kicad-developers team mailing list archive

[Marco Serantoni <marco.serantoni@xxxxxxxxx>]

On 06/dic/2012, at 20:58, Dick Hollenbeck wrote:

> On 12/06/2012 12:44 PM, Marco Serantoni wrote:
>> Sure Dick,
>> Seems that fseek returns 0L, as that is where the internal buffer points.
>> The setvbuff destroys the internal buffer and when is used the read call, file operations begins where the buffer manager has leaved the real file descriptor (6k ahead).
>> 
>> Being the read already buffered, #ifdef it out has almost no impact on performance.
>> 
>> Hoping this makes now more sense,
> 
> Yes more, but the picture is still not crystal clear. 
> 
> Does setvbuf() work under any circumstances on your OS?  If it simply never works, then
> 
> {
>    be sure and make it conditional in *both* FILE_LINE_READER constructors.
> 
>    consider filing a bug report with your OS (glibc) maintainers.
> 
> }
> 
> The newer FILE_LINE_READER::FILE_LINE_READER( const wxString& ) will be getting more and
> more use with time.
> 
Dick 

I can't file any kind of bug report because reading a netlist, the call is misused, in the other cases it worked.

As i've said in the first mail.

The code parsing the netlist does more or less this:

[Reads the header to identify the netlist type]
[Rewind the stream]
[Creates a FILE_LINE_READER]
----> calls setvbuf 
[Proceed to parse the file]

And the documentation about the call explicitly says:
"This function should be called once the stream has been associated with an open file, but before any input or output operation is performed with it." 
" The setvbuf function may only be used after opening a stream and before any other operations have been performed on it."

Ref:
http://linux.about.com/library/cmd/blcmdl3_setvbuf.htm
http://www.cplusplus.com/reference/cstdio/setvbuf/

And sadly is called after having identified the netlist kind (reads + rewind).

Marco