Changed fhandler_* read and raw_read methods throughout

Christopher Faylor
Sat Dec 14 10:01:00 GMT 2002

On Sat, Dec 14, 2002 at 05:55:44PM +0100, Corinna Vinschen wrote:
>Hi Chris,
>On Fri, Dec 13, 2002 at 11:05:32PM -0500, Chris Faylor wrote:
>> To accommodate my recent pipe changes, I've changed all of the
>> fhandler_* read and raw_read methods.  I've changed them to void
>> functions whose second parameter is both the length and the return
>> value.
>> I've done this so that if the ReadFile in raw_read succeeds but a signal
>> happens shortly thereafter, the number of bytes read will be available
>> to the caller even though the thread which did the read has been
>> terminated.
>> It's possible that I got one of the many fhandler functions wrong when I
>> went through making this change so if you see odd behavior this is the
>> place to check.
>your changes disable tcsh's ability to read script files correctly. 
>AFAICS, you overoptimized the part looking for CRLF in
>fhandler_base::read() by not writing the last character if it happens
>not to be a \r.  This can only be observed when reading in O_TEXT mode.

I can see why this would happen but I wonder how I missed this when I
tested it.  Must have been tired.

>Another point is the "else if (copied_chars <= 0)" stuff after the first
>raw_read call.  As I understand that part, it actually should return
>here if raw_read returned 0 or -1 and copied_chars is 0 as well.  Your
>changes in the lines above don't copy the correct count of bytes read
>back into copied_chars as the original code did so at this point, it
>should now ask for the value of 'len' shouldn't it?  I attached a patch
>for review.

Thanks for the patch.  I ended up reorganizing this function YA so that
it exits from the bottom.  There is also a possibility of dereferencing
a pointer beyond the end of the passed in buffer, which still needed to
be handled, AFAICT.

I've checked in a new patch.  Does tcsh work now?

Btw, there are still some problems with the pipe reader that I'm working
on.  It is possible to get cygwin to block indefinitely on a pipe read
when a pipe is opened in text mode.

I sure do love text mode.  And Windows 95.

Thanks for alerting me to this problem.


More information about the Cygwin-developers mailing list