Re: [PATCH 2/2] ios: Introduce IO stream device for stdin, stdout, stderr.

[Egeyar Bagcioglu]

On 5/10/20 9:53 PM, Jose E. Marchesi wrote:
>      +static char *
>      +ios_dev_stream_handler_normalize (const char *handler)
>      +{
>      +  if (strcmp (handler, "stdin")
>      +      && strcmp (handler, "stdout")
>      +      && strcmp (handler, "stderr"))
>      +    return NULL;
>      +
>      +  return strdup (handler);
>      +}
>
> I think we should introduce a new flag to use in ios_open,
> IOS_F_STREAM.  This way, ios_open can do the following:
>
> - If IOS_M_RDONLY & IOS_F_STREAM -> Use the stream IOD to open the given
>    filename.
>
> - If IOS_M_WRONLY & IOS_F_STREAM -> Use the stream IOD to open the given
>    filename.
>
> At the Poke level this translates to:
>
> open ("foo/bar", IOS_F_STREAM & IOS_M_RDONLY)
>   -> Open the file foo/bar as a read-only stream.
>
> open ("foo/bar", IOS_F_STREAM & IOS_M_WRONLY)
>   -> Open the file foo/bar as a write-only stream.
>
> open ("foo/bar", IOS_F_STREAM & IOS_M_RDWR)
>   -> ERROR.  Not possible to open a stream for both read and write.
>
> This means that handler of the stream IOD won't ever be used, so it can
> just return NULL in all cases...

The filename is the handler, right? So "foo/bar" in this case. Plus, it 
is used in .info ios too.
Also, if handler_normalize returns NULL, that is a rejection of the 
handler by that IO device. It would eventually be opened as a regular file.

> This will allow us to poke character devices, and named pipes created
> with mkfifo.

Right, I can add the flag IOS_F_STREAM. I need to check if they are read 
and written to differently. Assuming it wouldn't be too complicated, I 
can add the support for them too.

Thanks
Ege