On 7/25/2011 4:33, Warp wrote:
> Darren New<dne### [at] sanrrcom>  wrote:
>> On 7/24/2011 3:38, Orchid XP v8 wrote:
>>> On 23/07/2011 06:37 PM, Darren New wrote:
>>>
>>>> Interesting.
>>>
>>> I'm ticked by "C++ crashes a lot if it's immature".
>>>
>>> What, if C++ is immature? Or the C++ codebase is immature? Or perhaps you
>>> mean if the C++ developers are immature? ;-)
>
>> I think they meant your code crashes a lot if it's written in C++ and hasn't
>> had lots of debugging.
>
>    I think it depends a lot on the competence of the C++ programmer(s).
> Granted, becoming a very competent C++ programmer (whose code seldom crashes,
> or if it does, the crashes get caught very early on the development cycle)
> may require more work than with some other languages, but it's not impossible.

Well, I'd say it takes more time and more development. So if you count a 
single program, or if you count all the programs over your career, the 
comment is probably true. I.e., it also takes longer for a competent 
programmer to get expert at C++ than it does for a competent programmer to 
get expert at a simpler language.

>    I wonder if something similar ever happens with the "safer" programming
> languages (eg. you get a "null pointer exception" or some other similar
> error and have absolutely no idea why, and no debugging tool is helping).

Occasionally. Usually it's something like running out of file handles, if 
it's a low-level problem. I had that in Tcl once, and of course the thing 
would have to run for hours before it ran out of handles, and I just 
couldn't find where the problem was. The code was too complex in the error 
handling, and I for whatever reason bashed my head on it for a couple of 
days rather than refactor it, and I couldn't figure out which of the files 
it was running out of (i.e., which of the several "open" calls wasn't 
matching a "close").

The other kind of problem that's difficult is the large-scale logic problem. 
I had something that could rebuild the cache from the data store (to 
simplify the description). Then I added code to record changes to the data 
store and apply them incrementally to the cache.  And once every few days, 
that would crash and fall back to rebuilding from scratch. And it took me a 
while to figure out if someone added X and then deleted X all in one cycle, 
when it tried to propagate "add X" to the cache, it was already gone from 
the data store, so it gave up and fell back to the full rebuild. Fundamental 
mistakes in architecture or algorithm don't get any easier with safer or 
simpler languages.

The difference between the two is, you can reason from the program text. I 
knew I was running out of file handles because somewhere in the code that 
did open/read/write/close, one of the paths out of that code wasn't doing 
the close. I had very high confidence that it wasn't some other third party 
library stomping on my stack and changing the file handle variable to some 
other value.

But, in all honesty, the number of "I can't find this nightmare bug" I've 
encountered in safe languages is about the same as the number of "I found 
the bug, and it's in the compiler/interpreter" kinds of nightmare bugs.

I can't tell you the number of times in unsafe languages where I spent three 
days trying to figure out what was actually broken. To be fair, usually when 
I'm using unsafe languages, it's often because the machine isn't powerful 
enough or standard enough to support something like valgrind or even gdb, or 
you're missing so much source that such tools aren't useful, so it's purely 
reasoning from the code you actually have.

-- 
Darren New, San Diego CA, USA (PST)
   "Coding without comments is like
    driving without turn signals."

Post a reply to this message

On 7/25/2011 6:42, Warp wrote:
>    But if the null pointer is accessed in the depths of some obscure system
> library which gets called by the main runloop of the program (rather than
> being directly called by your code), you might end up in a situation where
> the stack trace is completely unhelpful and you have no idea whatsoever
> where that null pointer is coming from in your code (which might happen if
> you are not using that precise system library directly, but it's a library
> used by a library used by a library which you are using, but you don't know
> what this chain might be, as the implementation of the library you are using
> is hidden behind its public interface).

The stack trace in something like Java or Tcl is as useful as a full stack 
trace from gdb is in C, when you haven't actually clobbered memory in the 
unsafe languages.  I.e., you walk up the stack until you find your code that 
invokes the not-your-code, then you figure out what happened from there.

With the more OO stuff, the problem is figuring out which object you're 
talking about, IME. An interactive debugger that lets you examine things at 
the object-graph level solves that problem for the most part.

-- 
Darren New, San Diego CA, USA (PST)
   "Coding without comments is like
    driving without turn signals."

Post a reply to this message

Darren New <dne### [at] sanrrcom> wrote:
> The stack trace in something like Java or Tcl is as useful as a full stack 
> trace from gdb is in C, when you haven't actually clobbered memory in the 
> unsafe languages.  I.e., you walk up the stack until you find your code that 
> invokes the not-your-code, then you figure out what happened from there.

  But that's the problem: The buggy part of your code might not be in the
call stack at all anymore at the time the null pointer exception (or whatever
other symptom) happens.

  I know what a stack trace is and how to read it, and why I mentioned
several times that "the debugger is of no help whatsoever".

-- 
                                                          - Warp

Post a reply to this message

On 7/25/2011 9:20, Warp wrote:
>    But that's the problem: The buggy part of your code might not be in the
> call stack at all anymore at the time the null pointer exception (or whatever
> other symptom) happens.

Well, sure. I thought I had expressed it more clearly than that.

In the case that the broken code isn't the code in the call stack, you debug 
it the same way you'd debug code in C or C++ that wasn't doing undefined 
unsafe stuff - you'd figure out the infection that led to the error, figure 
out the defect that led to the infection, and correct the defect.[*]

For example, if the null came from an instance variable, you'd find the 
place the instance variable got initialized, breakpoint it, and see if it 
gets set. If not, figure out why not. If so, figure out where it gets set 
*again*, by breakpointing any assignment that sets it back to null. Etc.

The only difference in an unsafe language is you either need a debugger that 
can breakpoint not based on source code (i.e., that you can breakpoint based 
on writing to the memory via a wild pointer, which I imagine is most of them 
these days at least on normal desktop-class machines). It's harder to 
analyze because if you *can't* find the problem, you can't just say "find me 
every reference to this variable in the codebase."



[*] Those are apparently the technical terms.
"Error: Program gave wrong results."
"Infection: Program state that is not as it should be."
"Defect: Code that made the program state other than what it should be."

-- 
Darren New, San Diego CA, USA (PST)
   "Coding without comments is like
    driving without turn signals."

Post a reply to this message

Warp <war### [at] tagpovrayorg> wrote:
>   Of course with C++ (and C and other C-compatible languages) the rare
> circumstance where you get a nightmarish bug can be pretty fun.

  It's even more fun when you painstakingly go through the entirety of
your code, commenting out one piece at a time, every time testing if the
crash still happens (a very tedious task because the crashes happen at
random and are not guaranteed to happen every time), until you are left
with basically nothing else than a completely bare-bones "main" function,
and the crash is still happening. At that point you realize that the problem
is not in your code after all but somewhere else, and you still have no idea
where, or how it should be fixed.

  (On one hand it's a relief to know that I have not made an out-of-bounds
access or whatever, ie. my programming expertise really works. On the other
hand it sucks because the bug is still there, and now it's even more of a
mystery of why it happens and how it should be fixed.)

  (And yes, talking from experience here.)

-- 
                                                          - Warp

Post a reply to this message

On 7/27/2011 8:39, Warp wrote:
>    It's even more fun when you painstakingly go through the entirety of
> your code,

I did that once when we were having these terrible crashes in a 
large-for-the-time C program. Printed out an inch-thick stack of code and 
read it line by line to make sure I hadn't screwed up some allocation. 
(Found one bug that way, too, but it wasn't getting triggered by this case.) 
But the problem wasn't in my code, but rather the supervisor's. He said 
"What do we do about it?" I said "I tried to read your code, but I couldn't 
even follow the top-level procedures, so you have to go thru it line by line 
looking for the bug." He said "I don't have time to do that." I said "Then 
you don't have time to fix the bug. I can't do it for you, because I don't 
know what you expect your code to be doing."

Long story short, it turns out he'd been giving the wrong call-tree to the 
overlay linker. So he not only didn't know what his bug was, he wasn't even 
sure what order his code got called in.

-- 
Darren New, San Diego CA, USA (PST)
   How come I never get only one kudo?

Post a reply to this message