Re: Core utils 5.0 - "csplit" pipe memory error

[Nikola Milutinovic]

> "Nikola Milutinovic" <address@hidden> wrote:
> > I'm not subscribed to the list, so please CC to me.
> >
> > I believe I have located a problem with "csplit". I'm running as a part
of a pipe and it bombs my script with "pipe memory exhausted". This is the
script:
>
> Thanks for the report.
> A little more info would help track this down.

Sure.

> What was the exact diagnostic?

The pipe sequence before "csplit" builds an ASCII stream, quite large. There
are delimiter lines on regular distances. When "csplit" is in the pipe
sequence, I recieve two kinds of errors, depending on which shell is
running. With BASh-2.05 I recieve: "csplit: error, pipe memory exhausted".
With ZSh-4.1 I get "error: sed '1d' pipe broken". I have suspected shells as
culprits, but the same shells work on Tru64 UNIX and DEC's "csplit". Also,
using GNU core-utils-5.0 "csplit" gives the exactly same error.

> If csplit is failing, the diagnostic should start with `csplit:'.

It more refers to pipe, either being broken or memory exhausted.

> In any case, please tell us how big the input is.

The input is quite large 200 MB of ASCII data or more.

I have observed one other curiosity and difference in behaviour of two
"csplit"s. DEC's csplit waits for the "front" part of the pipe to finish
it's "thing". When I run it, I see nothing for a very long time, no file
gets created and then "csplit" starts and runs through the stream. GNU's
"csplit" tries to "run along", as the pipe's stream is formed. When I run
it, I see "csplit" running from the beginning of the whole operation, not
waiting for the "sed" and "awk" to finish pumping data and then it breaks.

Could it be that there is some bad pipe management at work here?

Nix.