Crash in pretty simple program involving GOSUB and RETURN

[soutbarr]

This is for the current Basic compiler (I was a part of the community program before it went vmdk), also this is on x86.
Adding version info
VSI BASIC x86-64 X1.8-005 (GEM 50Y3Q) on OpenVMS x86_64 V9.2-2


Alright, when the a program has a single subroutine it works perfectly fine. The moment it has more than one subroutine it breaks and crashes when it encounters it's first return call.

Code: Select all

OPTION TYPE = EXPLICIT
  GOSUB PRINT_HIYA
  PRINT "2nd GOSUB Upnext"
  GOSUB PRINT_HI
EXIT PROGRAM

PRINT_HIYA:
  PRINT "Hiya"
RETURN

PRINT_HI:
  PRINT "Hi"
RETURN

END

That is the Basic program, it compiles and links just fine on execution

Code: Select all

$ r testreturn
Hiya
%SYSTEM-F-ACCVIO, access violation, reason mask=15, virtual address=FFFFFFFF800000B0, PC=00002098010E0004, PS=0000001B

  Improperly handled condition, image exit forced by last chance handler.
    Signal arguments:   Number = 0000000000000005
                        Name   = 000000000000000C
                                 0000000000000015
                                 FFFFFFFF800000B0
                                 00002098010E0004
                                 000000000000001B
    Register dump:
    RAX = FFFFFFFF800000B0  RDI = 000000007ACB38D8  RSI = 0000000000000000
    RDX = 000000007ACB38A8  RCX = 0000000000000000  R8  = 000000007C05D790
    R9  = 0000000000000000  RBX = 000000007ACB3890  RBP = 000000007ACB3950
    R10 = FFFFFFFFFFFFFFFE  R11 = 0000000000000001  R12 = 000000007ACB39A8
    R13 = 0000000000000000  R14 = 000000007ACB3890  R15 = 0000000030F84606
    RIP = 00002098010E0004  RSP = 000000007ACB3880  SS  = 000000000000001B

[hb]

Tell the linker to put the code into P0. The FFFFFFFF800000B0 indicates a sign extension problem. To me 800000B0 looks like a P2 jump target. If you have the code in P0 this target is at 000020B0 and then the program works - for me.

I am inclined to suggest that for the time being all basic programs should be linked with code in P0.

Code: Select all

$ bas testreturn
$ link testreturn
$ r testreturn
Hiya
%SYSTEM-F-ACCVIO, access violation, reason mask=15, virtual address=FFFFFFFF800000B0, PC=00002098010E0004, PS=0000001B

  Improperly handled condition, image exit forced by last chance handler.
    Signal arguments:   Number = 0000000000000005
                        Name   = 000000000000000C
                                 0000000000000015
                                 FFFFFFFF800000B0
                                 00002098010E0004
                                 000000000000001B
    Register dump:
    RAX = FFFFFFFF800000B0  RDI = 000000007AC718D8  RSI = 0000000000000000
    RDX = 000000007AC718A8  RCX = 0000000000000000  R8  = 000000007C03D790
    R9  = 0000000000000000  RBX = 000000007AC71890  RBP = 000000007AC71950
    R10 = FFFFFFFFFFFFFFFE  R11 = 0000000000000001  R12 = 000000007AC719A8
    R13 = 0000000000000000  R14 = 000000007AC71890  R15 = 000000001F104606
    RIP = 00002098010E0004  RSP = 000000007AC71880  SS  = 0002868800000004
$ link testreturn/segment=code=p0
$ r testreturn
Hiya
2nd GOSUB Upnext
Hi
$ 

[cct]

Not even sure that is enough . I have given up testing my old suite. It all compiles, linked on P0 but there are so many run time failures - especially any error handling, unless trivial.
Also the debugger is currently toast.

Hoping my ambassador membership comes through soon - I quickly got an email saying I was accepted, but without the OS update, new version of Basic, and the the promised follow up, I cannot do any more

[neilrieck]

As HB already pointed out, you need to link differently (for now)

Code: Select all

kawc77:neil> type crash.bas
OPTION TYPE = EXPLICIT
  GOSUB PRINT_HIYA
  PRINT "2nd GOSUB Upnext"
  GOSUB PRINT_HI
EXIT PROGRAM

PRINT_HIYA:
  PRINT "Hiya"
RETURN

PRINT_HI:
  PRINT "Hi"
RETURN

END
kawc77:neil> bas crash    
kawc77:neil> link/segment=code=p0 crash
kawc77:neil> run crash
Hiya
2nd GOSUB Upnext
Hi
kawc77:neil> 

Remember that this is a limited field trial. That said, I have sent them approx a half dozen problems (one crash; some where output on x86 doesn't exactly match Itanium) and they have been very prompt at fixing them. IMHO, they ported BASIC last because they knew it would be messy.

[soutbarr]

Oh I know, I just don't know where to submit those bug reports.

[snadow]

Oh I know, I just don't know where to submit those bug reports.

As a "community" user (vs. a person holding a paid-for license with a paid-for support contract), you're in the right place already - here in the forum. VSI will open a case on your behalf if they view your bug as valid and not already submitted by someone else.

[jreagan]

We haven't updated the BASIC FT compiler yet. I think that linking with /SEG=CODE=P0 should be sufficient.

We are working on the compiler & RTL (they both share knowledge of run-time data structures) fix for this.