Boost C++ Libraries: Ticket #590: mkstemp

component, description changed; cc, severity set

Dave Abrahams — Fri, 06 Jul 2007 12:06:31 GMT

cc bdawes@… added
component None → filesystem
description modified (diff)
severity → Showstopper

Can't reassign to Beman until he logs into Trac at http://svn.boost.org and sets his email address.

owner, status changed

Eric Niebler — Wed, 11 Jul 2007 02:02:22 GMT

owner changed from nobody to Beman Dawes
status assigned → new

severity changed

Beman Dawes — Sun, 29 Jun 2008 21:31:57 GMT

severity Showstopper → Not Applicable

I'm considering #590 and #1090 as indications of interest and need, but feel that there are various details to be worked out first. mkstemp and similar functions seem pretty dated. The C++ approach suggested by #1090 is appealing, but more work needs to be done. Suggestions welcome.

--Beman

status changed; milestone set

Beman Dawes — Sat, 30 May 2009 09:54:54 GMT

status new → assigned
milestone → To Be Determined

No progress yet. Needs a design that abstracts away differences between POSIX and Windows. Contributions and/or suggestions welcome!

--Beman

Andrey Semashev — Fri, 03 Jul 2009 19:36:00 GMT

Not sure if this applies to this ticket, but I've been in need of temp files support many times, too. So here are my thoughts.

Maybe, on the low level, such functionality could be reached through fstreams extension: if it was possible to construct fstream from FILE* then the tmpfile function could be used to create temporary files. On the higher level a wrapper function could be provided, such as:

void open_temp_file(fstream& strm);

The function would internally call tmpfile or whatever other API to create the file and attach it to the stream.

Scoped files, while they look appealing for some cases, are limiting when the lifetime of the stream has to span across scopes. A simple wrapper derived from the fstream and on top of open_temp_file can be provided, but it should not be the only way to do it.

status, resolution changed

Beman Dawes — Fri, 11 Jun 2010 15:48:42 GMT

status assigned → closed
resolution None → fixed

The unique_path() function in Version 3 addresses the problem. Version 3 will ship with 1.44.

Thanks,

--Beman