wget/TODO

                                                       -*- outline -*-

This is the to-do list for GNU Wget.  There is no timetable of when we
plan to implement these features -- this is just a list of features
we'd like to see in Wget, as well as a list of problems that need
fixing.  Patches to implement these items are likely to be accepted,
especially if they follow the coding convention outlined in PATCHES
and if they patch the documentation as well.

The items are not listed in any particular order (except that
recently-added items may tend towards the top).  Not all of these
represent user-visible changes.

* Change the file name generation logic so that redirects can't dictate
  file names (but redirects should still be followed).  By default, file
  names should be generated only from the URL the user provided.  However,
  with an appropriate flag, Wget will allow the remote server to specify
  the file name, either through redirection (as is always the case now)
  or via the increasingly popular header `Content-Disposition: XXX;
  filename="FILE"'.

  The file name should be generated and displayed *after* processing
  the server's response, not before, as it is done now.  This will
  allow trivial implementation of -nc, of O_EXCL when opening the
  file, --html-extension will stop being a horrible hack, and so on.

* -O should be respected, with no exceptions.  It should work in
  conjunction with -N and -k.  (This is hard to achieve in the current
  code base.)  Ancillary files, such as directory listings and such,
  should be downloaded either directly to memory, or to /tmp.

* Implement digest and NTLM authorization for proxies.  This is harder
  than it seems because it requires some rethinking of the HTTP code.

* Rethink the interaction between recur.c (the recursive download code)
  and HTTP/FTP code.  Ideally, the downloading code should have a way
  to retrieve a file and, optionally, to specify a list of URLs for
  continuing the "recursive" download.  FTP code will surely benefit
  from such a restructuring because its current incarnation is way too
  smart for its own good.

* Both HTTP and FTP connections should be first-class objects that can
  be reused after a download is done.  Currently information about both
  is kept implicitly on the stack, and forgotten after each download.

* Restructure the FTP code to remove massive amounts of code duplication
  and repetition.  Remove all the "intelligence" and make it work as
  outlined in the previous bullet.

* Add support for SFTP.  Teach Wget about newer features of more
  recent FTP servers in general, such as receiving reliable checksums
  and timestamps.  This can be used to implement really robust
  downloads.

* Wget shouldn't delete rejected files that were not downloaded, but
  just found on disk because of `-nc'.  For example, `wget -r -nc
  -A.gif URL' should allow the user to get all the GIFs without
  removing any of the existing HTML files.

* Be careful not to lose username/password information given for the
  URL on the command line.  For example,
  wget -r http://username:password@server/path/ should send that
  username and password to all content under /path/ (this is apparently
  what browsers do).

* Don't send credentials using "Basic" authorization before the server
  has a chance to tell us that it supports Digest or NTLM!

* Add a --range parameter allowing you to explicitly specify a range
  of bytes to get from a file over HTTP (FTP only supports ranges
  ending at the end of the file, though forcibly disconnecting from
  the server at the desired endpoint would work).  For example,
  --range=n-m would specify inclusive range (a la the Range header),
  and --range=n:m would specify exclusive range (a la Python's
  slices).  -c should work with --range by assuming the range is
  partially downloaded on disk, and contuing from there (effectively
  requesting a smaller range).

* If multiple FTP URLs are specified that are on the same host, Wget should
  re-use the connection rather than opening a new one for each file.
  This should be easy provided the above restructuring of FTP code that
  would include the FTP connection becoming a first-class objects.

* Try to devise a scheme so that, when password is unknown, Wget asks
  the user for one.  This is harder than it seems because the password
  may be requested by some page encountered long after the user has
  left Wget to run.

* If -c used with -N, check to make sure a file hasn't changed on the server
  before "continuing" to download it (preventing a bogus hybrid file).

* Generalize --html-extension to something like --mime-extensions and
  have consult mime.types for the preferred extension.  Non-HTML files
  with filenames changed this way would be re-downloaded each time
  despite -N unless .orig files were saved for them.  (#### Why?  The
  HEAD request we use to implement -N would still be able to construct
  the correct file name based on the declared Content-Type.)

  Since .orig would contain the same data as non-.orig, the latter
  could be just a link to the former.  Another possibility would be to
  implement a per-directory database called something like
  .wget_url_mapping containing URLs and their corresponding filenames.

* When spanning hosts, there's no way to say that you are only
  interested in files in a certain directory on _one_ of the hosts (-I
  and -X apply to all).  Perhaps -I and -X should take an optional
  "hostname:" before the directory?

* --retr-symlinks should cause wget to traverse links to directories too.

* Make wget return non-zero status in more situations, like incorrect HTTP auth.
  Create and document different exit statuses for different errors.

* Make -K compare X.orig to X and move the former on top of the latter if 
  they're the same, rather than leaving identical .orig files laying around.

* Make `-k' check for files that were downloaded in the past and convert links 
  to them in newly-downloaded documents.

* Devise a way for options to have effect on a per-URL basis.  This is very
  natural for some options, such as --post-data.  It could be implemented
  simply by having more than one struct options.

* Add option to clobber existing file names (no `.N' suffixes).

* Add option to only list wildcard matches without doing the download.  The same
  could be generalized to support something like apt's --print-uri.

* Handle MIME types correctly.  There should be an option to (not)
  retrieve files based on MIME types, e.g. `--accept-types=image/*'.
  This would work for FTP by translating file extensions to MIME types
  using mime.types.

* Allow time-stamping by arbitrary date.  For example,
  wget --if-modified-after DATE URL.

* Make quota apply to single files, preferrably so that the download of an
  oversized file is not attempted at all.

* When updating an existing mirror, download to temporary files (such as .in*)
  and rename the file after the download is done.

* Add an option to delete or move no-longer-existent files when mirroring.

* Implement uploading (--upload=FILE URL?) in FTP and HTTP.  A beginning of
  this is available in the form of --post-file, but it should be expanded to
  be really useful.

* Make HTTP timestamping use If-Modified-Since facility.

* Add more protocols (such as news or possibly some of the streaming
  protocols), implementing them in a modular fashion.

* Add a "rollback" option to have continued retrieval throw away a
  configurable number of bytes at the end of a file before resuming
  download.  Apparently, some stupid proxies insert a "transfer
  interrupted" string we need to get rid of.