Re: new module 'fpurge'

[Eric Blake]

Bruno Haible <bruno <at> clisp.org> writes:

> 
> This generalizes the fpurge() function from the BSDs (FreeBSD, MacOS X,
> OpenBSD, NetBSD).

> #elif defined __sferror             /* FreeBSD, NetBSD, OpenBSD, MacOS X, 
Cygwin */
>   fp->_p = fp->_bf._base;
>   fp->_r = 0;
>   fp->_w = ((fp->_flags & (__SLBF | __SNBF) == 0) /* fully buffered? */
>           ? fp->_bf._size
>           : 0);
> #elif defined _IOERR                /* AIX, HP-UX, IRIX, OSF/1, Solaris, 

Unfortunately, this implementation of fpurge crashes cygwin.  It violates 
assumptions made elsewhere in stdio, to the point that running tests-fpurge 
under gdb hangs up the debugger, making it very difficult for me to figure out 
what went wrong.  I'm suspecting that the debugger does some sort of fflush
(NULL) between commands, but because the fpurge corrupted the test stream, the 
stdin/out/err streams also get messed up.  I'm trying to look more closely at 
newlib sources, and figure out what fflush does that fpurge needs to copy, 
while still leaving stdio in a happy state, but it may take a while.

It also lacks the HASUB/FREEUB macro usage present in the MacOS version of 
fpurge, which means you may have introduced a memory leak.

-- 
Eric Blake