This is the mail archive of the mailing list for the Cygwin project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: cygwin-1.3.21-1, problem with sparse file creation as default

> On Thu, Mar 13, 2003 at 04:09:54PM +0100, Markus Mauhart wrote:
> > I have a problem with the following new feature of cygwin-1.3.21-1
> > 
> > > - Create sparse files by default, when possible.  (Vaclav Haisman)
> > 
> > Couldnt it be made configurable, or removed ?
> > 
> > 
> > 1) good old file manager (winfile.exe from NT4 system) does not display
> > sparse files - so all newly created files (through gcc, or make,
> > or "cp con 123.txt") are now invisible. (beside browsing, I didnt test real
> > file operations like move/copy a folder containing some sparse files)
> You're actually still using winfile?!?  This is older than dirt.  AFAIK,
> winfile already collapsed on some features of NT4 SP4...

Never had a problem. Do you really think that NT4sp4 broke it (without
eventually fixing it in sp5,6) ? Works w/o problems on w2k; didnt work
with XPbeta/RC, but this has been fixed for XP's release ( -> probably
some VIPs at MS use winfile.exe ;-)
Allthough neither on w2k nor on XP I used it with sparse files or
encrypted files (but we know that no file GUI or text-app can support
every feature of every file system), but IIRC "reparse points" work.

> I've checked it, it's no problem to view the files in explorer under NT4. 

You mean w2k or wXp (or does NT4spx support ntfs5 including read/write
sparse files) ? Anyway, explorer on XP has no problem with the new
cygwin's sparse files.

> > 2) AFAICS its advantages are very sparse ;-) Only when extending
> > a file's size the sytem (ntfs5+) automatically adds 'sparse' clusters.
> > Otherwise (even when writing 10G of all zero) not a single sector
> > is spared. Only programs aware of win32-sparse files profit from this
> > existing file-attribute when explicitely marking a range as zero,
> > but IMHO this is a micro-profit: such program can replace the following
> > code ....
> Nope.  All applications using seek instead of blindly writing zeros
> to the file do profit.  And also this is default on modern UNIX boxes.

Now i'm confused: with "seek", did you mean the case I called "extending
a file's size" ? IMHO less than 0,01% of such file expansions really end
up with all-zero-clusters ... maybe first some KB zero are setup in the
file cache, but before the file is closed this changes.
"this is default on modern UNIX boxes" ... what ? And is it a property
of the filesystem-data, FS-driver or an OS feature ?

Another reason that makes me suspicious: ntfs5 with sparse files is
released since 3/2000, but nevertheless neither w2k nor wxp nor any of
the servers AFAIK provide even the option of creating all new files
in a directory or volume as sparse files - have the guys at MS missed
the performance benefits that cygwin-1.3.21-1 now claims, or do they
know it (their NTFS5x !) better ?

But note, after reading your remarks concerning the previous discussions
in the patches list I've found it and will go through it, maybe this
thread has enough new & good arguments to convince me and make me smarter.


Unsubscribe info:
Bug reporting:

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]