httpd-apreq-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Joe Schaefer <joe+gm...@sunstarsys.com>
Subject Re: Apache::Upload API
Date Thu, 17 Jun 2004 13:41:52 GMT
Markus Wichitill <mawic@gmx.de> writes:

> One is the common case of simply storing an uploaded file. Under MP1,
> this was easy, first I try to create another hard link() for the spool
> file, and if that fails because the destination is on another
> filesystem, I use the filename to make a copy.
> 
> With the MP2 version, if link() fails, 

Bear in mind that link() fails far less in the MP2 version (since it
*will* copy the file across mount points if necessary).

> I either have to slurp() the whole file into memory and then save it,
> which is very inefficient with big files, or I have to save the data
> in chunks using the brigade API, which is rather inconvenient
> (especially for lazy Perl progammers) and probably not as efficient as
> letting the OS copy the file.
>
> I can't force the spool file to be created on the same filesystem where
> I want to store my files, since TEMP_DIR is usually ignored if there's
> already a temp directory specified by the environment.

See above remarks (sorry this change wasn't noted in CHANGES, it 
should have been there).

> The other scenario is when I want to process an uploaded file with a
> random CPAN module. These don't care for the brigade API, they usually
> accept file names, file handles, and maybe in-memory files. Since I
> can't rely on link(), I have to pass the latter. But some modules don't
> need to access the whole file, they only look at parts of it, like the
> directory at the end of a Zip archive. So again this is inefficient.
> 
> When I last looked at the apreq2 source a few months ago, I was under
> the impression that the code stores small uploads in memory, and only
> writes them into spool files from a certain size upwards. Which might be
> the reason why the filename() and fh() methods are not available
> anymore. But it seemed to me that to provide link(), the code does
> create files for smaller uploads when required. Couldn't this also be
> done to provide filename() and fh() again?

I'd *really* prefer to stay out of the IO game in Apache::Request
(moving filehandles between apache and perl is a gigantic mess).
I'd rather argue for some sort of brigade-as-filehandle mechanism
in APR::Brigade, precisely for the backwards-compatibility +
interoperability reasons above.

> I'd like to think that a lot of people would appreciate this, for
> efficiency, simplicity and last not least backwards compatibility.

-1 on tempname() magically reappearing, since it was never meant to
be in the public API.  fh() might be possible if there's a way to
graft it onto an APR::Brigade.

-- 
Joe Schaefer


Mime
View raw message