Re: Exited with return code -1073741819

[David Wright]

On Thu 05 Apr 2018 at 06:30:52 (+0200), Helge Kruse wrote:
> Am 05.04.2018 4:35 vorm. schrieb "David Wright" <address@hidden>:
> > > It seems that beyond a certain level of complexity I get Exit. In the
> > > current example it's return code -1073741819, but it can be other codes.
> 
> > Googling 1073741819 is jaw-dropping. I don't think you'll find out
> > anything specific from that particular code.
> 
> Well, using a search machine to find an error code interpretation is not
> necesdarily sensible.

Eminently sensible in my experience. The days of consulting a shelf of
printed manuals are long gone.

But the hits returned have to be interpreted carefully. There may be
cases that are easily solved, but the wide variety of symptoms and
fixes shows that most of them will not apply to any individual case.
As you have spotted, the message is generic.

> The number -1073741819 (a bit different)

Identical to google, except that you have to escape the - (which
carries the meaning "exclude pages that contain this string) either
by quoting it "-1073741819", or by deleting it. The latter is
easier of course.

> can also
> represented as 0xC0000005 assuming thar the program use a 32 bit number
> range.
> That error code is a composition of the error number (5: access denied) and
> the error class. 0xC0000000 stands for generic error.
> So all you get from the code is "access denied". The actual error can only
> be found if Lilypond issues the function that returned this code. Do you
> think that the complete error message could be shared?

This argument has already been rehearsed:

http://lists.gnu.org/archive/html/lilypond-user/2017-07/msg00313.html

Tracing errors like this (assuming "access" here refers to something
resembling an out-of-bounds error) is notoriously difficult. Even when
you find some problem code that provokes it, just adding debug code
can make the problem apparently disappear or manifest itself in a
more benign manner (hitting the wrong place but still within bounds).

You could try running the program with strace (system trace) but this
only finds errors connected with system calls (eg many of the
permissions errors reported on those google hits). If the problem is
deep within some mathematical computations, say, then strace will be
no help at all.

So sometimes the answer is to look to ones own behaviour and see if
the way one's LP source is constructed could be "improved" so as not
to provoke the error. Unfortunately it's easy for this to look like
cargo cult behaviour.

> I also wonder why the error codes are displayed as decimal numbers.
> Wouldn't it improve tge user experience if the code is readable?

Yes, unfortunately when these composite numbers get passed around, the
receiver might not be aware of the significance of the composition,
or care.

PS to Karlin High. Unless there's a manifest cause of the error,
any reduction in the problem size is likely to "fix" it.
If a car's suspension rattles because it's overloaded, then any
attempt at diagnosis by removing potential sources of the rattle
will thereby reduce the weight and the rattling will stop: labour
costs spiral. That said, I did wonder about how the "splitting"
is being done and whether it's leading to unintended complexity.

Cheers,
David.