[9fans] acid question

[9fans] acid question

[Tim Newsham]

I have two variables, "a" and "b".  One comes from a symbol name
(ie: a = main), and the other comes from data returned from strace().
Normal stuff:

    print	shows
    -----------------------------
    a		0x800529a4
    b		0x800529a4
    whatis a	integer variable format X
    whatis b    integer variable format X
    a - b	0x00000000
    b - a	0x00000000

now weirdness:
    a == b	0
    a < b	0x00000001
    b < a	0x00000000
    a > b	0
    b > a	1

Whats going on here?  Signed vs. unsigned issues?

Tim N.

Re: [9fans] acid question

[Rob Pike]

Acid can be surprising.

acid; c=a>b
acid; whatis c
integer variable format D
acid; c=a<b
acid; whatis c
integer variable format X

This at least can be fixed by putting
	res->store.fmt = 'D';
in a number of places in src/cmd/acid/expr.c analogous
to its placement in ogt().

-rob


On Fri, 7 Jan 2005 17:36:17 -1000 (HST), Tim Newsham <newsham@lava.net> wrote:
> I have two variables, "a" and "b".  One comes from a symbol name
> (ie: a = main), and the other comes from data returned from strace().
> Normal stuff:
> 
>     print       shows
>     -----------------------------
>     a           0x800529a4
>     b           0x800529a4
>     whatis a    integer variable format X
>     whatis b    integer variable format X
>     a - b       0x00000000
>     b - a       0x00000000
> 
> now weirdness:
>     a == b      0
>     a < b       0x00000001
>     b < a       0x00000000
>     a > b       0
>     b > a       1
> 
> Whats going on here?  Signed vs. unsigned issues?
> 
> Tim N.
>

Re: [9fans] acid question

[Tim Newsham]

> This at least can be fixed by putting
> 	res->store.fmt = 'D';

Sure, but I'm really less concerned about the format associated
with the result of the comparison than I am about the fact that
although "a" and "b" show the same value, and "a-b" shows zero,
the "==" operator says they are not the same.  (Worse yet, if
I compare "a-b" with zero, it also fails).

> -rob

> >     a           0x800529a4
> >     b           0x800529a4
> >     a - b       0x00000000
> >     b - a       0x00000000
> >     a == b      0
> >     a < b       0x00000001
> >     b < a       0x00000000
> >     a > b       0
> >     b > a       1

Re: [9fans] acid question

[Tim Newsham]

>     a		0x800529a4
>     b		0x800529a4
>     a == b	0
>
> Whats going on here?  Signed vs. unsigned issues?

Indeed a signed issue -- acid integer values are stored
as vlongs.  The value of a symbol such as "main" is being
converted to a vlong by zero-extension, while the symbol
value from strace() (Symbol.value) is a signed long which
is sign-extended.  One fix is to change
/sys/src/cmd/acid/list.c from

   l->ival = sym->value;

to

   l->ival = (ulong)sym->value;

Or perhaps the Symbol value field in <mach.h> should be a ulong?

Tim N.

Re: [9fans] acid question

[Philippe Anel]

  Not needed in my opinion. Acid can also helps.
If we decide that function marked for unit testing with a known regexp
such as ending by _Test ...

defn utest()
{
     sl = symbols;
     while sl do {
         s = head sl;
         if (regexp(".*_Test", s[0])) then {
             call(s[2]);
         }
         sl = tail sl;
     }
}

This would call every function whose name is ending with _Test.

Phil;


On 7/17/2010 7:23 PM, Skip Tavakkolian wrote:
> nice.  i'll give it a try later.
>
> thinking outloud here; it seems that one could extend the capability
> of ?c for generating acid code to include creating call wrapper
> for any function that is marked for unit testing (perhaps another
> #pragma?)
>
>> This one sounds like to work (except it does not save registers) ...
>>
>> defn call(addr)
>> {
>> 	local pco;
>>
>> 	// we need stack space, and thus process stopped.
>> 	if status(pid) != "Stopped" then {
>> 		print("Waiting...\n");
>> 		stop(pid);
>> 	}
>>
>> 	// Backup orginal instruction
>> 	pco = fmt(*PC, bpfmt);
>>
>> 	// Put a temporary breakpoint
>> 	**PC = bpinst;
>>
>> 	// update stack pointer
>> 	*SP = *SP - 4;
>>
>> 	// make called function returning to this point
>> 	**SP = *PC;
>>
>> 	// set PC to function to call
>> 	*PC = addr;
>>
>> 	// do the call
>> 	startstop(pid);
>>
>> 	// restore original instruction
>> 	*PC = pco;
>> }
>>
>> Phil;
>
>

Re: [9fans] acid question

[Skip Tavakkolian]

nice.  i'll give it a try later.

thinking outloud here; it seems that one could extend the capability
of ?c for generating acid code to include creating call wrapper
for any function that is marked for unit testing (perhaps another
#pragma?)

> This one sounds like to work (except it does not save registers) ...
>
> defn call(addr)
> {
> 	local pco;
>
> 	// we need stack space, and thus process stopped.
> 	if status(pid) != "Stopped" then {
> 		print("Waiting...\n");
> 		stop(pid);
> 	}
>
> 	// Backup orginal instruction
> 	pco = fmt(*PC, bpfmt);
>
> 	// Put a temporary breakpoint
> 	**PC = bpinst;
>
> 	// update stack pointer
> 	*SP = *SP - 4;
>
> 	// make called function returning to this point
> 	**SP = *PC;
>
> 	// set PC to function to call
> 	*PC = addr;
>
> 	// do the call
> 	startstop(pid);
>
> 	// restore original instruction
> 	*PC = pco;
> }
>
> Phil;

Re: [9fans] acid question

[cinap_lenrek]

[-- Attachment #1: Type: text/plain, Size: 86 bytes --]

accedently hit Post... ignore the previous mail. nice trick by the way.

--
cinap

[-- Attachment #2: Type: message/rfc822, Size: 3537 bytes --]

From: Philippe Anel <xigh@bouyapop.org>
To: 9fans@9fans.net
Subject: Re: [9fans] acid question
Date: Sat, 17 Jul 2010 17:56:22 +0200
Message-ID: <4C41D2A6.7080402@bouyapop.org>



This one sounds like to work (except it does not save registers) ...

defn call(addr)
{
	local pco;

	// we need stack space, and thus process stopped.
	if status(pid) != "Stopped" then {
		print("Waiting...\n");
		stop(pid);
	}

	// Backup orginal instruction
	pco = fmt(*PC, bpfmt);

	// Put a temporary breakpoint
	**PC = bpinst;

	// update stack pointer
	*SP = *SP - 4;

	// make called function returning to this point
	**SP = *PC;

	// set PC to function to call
	*PC = addr;

	// do the call
	startstop(pid);

	// restore original instruction
	*PC = pco;
}

Phil;

Re: [9fans] acid question

[cinap_lenrek]

[-- Attachment #1: Type: text/plain, Size: 134 bytes --]

defn call(addr)
{
	local opc;

	opc = *PC;
	bpset(*PC);
	*SP = *SP - 4;
	**SP = *PC;
	*PC = addr;
	cont();
	bpdel(*PC);
}

[-- Attachment #2: Type: message/rfc822, Size: 3537 bytes --]

From: Philippe Anel <xigh@bouyapop.org>
To: 9fans@9fans.net
Subject: Re: [9fans] acid question
Date: Sat, 17 Jul 2010 17:56:22 +0200
Message-ID: <4C41D2A6.7080402@bouyapop.org>



This one sounds like to work (except it does not save registers) ...

defn call(addr)
{
	local pco;

	// we need stack space, and thus process stopped.
	if status(pid) != "Stopped" then {
		print("Waiting...\n");
		stop(pid);
	}

	// Backup orginal instruction
	pco = fmt(*PC, bpfmt);

	// Put a temporary breakpoint
	**PC = bpinst;

	// update stack pointer
	*SP = *SP - 4;

	// make called function returning to this point
	**SP = *PC;

	// set PC to function to call
	*PC = addr;

	// do the call
	startstop(pid);

	// restore original instruction
	*PC = pco;
}

Phil;

Re: [9fans] acid question

[Philippe Anel]

This one sounds like to work (except it does not save registers) ...

defn call(addr)
{
	local pco;

	// we need stack space, and thus process stopped.
	if status(pid) != "Stopped" then {
		print("Waiting...\n");
		stop(pid);
	}

	// Backup orginal instruction
	pco = fmt(*PC, bpfmt);

	// Put a temporary breakpoint
	**PC = bpinst;

	// update stack pointer
	*SP = *SP - 4;

	// make called function returning to this point
	**SP = *PC;

	// set PC to function to call
	*PC = addr;

	// do the call
	startstop(pid);

	// restore original instruction
	*PC = pco;
}

Phil;

Re: [9fans] acid question

[Philippe Anel]

  On 7/17/2010 10:33 AM, Skip Tavakkolian wrote:
> is there a way to do in acid the equivalent to gdb 'call'?
>
>
Did you try to write an acid function with something like this :

*PC=test

where 'test' is the name of the function to call such as:

void test(void)
{
     print("hello from test\n");
}

?

You'll have to save SP, PC, and maybe add a breakpoint somewhere and
even push the address of this breakpoint to SP before calling 'cont()'
acid function. But I think it will do the trick.

Phil;

Re: [9fans] acid question

[EBo]

> i've spent some time with acid and i think the answer is no.

What is the hold up?

> i wasn't referring to your reply to my acid suggestion on gsoc -- but
> my question is related to what you asked about unit testing.  if there
> is a way to call arbitrary functions in a program under acid's
> control, then it should be easy to generate acid scripts to do unit
> tests.

I would not say arbitrary, but you can start a program and play around in
the internals.  So, I think we agree.  One the other hand, a mixture of
little test programs and acid scripts might be able to do the same thing.

  EBo --

Re: [9fans] acid question

[Skip Tavakkolian]

i've spent some time with acid and i think the answer is no.

i wasn't referring to your reply to my acid suggestion on gsoc -- but
my question is related to what you asked about unit testing.  if there
is a way to call arbitrary functions in a program under acid's
control, then it should be easy to generate acid scripts to do unit
tests.

-Skip

>> is there a way to do in acid the equivalent to gdb 'call'?
>
> I have not spent any time with acid so I cannot say definitively, but if
> you are referring to the trick I did to develop an embedded regression test
> suite was to set break points and then test specific variables or continue
> to other break points which revealed the proper behavior as a pass or fail.
> I was able to put all this into a series of .gdbinit's.   Looking at the
> acid documentation it appears that there is an equivalent functionality.
> So as long as there is a way to read a configuration file when acid starts,
> then yes.  If not, then it should be relatively easy to add.
>
> I hope this answered your question, assuming I was the one you were asking
> the question of.
>
>   EBo --

Re: [9fans] acid question

[EBo]

> is there a way to do in acid the equivalent to gdb 'call'?

I have not spent any time with acid so I cannot say definitively, but if
you are referring to the trick I did to develop an embedded regression test
suite was to set break points and then test specific variables or continue
to other break points which revealed the proper behavior as a pass or fail.
I was able to put all this into a series of .gdbinit's.   Looking at the
acid documentation it appears that there is an equivalent functionality.
So as long as there is a way to read a configuration file when acid starts,
then yes.  If not, then it should be relatively easy to add.

I hope this answered your question, assuming I was the one you were asking
the question of.

  EBo --

[9fans] acid question

[Skip Tavakkolian]

is there a way to do in acid the equivalent to gdb 'call'?