Re: [gnugo-devel] question about GTP

[aquarius]

Hallo Arend,

Am Wed, 26 Nov 2003 17:20:20 +0100 (CET) schrieb Arend Bayer <address@hidden>:

> Hallo aqaurius,
>
> wer Du auch immer bist: :-)

A happy Gnu Go user :-)

...
> Thanks for reminding us about this issue. I have nothing against your
> patch, but I'd like to understand the issue. On my linux system, the man page
> for setvbuf says:
>
> The setlinebuf function is exactly equivalent to the call:
>
>               setvbuf(stream, (char *)NULL, _IOLBF, 0);
>
> In the 2nd mail you cite above, you say that exactly this replacement
> has been the problem.
> setvbuf() is conforming to ANSI, while as setlinebuf() is not considered
> portable.
> Is MingW not ANSI-conformant in this respect?
>
> Arend

I think it tries hard to be ... but there must be a bug in MingW's
buffering related code since *the*same*(unchanged)*Gnu*Go*binary*
works well inside cmd.exe or command.com but not inside rxvt.exe or
when connected with pipes. (One can easily try out Gnu Go's GTP-Mode
manually in a shell / cmd.exe / rxvt.exe to reproduce this error)
As I noted in my message msg00004.html:

The underlying reason seems to be that MingW/Msys's isatty() is
clueless sometimes about the nature of the communication partner.

So MingW seems to refuse to obey to the setvbuf statement sometimes.
(at runtime, not at compile time)
I will try to investigate that further tomorrow ... perhaps the MingW
folks already have come to a conclusion ... but (as i noted in msg00004)
I personally do not seem to have enough C knowledge to pin that one
down and to advice the MingW developers ... sorry :-(

Happy Go - ing,
 aquarius
--