X-Authentication-Warning: delorie.com: mail set sender to djgpp-bounces using -f
From: Jason Mills <jmills AT cs DOT mun DOT ca>
Newsgroups: comp.os.msdos.djgpp
Subject: Re: fstream and fflush
Date: Fri, 22 Oct 2004 08:08:12 -0230
Organization: Memorial University of Newfoundland
Lines: 50
Message-ID: <clanul$8p9$1@coranto.ucs.mun.ca>
References: <cl92jr$3ro$1 AT coranto DOT ucs DOT mun DOT ca> <cl96k3$45r$1 AT coranto DOT ucs DOT mun DOT ca> <c5jhn011q2617u650ginng48u9dnrd5b3j AT 4ax DOT com>
NNTP-Posting-Host: jmills.ccore.mun.ca
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit
X-Trace: coranto.ucs.mun.ca 1098441493 9001 134.153.38.195 (22 Oct 2004 10:38:13 GMT)
X-Complaints-To: usenet AT coranto DOT ucs DOT mun DOT ca
NNTP-Posting-Date: Fri, 22 Oct 2004 10:38:13 +0000 (UTC)
User-Agent: Mozilla Thunderbird 0.7.3 (Windows/20040803)
X-Accept-Language: en-us, en
In-Reply-To: <c5jhn011q2617u650ginng48u9dnrd5b3j@4ax.com>
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Reply-To: djgpp AT delorie DOT com
Errors-To: nobody AT delorie DOT com
X-Mailing-List: djgpp AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com
Precedence: bulk

Brian Inglis wrote:

> On Thu, 21 Oct 2004 18:06:19 -0230 in comp.os.msdos.djgpp, Jason Mills
> <jmills AT cs DOT mun DOT ca> wrote:
> 
> 
>>Jason Mills wrote:
>>
>>>After a few hours or debugging, I reached the conclusion
>>>that to flush a C++ fstream, you must either fstream::close(),
>>>or call fflush(NULL). Is this normal? Or is it a bug?
>>
>>Actually, the above is not correct. Sorry.
>>
>>After more tests, I discovered that fstream::flush() and fflush(NULL)
>>both will not flush the stream if I am writing rapidly to the stream.
>>If I sleep(), (maybe if I a do other work, I have not checked)
>>the stream is eventually flushed. Can someone explain?
> 
> 
> How are you testing this? From another window, or another system on a
> network?

In the program, I noticed stat() returns a file size of 0 until the
file is closed, or a sleep() is called. You could simply test this from
another window as well (if you mean doing a 'dir' in that window).

> The flush() operations just write the program's buffer(s) to the DOS
> file buffer(s), it doesn't write the DOS file buffer(s) to the file
> system disk cache, nor flush the file system disk cache to the disk
> immediately. You need to call fsync(), and/or sync(), and/or
> _flush_disk_cache(), to do the rest.

I was unaware of this because the program is being ported from Linux,
where it worked flawlessly: fsync(), sync(), etc. is apparently
unecessary.

> 
> Even then I'm not sure that Windows makes cache updates visible
> immediately (like Unix) to other processes or systems on a network
> before the cache is flushed to disk and/or the file is closed,
> depending on when it updates the clusters in the FAT or blocks used in
> the FS. 

"Like UNIX" is the key. The program is port of a linux program. There
the file size is reported correctly as soon as a flush is called. (see
related reply to Hans-Bernhard Broeker for details of what I'm trying
to do)

Jason