[9fans] sed oddity

[9fans] sed oddity

[hugo rivera]

Hi,
is there some reason why sed doesn't check for write errors on its
stdout? (or at least it doesn't report them)
I am implementing a fs, and I wasted my whole afternoon trying to figure out why
sed 300q file > mnt/data
doesn't say anything about the write error I was expecting.
Note that
sed 300q file | cat > mnt/data
shows my error message ;-)
Am I missing something?
Saludos

--
Hugo

Re: [9fans] sed oddity

[Uriel]

Interesting, this reminds me of a question I had: is there any command
that would read from stdin, and write to stdout, but if there was an
error when writing to stdout it would ignore it and continue reading
stdin? It is trivial to do it in C, but don't want to require an extra
program just to keep werc's error logs clean of spurious noise.

I guess based on what you said, that I might be able to use sed, but
only until your complaint is fixed :)

uriel

On Tue, Aug 18, 2009 at 4:41 PM, hugo rivera<uair00@gmail.com> wrote:
> Hi,
> is there some reason why sed doesn't check for write errors on its
> stdout? (or at least it doesn't report them)
> I am implementing a fs, and I wasted my whole afternoon trying to figure out why
> sed 300q file > mnt/data
> doesn't say anything about the write error I was expecting.
> Note that
> sed 300q file | cat > mnt/data
> shows my error message ;-)
> Am I missing something?
> Saludos
>
> --
> Hugo
>
>

Re: [9fans] sed oddity

[Gorka Guardiola]

On Tue, Aug 18, 2009 at 4:55 PM, Uriel<uriel99@gmail.com> wrote:
> Interesting, this reminds me of a question I had: is there any command
> that would read from stdin, and write to stdout, but if there was an
> error when writing to stdout it would ignore it and continue reading
> stdin? It is trivial to do it in C, but don't want to require an extra
> program just to keep werc's error logs clean of spurious noise.
>
> I guess based on what you said, that I might be able to use sed, but
> only until your complaint is fixed :)
>
> uriel
>

I am not sure by your description, it looks to me like you want either:

{cat || cat > /dev/null}

or

while(;){
      cat
}

--
- curiosity sKilled the cat

Re: [9fans] sed oddity

[erik quanstrom]

> is there some reason why sed doesn't check for write errors on its
> stdout? (or at least it doesn't report them)

/n/sources/patch/sederrors

while i agree that sed seems more complicated that i would
like when debugging, eating errors seems like the wrong thing
to do.  it seems easy enough >[2]/dev/null if the old behavior
is wanted.

the problem was that B* functions (Bopen, Bputc, Bputrune, Bterm)
were not consistently being checked for errors.

i wonder if it would make sense to add a ->error() member
to Biobufhdr.  since most programs just want to quit if they
have a bio error, they could just quit there without needing
lots tedious error checking.

perhaps ->error could just be a boolean.

- erik

Re: [9fans] sed oddity

[roger peppe]

2009/8/18 Uriel <uriel99@gmail.com>:
> Interesting, this reminds me of a question I had: is there any command
> that would read from stdin, and write to stdout, but if there was an
> error when writing to stdout it would ignore it and continue reading
> stdin? It is trivial to do it in C, but don't want to require an extra
> program just to keep werc's error logs clean of spurious noise.

it seems to me that dd -conv noerror
should do what you're after, but a quick inspection
of the source shows me that i'm wrong.

i think that could be construed as a bug.

gorka's suggestion isn't quite right, i think, as
you probably do want the cat to terminate when it
reads eof or an error, so something like:

e = write
while(~ $e write*){
    cat
    e = $status
}

might do the job better (untested).

Re: [9fans] sed oddity

[hugo rivera]

Sadly things like sed are out of my reach; I have no idea how to
program languages.
Thanks for the reply.

2009/8/18, erik quanstrom <quanstro@quanstro.net>:
> > is there some reason why sed doesn't check for write errors on its
>  > stdout? (or at least it doesn't report them)
>
>
> /n/sources/patch/sederrors
>
>  while i agree that sed seems more complicated that i would
>  like when debugging, eating errors seems like the wrong thing
>  to do.  it seems easy enough >[2]/dev/null if the old behavior
>  is wanted.
>
>  the problem was that B* functions (Bopen, Bputc, Bputrune, Bterm)
>  were not consistently being checked for errors.
>
>  i wonder if it would make sense to add a ->error() member
>  to Biobufhdr.  since most programs just want to quit if they
>  have a bio error, they could just quit there without needing
>  lots tedious error checking.
>
>  perhaps ->error could just be a boolean.
>
>
>  - erik
>
>


--
Hugo

Re: [9fans] sed oddity

[Uriel]

On Tue, Aug 18, 2009 at 6:00 PM, roger peppe<rogpeppe@gmail.com> wrote:
> 2009/8/18 Uriel <uriel99@gmail.com>:
>> Interesting, this reminds me of a question I had: is there any command
>> that would read from stdin, and write to stdout, but if there was an
>> error when writing to stdout it would ignore it and continue reading
>> stdin? It is trivial to do it in C, but don't want to require an extra
>> program just to keep werc's error logs clean of spurious noise.
>
> it seems to me that dd -conv noerror
> should do what you're after, but a quick inspection
> of the source shows me that i'm wrong.
>
> i think that could be construed as a bug.
>
> gorka's suggestion isn't quite right, i think, as
> you probably do want the cat to terminate when it
> reads eof or an error,

Exactly.

> so something like:
>
> e = write
> while(~ $e write*){
>    cat
>    e = $status
> }
>
> might do the job better (untested).

Nice trick, will try it out, thanks!

uriel