Re: [Boost-users] boost::filesystem problems

From: Carlo Wood (carlo_at_[hidden])
Date: 2004-08-18 07:19:57

On Tue, Aug 17, 2004 at 11:55:22PM -0700, Victor A. Wagner Jr. wrote:
> At Tuesday 2004-08-17 17:16, you wrote:
> >On Tue, Aug 17, 2004 at 02:40:14PM -0700, Delfin Rojas wrote:
> >> Hi Carlo,
> >>
> >> I just wanted to point out native_file_string is part of
> >> boost::filesystem::path and copy_file and rename_file (actually called
> >> rename) are part of boost::filesystem (operations.hpp).
> >
> >I am aware of that.  The point is that I had a need to rewrite them.
> >That seems to indicate that the boost implementation is flawed.
> 
> perhaps you could enlighten us further?
> "doesn't work" is a lousy bug report
> "flawed" isn't a lot better.

My problem is that Beman is not here.  It seems to make little
sense to try to convince people when in the end nobody is going
to make changes to boost when he is not here.

But I can tell you what I improved of course.  Lets start with
just one case.

boost::filesystem::copy_file starts with truncating the target
file - and when an error occurs, it throws - but STILL did truncate
the target file (assuming it already existed) (the exception at
the bottom of the function).

This is not how I want exceptions to work; when an exception is
thrown from a function then nothing must have been changed - as
if the exception is thrown at the start of the function.

My implementation of copy_file either succeeds completely or doesn't
do anything: it is atomic.

I need this for robustness.  I don't write applications that '"Oops"
some error occured - sorry, your configuration file was erased - no
there is no backup, I was trying to make one namely'.  I have a gard my
reputation to write robust and bug free code ;).

The other changes are mostly also related to the need for a better
defined post-condition in case of errors.

-- 
Carlo Wood <carlo_at_[hidden]>