[5581] in Athena Bugs
Re: rt 6.4R: MATLAB, crash in sprintf routine
daemon@ATHENA.MIT.EDU (sethf@ATHENA.MIT.EDU)
Thu Jul 19 21:16:57 1990
From: sethf@ATHENA.MIT.EDU
Date: Thu, 19 Jul 90 21:16:45 -0400
To: jik@ATHENA.MIT.EDU
Cc: bugs@ATHENA.MIT.EDU, sethf@ATHENA.MIT.EDU
In-Reply-To: bugs[5548]
>> What were you trying to do?
>> Determine how MATLAB's sprintf routine worked, and whether a
>> string was a legal argument. I typed
>> sprintf('Foo %s','foo')
>> What's wrong:
>> Segmentation violation
>>
>> What should have happened:
>> A less drastic error message.
> Sorry, but I disagree. You told matlab to write into a constant
> string, probably overwriting the memory reserved for that string.
> There is no way for it to check that, and in fact, the same thing
> would result if you did the same thing in a C program with the C
> sprintf() function.
> I consider segmentation faulting in response to what you did to fall
> quite soundly within the bounds of "wreaking havoc." They documented
> the function that way *because* they know that it's not possible to
> check for things such as what you did.
Sorry, I disagree with the disagreement :-). Your reply seems to
me to be based on a confusion between the forms of the C and MATLAB
sprintf. The C sprintf has the usage sprintf(buf, format, args), while
MATLAB's sprintf takes the form sprintf(format, args), with buf being
internal. That MATLAB statement is not equivalent to identical C
statement (which would indeed be erroneous), but rather to the C
sprintf(buf,"Foo %s","foo"), which is perfectly legal (buf being
previously properly declared). Or at least so my thoughts ran.
What are interpreters for, if not to check possibly
dangerous errors? I don't think what I typed was unreasonably odd or
unpredictable. In this context, I thought "wreak havoc" meant produce
an utter garbage string, not crash the entire program.
================
Seth Finkelstein
sethf@athena.mit.edu
