RiscLua 6.0

It works OK for me with StrongED 4.70a9 (01 Aug 2015).

Didn’t know there was a new version out. Will try it and see what happens.

Have tried it. Still the same—causes !StrongEd to crash when backspace key is pressed.

The ^H^H phenomenon occurs with !DeskEdit, !Edit and !Zap as well. Just in case it made a difference, I’ve tried !Nettle as well but the samw happens there too.

I’ve restored the previous version of RiscLua and in that, the behaviour on input is correct.

Looks like this is a bug in the version of !GCC you’ve used.

What machine and what version of RISC OS are you using? RiscLua 6 is compiled with GCC 4.7.4 rel2 , not using the shared C library. When you type in text at the > prompt it is fgets that reads in the string from stdin and puts that echoes it back. The problem looks like it lies with one or other of these. What happens if you make a text file containing just

print "hello"

Mmm. Needs work. :-)

Downloaded from !Store.
SharedLibs put in Resources, SharedULib put in !System.
Lua extracted to $.risclua.

Lua starts, places an icon on the iconbar.
Nothing works – “File 'stick' not found”.

It seems that the Obey files are like:

wimpslot -32K
stick rlua6:Tools.<something>

Either prefix “stick” with “rlua6:” or add RiscLua’s path to Run$Path; as the method you’re using is only going to work if RiscLua’s directory is the CSD.

Tweaked TaskW to run. It says to drag a file to the icon bar icon to run it in a TaskWindow. Well, uh, all of the examples cause “Window Manager is currently in use”.

So I decide to open a TaskWindow and try Lua directly.
“SDFS::blahblahblah.lua6: can't load library 'libgcc_s.so'”.

Perhaps you might like to probe for these things on first run, you shouldn’t assume that it is all present as standard (it wasn’t on my machine).

I created “abi-2/0” and copied in libgcc* and libunix* that I downloaded a couple of weeks ago for trying Otter.

Tried “lua” in a TaskWindow again. Crash. Oops.
In case it was running out of memory with the 6MiB of libraries (gee, just like VB! ;-) ) I dragged Next up to 50-something MiB and tried again.
[it also does this if I try running a Lua file containing print "hello"…]

Fatal signal received: Illegal Instruction

Stack backtrace:

Running thread 0x40ac4
  ( 3587f34) pc: 5ace6014 lr: 5ace6428 sp:  3587f38  __write_backtrace()
  ( 3587fa0) pc: 5ace615c lr: 5ace7fe0 sp:  3587fa4  __unixlib_raise_signal()
  ( 3587fb0) pc: 5ace7e94 lr:     b958 sp:  3586e1c  __h_cback()

  Register dump at 03587fb4:

    a1:    68dfc a2:        0 a3: 407f7000 a4:       88
    v1:    68dfc v2:    68d80 v3:        1 v4:    6918c
    v5:        1 v6:        c sl:  3586208 fp:  3586e78
    ip:  3586e38 sp:  3586e1c lr:     b958 pc:    20980
    cpsr: 60000110

  0002096c : n_.. : 00005f6e : ANDEQ   R5,R0,R14,ROR #30
  00020970 : ...ÿ : ff000014 : Undefined instruction
  00020974 : .À á : e1a0c00d : MOV     R12,R13
  00020978 : pØ-é : e92dd870 : STMDB   R13!,{R4-R6,R11,R12,R14,PC}
  0002097c : .⇑-í : ed2d8b02 : VPUSH   {D8}               ; VFP or ASIMD required
  00020980 : .°Lâ : e24cb004 : SUB     R11,R12,#4
  00020984 : ..]á : e15d000a : CMP     R13,R10
  00020988 : ʧÿ» : bbffa7ca : BLLT    &0000A8B8
  0002098c : .ÐMâ : e24dd008 : SUB     R13,R13,#8

  ( 3586e78) pc:     b8b4 lr:    10728 sp:  3586e7c  pmain()
  ( 3586ea8) pc:    104cc lr:    10a70 sp:  3586eac  luaD_precall()
  ( 3586ec0) pc:    10a20 lr:     c048 sp:  3586ec4  luaD_call()
  ( 3586ed0) pc:     c034 lr:     fe0c sp:  3586ed4  f_call()
  ( 3586f6c) pc:     fdb4 lr:    10d6c sp:  3586f70  luaD_rawrunprotected()
  ( 3586f98) pc:    10d34 lr:     dde4 sp:  3586f9c  luaD_pcall()
  ( 3586fc8) pc:     dd60 lr:     ad9c sp:  3586fcc  lua_pcallk()
  ( 3586fec) pc:     ad34 lr: 5acfcbe4 sp:  3586ff0  main()

*

I’ll say something for GCC – it’s a nice looking backtrace.

“Oh! Suits you sir”

“Oh! Suits you sir”

Irrelevant to RISC OS, but that should, IIRC, be “Suit you, Sir!”

Apparently often misquoted.

Have tried it. Still the same—causes !StrongEd to crash when backspace key is pressed.

It’s fine here, running the latest StrongED.

The ^H^H phenomenon occurs with !DeskEdit, !Edit and !Zap as well.

RiscLua 5.71 does not exhibit this behaviour so it’s likely down to the switch to GCC. One thing to check is how characters read are typed, as char or int?

Downloaded from !Store.

It looks like SharedLibs hasn’t been updated for GCC 4.7.2. I upgraded my copy using PackMan, it has additional directories; abi-1/0 and abi-2/0.

“Oh! Suits you sir”

(and John’s correction)

Sorry, I never found The Fast Show to be either funny or worth watching, so while I recognise the quote, I don’t know its context…

needs more work :)

Apologies for that. I have been uploading corrections to my website and plingstore fairly frequently – not because RiscLua itself has changed but the libraries and examples have. All information about how it fails to run on other setups than my own are very useful. Maybe it would make sense to have separate downloads for examples and for libraries.

"File 'stick' not found".

My copy has rlua6:stick <rlua$dir> in the !Run file. Did I really put out a copy without the rlua6: prefix? If so, apologies. I should have tested more carefully what bits of !SharedLibs are needed and what are not. I have all sorts of junk in there.

At the moment the only examples which are wimp tasks (i.e. cannot be run in a taskwindow) are !Eph and !Snerp. !Eph does what Acorn’s !Msg did, as found in the C/C++ package’s !SetPaths. That is, it puts out a temporary message – the message and the delay time in seconds are RunImage’s command line parameters.

!Snerp is different from the application of that name in earlier distributions. It is a demonstration of how to use the libraries task, window, menu and ibar. It puts an icon on the iconbar, which has a menu – the usual info/help/quit. Clicking Select opens a window. Dragging an object into the window will report the object’s filetype. The code demonstrates how to make objects respond to clicks and drags and how to give menu-items actions. The terminology has been chosen to be reminiscent of some of the wimp library of Acorn C release 4 – but without the types :).

In some sense this has been an experiment in trying to separate wimp programs into two components – the detail hidden in libraries, the structure exposed in the !RunImage that uses the libraries. Whether the result is any more intelligible in RiscLua than in BASIC or C I am not sure. For example, what happens in the main window is given by

snerp:register "main" {
     on_Menu = menu.open (mainmenu);
     on_Select = say "Select clicked in Main";
     on_Adjust = say "Adjust clicked in Main";
     on_Drag = \ (self, _, ftype)
                 self:alert (dr:format (ftype))
               end; 

local imenu = {
    title = "Snerp";
    { name = "info"; submenu = snerp.object.info; };
    { name = "help";};
    { name = "quit"; };
    click = {
        quit = QUIT;
        help = \ (self)
          os.execute (filer_run:format (help_file))
          end;
            }; -- click
   } -- imenu

In response to Gavin above, I was using VRPC with RO 4.39. If I used

print("hello")

it works as it should—but if there is an input statement in the Lua code, it also exhibits the ^H problem. I notice that a program compiled using Norcroft C and using gets() works correctly. I think next I’ll check how GCC behaves on my Raspberry Pi.

I was using VRPC with RO 4.39.

Did you remember to drag the contents of the v3 directory over !rlua6? In the zipped distribution the lua6 and luac6 files in !rlua6 have been compiled for ARMv7 platforms with vfp. The lua6 and luac6 files in v3.!rlua6 have been compiled for ARMv3 platforms with softfloat.

I didi drag the V3 directory over !rlua6 (actually, I forgot to do so at first, and got a message saying something like “this version needs ARM v6”).

Gavin:

Further to the original (^H^H^H) problem, you could use the OS routines to do the input—in this case OS_Read_Line32. This works correctly with GCC and allows the usual RISC OS line editing facilities to work.

It can be called either using _swi() or _swix() from C or alternatively using OSLib.

If you’re using fgets() because you sometimes need to read from a file you could have a routine which only calls the SWI if the ‘file’ is standard input.

OS_Read_Line32

Thanks. Yes I tried this, but I must have made a pig’s ear of it, because it caused a crash, so I left well alone. Also, I wanted to keep the lua binary as close as possible to the standard distribution, and restrict everything to do with RISC OS to the riscos library, if possible. Obviously, allowing the use of !,?,$ in identifiers (a nod to BASIC) could not very well be so confined, so had to be done with a #ifdef RISCOS … #endif patch in llex.c.

I’ve tried this just now, actually using OS_Read_Line (on RO 4.39) on the ARM v3 version. It works fine, the only difference is that typing CTRL-d at the prompt doesn’t terminate lua6, instead you have to press Escape instead.

The code changes (to “lua.c” i.e. c.lua) are minor. I’ve added an include:

#include <swis.h>

changed a #define:

#define lua_readline(L,b,p) \
        ((void)L, fputs(p, stdout), fflush(stdout),  /* show prompt */ \
        my_gets(b, LUA_MAXINPUT) != NULL)       /* get line */

and added the following code:

/* Read line -- SWI &0E (14) */
void ro_read_line(char* buf, int size, int lo, int hi, int* count) {
  _swi(14,_IN(0)|_IN(1)|_IN(2)|_IN(3)|_OUT(1),buf,size,lo,hi,count);
}

/* Routine to emulate fgets(str,n,stdin) but using RO routines */
char* my_gets(char* str, int n) {
  int count;  /* number of characters read */

  ro_read_line(str,n,0,255,&count);
  str[count]=0;
  return str;
}

Just to add with respect to the problems I encountered modifying the TaskWindow mode file for StrongEd, using CharstoBuffer("<127>"), as stated earlier I get “abort on data transfer” — the address is &03807B64. It doesn’t make any difference if I use “8” instead of “127”.

Note: This change can be made successfully on RO 5.21/Pi version 1. The problem encountered here is on RO 4.39/VRPC.

I’ve modified the code for the revised version of “lua.c” to call the X version of the SWI:

/* Read line -- SWI &0E (14) */
_kernel_oserror* ro_read_line(char* buf, int size, int lo, int hi,
                              int* count) {
  return _swix(14,_IN(0)|_IN(1)|_IN(2)|_IN(3)|_OUT(1),buf,size,lo,hi,count);
}

/* Routine to emulate fgets(str,n,stdin) but using RO routines */
char* my_gets(char* str, int n) {
  int count;                 /* number of characters read */
  _kernel_oserror* err;      /* error return -- NULL if OK */

  err = ro_read_line(str,n,0,255,&count);
  if (err) {
    fprintf(stderr, "Error %d: %s\n", err->errnum, err->errmess);
    return NULL; /* fgets() returns NULL if an error occurs */
  } else {
    str[count]=0;
    return str;
  }
}

Everything else stays the same as in the previous post (i.e. the altered #define, and the additional #include).

I’ve found while these changes work OK in running Lua interactively, they make no difference if you write a program in Lua—you still get ^H if you backspace.

I can’t find any other instances of fgets (or fscanf, for that matter) in the code—am I missing something?

That is nothing to do with Lua, surely, but the text editor you use to write the program. I tend to use StrongED for creating text (currently version 4.70a9) – whether for programs or for anything else – and I have no such problem. What version are you using?

A misunderstanding, I think.

What I meant was, if I write a program in Lua that uses io:input() to read in a string! it displays the ^H phenomenon still even though the modified version of Lua doesn’t do so when used interactively.

Only if you run it in a command window. If you run it in a taskwindow there should be no problem.

I was wondering where in the Lua source is the code that is used by io:input? I can’t use a TaskWindow successfully since StrongEd crashes if I modify the mode file as suggested (see post above).

liolib.c is what you want, I think.

If you want to know more about RiscLua, Gavin will be telling us all about it at the next ROUGOL meeting on Monday 16th November:

http://www.rougol.jellybaby.net/

Entance is free so come along and ask lots of questions!