| msg50852 - (view) |
Author: Damien Miller (djmdjm) |
Date: 2006-08-10 06:57 |
| Hi, tempfile.NamedTemporaryFile provides a good interface to creating temporary files, but its insistence on deleting the file upon close is limiting. The attached patch adds an optional parameter to NamedTemporaryFile that allows persistence of the temp file after it has been closed. One use-case that demonstrates where keeping the temporary file around is handy would be when you need to safely create and fill a temp file before it is atomically moved into place with os.rename(). E.g. def update_conf(conf_path): old = open(conf_path) tmp = tempfile.NamedTemporaryFile(prefix = os.basename(conf_path), \ dir = os.dirname(conf_path), delete = False) for l in old: tmp.write(l.replace('war', 'peace')) close(old) close(tmp) os.link(conf_path, conf_path + ".old") os.rename(tmp.name, conf_path) |
|
|
| msg50853 - (view) |
Author: Damien Miller (djmdjm) |
Date: 2006-08-10 06:58 |
| Logged In: YES user_id=1359232 oops, wrong Category: this should be Lib and not Modules |
|
|
| msg50854 - (view) |
Author: Brett Cannon (brett.cannon) *  |
Date: 2006-08-24 23:52 |
| Logged In: YES user_id=357491 Why can't you store into an instance of StringIO instead of a temp file? |
|
|
| msg50855 - (view) |
Author: Damien Miller (djmdjm) |
Date: 2006-08-25 00:22 |
| Logged In: YES user_id=1359232 As far as I can tell, StringIO doesn't actually create a filesystem object that can be manipulated. |
|
|
| msg50856 - (view) |
Author: Damien Miller (djmdjm) |
Date: 2006-08-25 04:27 |
| Logged In: YES user_id=1359232 Here is an diff that includes a regress test |
|
|
| msg50857 - (view) |
Author: Brett Cannon (brett.cannon) * |
Date: 2006-08-26 06:40 |
| Logged In: YES user_id=357491 Right, it doesn't create a filesystem file. But that is the point. You work in memory and then write to your final destination as needed. Your code you have pasted in the description does nothing special that requires the use of a temporary file. You can just write into a StringIO object, skip the os.link call, and then just write out to the final file location. |
|
|
| msg50858 - (view) |
Author: Damien Miller (djmdjm) |
Date: 2006-08-26 06:47 |
| Logged In: YES user_id=1359232 Well, that would a) not be an atomic replacement and b) you miss (or would have to reimplement) the mkstemp() like behaviour. |
|
|
| msg50859 - (view) |
Author: Georg Brandl (georg.brandl) * |
Date: 2007-03-13 18:32 |
| Thanks for the patch, committed with doc changes as rev. 54344. |
|
|