[tahoe-dev] wanted in webapi: a way to download a file only if it is a file

[Brian Warner]

> Still, this is non-obvious and I think the webapi would be improved  
> by providing a way to express "Just get me the file." without saying  
> "Get me the file or else some human-readable web page if it turns out  
> it isn't a file.".

Yeah. It's a question of what seems most obviously correct, and it 
depends upon how much time you spend hanging around apache, I guess. I 
think we can agree that a GET of a file should just return its contents. 
I think there's also a strong argument that a GET of a directory should 
give you a way to examine and/or manipulate that directory. I think I'm 
invoking Zipf's Law here: the shortest command should be the most 
frequent one, in this case it's a question of what URL the human types 
into their browser and what they expect that URL to do. So GET of a file 
gets the file, GET of a directory shows the directory in a 
browser-useful (ie HTML for humans) way.

But I quite agree that the nondeterministic behavior of GET on a node of 
  unknown type is a big problem for a program that's trying to behave 
correctly. Channeling Tyler for a moment, (and Tyler, forgive me if I'm 
misrepresenting your design aesthetic here), I can imagine an argument 
that a GET of a directory should return a special XML representation of 
that directory, marked by the Content-Type being something like 
x-application/allmydata-tahoe-dirnode-xml or something, and have it be 
delivered with an XSLT stylesheet that transforms the machine-meaningful 
directory contents into a human-manipulable HTML page. This approach 
would basically merge the GET(dirnode?t=json) with the GET(dirnode) 
(i.e. t=html) and make a single URL useful for both purposes. The 
program would do a GET of the URL, then examine the Content-Type to 
determine whether it should save the data to a file or if it should 
interpret it as a directory (and then either indicate an error to the 
user or perhaps to a recursive download if that weren't too surprising, 
say perhaps because the user had passed in a "-r" flag).

However, this sounds a bit crazy to me, partially because I don't know 
enough XSLT to even attempt to implement it. So I'd be happy with an 
indicator that the program could use to point out that it is only going 
to be happy with a file. GET(url?t=file) would signal an error if the 
url actually pointed to a dirnode. If the url refers to a file, then 
GET(url?t=file) would behave exactly the same as GET(url).

(thinking about further, it might be the case that the HTTP header that 
says which content-types you are willing to accept might be relevant 
here, except that then you'd want the file-fetch to say that it would 
accept everything *except* the magic directory type, and I don't know if 
the syntax allows for that sort of thing).

cheers,
  -Brian



> 
> Regards,
> 
> Zooko
> 
> _______________________________________________
> tahoe-dev mailing list
> tahoe-dev at allmydata.org
> http://allmydata.org/cgi-bin/mailman/listinfo/tahoe-dev