Re: [Bug-wget] [PATCH] Support metalink:file elements with a "path/file" format

[Tim Ruehsen]

On Tuesday, August 2, 2016 10:06:42 AM CEST Matthew White wrote:
> On Sat, 30 Jul 2016 21:23:56 +0200
> 
> Matthew White <address@hidden> wrote:
> > Hello!
> > 
> > I noticed that wget doesn't use the metalink:file element as the RFC5854
> > suggests.
> > 
> > https://tools.ietf.org/html/rfc5854#section-4.1.2.1
> > 
> > The RFC5854 specifies that the metalink:file could have a "path/file"
> > format. In this case wget should create the "path" tree and save the file
> > as "path/file", but it doesn't. Instead wget saves the file in the
> > working directory.
> > 
> > e.g. <file name="dirA/dirB/file.gz">
> > 
> > With this patch wget conforms to the RFC5854.
> > 
> > I made this patch working on the following branch:
> > master (latest 20cac2c5ab3d63aacfba35fb10878a2d490e2377)
> > git://git.savannah.gnu.org/wget.git
> > 
> > Let me know if this helps.
> 
> Hi,
> 
> After the suggestions of Tim, I fixed the patch (nice! fix the fix...). So,
> scratch the previous patch and use this one instead.
> 
> The function concat_strings() replaces the combination of malloc(),
> strlen(), and strcpy(). (Thanks Tim.)

In this special case (just one string to clone), please use xstrdup(). That is 
much less overhead.

> The description now follows the style of the other patches. (Thanks again
> Tim!)
> 
> You may consider this patch a Bugfix:
> * src/utils.c (unique_create, fopen_excl): cannot create a directory tree
> like "path/file".
> 
> The problem is that fopen_excl() doesn't create non-existing directories.
> What do you suggest?

IMO, saving metalink files should work 'as usual and expected'. That means, all 
the directory options should apply (see man wget / Directory Options). We also 
have to deal with 'escaping' file name sequences like ../ etc.
Not to forget character set conversions.

What about cases where you download https://myserver/file.tgz, and in the 
metalink file 'name' is set to 'xyz.doc' ? IMO, we should keep file.tgz, except 
the user stated --trust-server-names.

Maybe we should set up a a metalink document where we define how everything 
should work including corner cases. Next step would be to set up appropriate 
tests and fix the wget metalink code to survive the tests.

> Can we make this patch obsolete?

Basically yes.
But we see, there is still much to do around metalink ... but most of the 
needed code is already there. Most work will be the definition and the tests.

Regards, Tim

signature.asc
Description: This is a digitally signed message part.