differences of Intel and Apple Silicon, and Rosetta's working, I forgot it at all though I used it, too !

2026年7月16日木曜日 16:49:49 UTC+9 ben: