Oleg,
How is your idea an improvement over James' idea? It seems to me that a call to translate-logical-pathname would be far simpler and easier to review.
I do not read iterate, so if we are moving forward with your approach, please use loop or other standard CL iteration facilities.
Thanks, Hans Am 01.02.2014 10:29 schrieb "Left Right" olegsivokon@gmail.com:
Hi Hans,
The below hardly qualifies as a patch, just a general idea (I'll give it a try for now and will see if I can make it better):
(defun tbnl:request-pathname (&optional (request tbnl:*request*) (uri-prefix ".")) (let* ((path (tbnl:url-decode (tbnl:script-name request))) (parts (parse-namestring path)) (uri-parts (parse-namestring uri-prefix))) (iter (with pick := nil) (with uri-dirs := (cdr (pathname-directory uri-parts))) (for uri-part := (pop uri-dirs)) (for part :in (cdr (pathname-directory parts))) (when (or pick (not (equal uri-part part))) (setf pick t) (collect part :into uri)) (finally (return (make-pathname :host (pathname-host (tbnl:acceptor-document-root (tbnl::request-acceptor request))) :directory (cons :relative uri) :name (pathname-name parts) :type (pathname-type parts)))))))
------ test -----
(let ((tbnl:*acceptor* (make-instance 'rgol-acceptor :port 4242 :request-class 'rgol-request :document-root #p"rgol:www;" :message-log-destination #p"rgol:logs;messages.log" :access-log-destination #p"rgol:logs;access.log"))) (let ((r (make-instance 'tbnl:request :acceptor tbnl:*acceptor* :headers-in '(("Accept:*/*")) :uri "/foo/bar/qux/baz.ext"))) (let ((no-prefix (tbnl:request-pathname r)) (prefix (tbnl:request-pathname r "/foo/"))) (format t "~&no-prefix: ~s ~ with-prefix: ~s ~%~ resolved no-prefix: ~s ~%~ resolved prefix: ~s" no-prefix prefix (merge-pathnames no-prefix #p"rgol:www;") (merge-pathnames prefix #p"rgol:www;")))))
;; no-prefix: #P"RGOL:;FOO;BAR;QUX;BAZ.EXT" with-prefix: #P"RGOL:;BAR;QUX;BAZ.EXT" ;; resolved no-prefix: #P"RGOL:WWW;FOO;BAR;QUX;BAZ.EXT.NEWEST" ;; resolved prefix: #P"RGOL:WWW;BAR;QUX;BAZ.EXT.NEWEST"
If I can suggest: would it be worth to make request-pathname a generic function, rather then a plain function? I would imagine few have used this mechanism to implement their own path resolution scheme, but theoretically it could be useful. Its name and the way it's been used seem to ask for it to be a method of request class.
Best,
Oleg
On Sat, Feb 1, 2014 at 2:11 PM, Hans Hübner hans.huebner@gmail.com wrote:
Oleg,
James Anderson privately shared the observation that if the document root logical pathname is translated (using translate-logical-pathname) before
it
is merged to the URI's path, the problem would be solved. Can you give
that
a try? If that works for you, then I'd like to see Hunchentoot be
changed
so that it call translate-logical-pathname for the configuration paths
that
the user supplies, in the constructor. The behavior should be
documented.
I've also pondered whether it would be better to do the translation for every request, but James rightfully pointed out that one would want to
see
erroneous logical pathnames early on. Also, the overhead might be considerable, so I'm tending towards doing it at server initialization
time.
This means that if a logical host definition is changed at run time, the server must be restarted to pick up the new definition.
-Hans
2014-02-01 Left Right olegsivokon@gmail.com:
Ha! I see, I think I've given up on pathnames once, but now I've got more time to try to figure them out, so maybe one day I'll have that patch ;)
Thanks for the info, though!
Best,
Oleg
On Sat, Feb 1, 2014 at 12:32 PM, Hans Hübner hans.huebner@gmail.com wrote:
Oleg,
what would be needed is complete support for logical pathnames (which
is
not present at the moment). I believe that if the document root is a logical pathname, the incoming URL would need to be converted to a relative logical pathname before it is merged to the document root.
At this point, I am not prepared to work on this myself. As pathnames are hairy business, I would only want to merge a patch to support logical pathnames if it comes with thorough tests.
Alternatively, I would accept a documentation patch that says that if the document root is a logical pathname, it must not contain a directory
or
file component. :)
-Hans
2014-02-01 Left Right olegsivokon@gmail.com:
Thanks for the answer, Hans,
I'm looking now at the misc.lisp, create-folder-dispatcher-and-handler, am I right assuming this is the function that handles paths merging?
If so, I was thinking to override request-pathname on request to create a URI with the required bits of the pathname. This would seem to me like a better way to handle it. But I'm not certain what would it mean in terms of other bits of the code, is it likely to break other things?
Example:
(merge-pathnames (make-pathname :host "rgol" :name "game" :type "html") #p"rgol:www;") #P"RGOL:WWW;GAME.HTML.NEWEST"
The reason I want to do it this way is because merge-pathnames will only concatenate paths if the default-pathname is trivial: no directories, wildcards etc. So it seems to me that the original intention was to simply concatenate paths, but because for trivial pathnames merge-pathnames worked as concatenation, it was used. Does it make sense?
Best,
Oleg
On Sat, Feb 1, 2014 at 10:59 AM, Hans Hübner <hans.huebner@gmail.com
wrote:
Oleg,
the problem seems to be related how logical pathnames are merged. The partial pathname that is created by Hunchentoot and then merged
with
the document root pathname is considered to contain a directory
component
by CL. Therefore, the directory component that you are matching in your logical pathname host definition is not present when the two partial pathnames are merged. To work around this behavior, I'd recommend that you
define
a separate logical pathname host for your document root.
TEST-LOGICAL-PATHNAMES> (setf (logical-pathname-translations
"TEST")
'(("foo;*.*.*" "/tmp/*"))) (("foo;*.*.*" "/tmp/*")) TEST-LOGICAL-PATHNAMES> (directory (merge-pathnames "test.txt" #P"test:foo;")) NIL TEST-LOGICAL-PATHNAMES> (setf (logical-pathname-translations
"TEST")
'(("*.*.*" "/tmp/*"))) (("*.*.*" "/tmp/*")) TEST-LOGICAL-PATHNAMES> (directory (merge-pathnames "test.txt" #P"test:foo;")) (#P"/private/tmp/test.txt") TEST-LOGICAL-PATHNAMES> (directory (merge-pathnames "test.txt" #P"test:")) (#P"/private/tmp/test.txt")
-Hans
2014-01-31 Left Right olegsivokon@gmail.com:
> Sorry, I've copied the wrong log entry, this is the correct one: > > 127.0.0.1 - [2014-01-31 17:35:44] "GET /game.html HTTP/1.1" 404
188
> "-" > "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like > Gecko) > Chrome/28.0.1500.95 Safari/537.36" > > > On Fri, Jan 31, 2014 at 5:33 PM, Oleg Sivokon > olegsivokon@gmail.com > wrote: >> >> Hello list, >> >> There should be something very simple I've overlooked, yet I
can't
>> find >> it. Hunchentoot seems not to be able to locate the root
directory,
>> where >> I have my static content, and I can't get it to print any useful >> information about it. That's why I'm asking for your help. >> >> Below is my setup: >> >> (setf (logical-pathname-translations "rgol") >> '(... >> ("WWW;*.*.*" "/home/wvxvw/.../www/") >> ("WWW;*;*.*.*" "/home/wvxvw/.../www/*") >> ...)) >> >> (make-instance 'hunchentoot:acceptor :port 4242 >> :document-root #p"rgol:www;" >> :message-log-destination #p"rgol:logs;messages.log" >> :access-log-destination #p"rgol:logs;access.log") >> >> I've defined another handler, which doesn't depend on static
files,
>> and >> it works fine, however, when I try to access static files, the
log
>> record looks like this: >> >> 127.0.0.1 - [2014-01-31 17:12:40] >> "GET /img/made-with-lisp-logo.jpg HTTP/1.1" >> 404 206 "http://localhost:4242/game.html" >> "Mozilla/5.0 (X11; Linux x86_64; rv:23.0) Gecko/20100101 >> Firefox/23.0" >> >> But the file is definitely there, because if I try this in REPL: >> >> (directory #p"rgol:www;*.*") >> (#P"/home/wvxvw/.../www/game.html" >> ... more files ...) >> >> My version of Hunchentoot is: >> >> hunchentoot:*hunchentoot-version* >> "1.2.17" >> >> $ sbcl --version >> SBCL 1.1.2-1.fc18 >> >> Best, >> >> Oleg > >