Add the hosting wish list relevant for #53

[privoxy.git] / TODO

$Id: TODO,v 1.105 2014/05/13 11:15:28 fabiankeil Exp $

Some Privoxy-related tasks, sorted by the time they
have been added, not by priority.

The latest version should be available at:
http://ijbswa.cvs.sourceforge.net/viewvc/ijbswa/current/TODO

1)  Add some regression tests. Filters should be tested automatically
    (variables too). Could probably reuse large parts of Privoxy-Filter-Test.
    Note that there is currently work in progress to leverage curl's
    test suite which could be used for this as well.

3)  Fix some more XXX: comments.

6)  Remove actions that aren't needed anymore:

    content-type-overwrite should probably stay as it's also
    used by some of the CGI pages (XXX: name them).

    crunch-client-header and crunch-server-header should probably
    go, their only advantage is that their search strings can be
    controlled through the CGI pages, other than that they only
    have disadvantages.

    crunch-if-none-match can be replaced with a header filter.

    prevent-compression has a misleading name and could
    be replaced with a header filter.

7)  force-text-mode has a stupid name and should probably
    be renamed to force-filter-mode.

8)  handle-as-empty-document and handle-as-image should
    be merged to something like handle-as{something} to
    prevent them from being activated at the same time.

10) There's a bug in the CGI editor that turns the
    first section's "Insert new section below" into
    a "Insert new section above" button.

11) CGI templates should use semantically-correct HTML
    and scale properly.

12) Support pipelining for outgoing connections.

14) Allow to filter POST parameters.

15) If trusted CGI pages are requested without trusted referrer,
    set the status code to 403 instead of 200.

16) Filter SSL encrypted content as well.

    At the beginning we could use a unencrypted connection between
    client and Privoxy, and use an encrypted connection between
    Privoxy and the server.

    This should be good enough for most of the content the
    user would want to filter.

19) enable-forward-fallback. Syntax? Suggested by K.R.

21) User Manual delivery doesn't accept multiple slashes. Should it?

22) Verify action files properly (Including arguments) and
    act accordingly (should probably intercept all requests
    with a "Invalid option foo detected" CGI page).

23) Do the same in case of syntax errors in the configuration file,
    instead of just exiting or ignoring the problem.

25) Handle multiple filters with the same name better. Reject them?

26) Let show-url-info detect clearly invalid URLs.

27) Make errno logging less thread-unsafe.
    Verify that it's really an improvement.

28) Don't take default ports in case of invalid forwarding ports.

31) If a string action foo is disabled csp->action->string[ACTION_STRING_FOO]
    doesn't necessarily contain NULL, but may contain the string of an
    enabled foo action in an overruled section. Is it a bug? Does it matter?

32) In case of forwarding failures with socks port == 9050,
    show extra info about Tor (the whole FAQ entry?).

36) Unload unused action files directly, even if they are
    disabled without replacement.

38) In the final results, explicitly list disabled multi actions
    with their parameters. Not as trivial as it sounds.

40) When running in daemon mode, Privoxy's working directory is '/'
    which means it may not have permissions to dump core when necessary.
    Figure out a way to solve this. Introduce a cwd config option?

41) Change documentation framework to one that works cross-platform.
    Evaluate WML.

42) Add a DTrace USDT provider. Now that FreeBSD has userland DTrace
    support there's no longer any reason not to.

43) Write a tool to check URL patterns against URLs in the log file.
    This could be included in Privoxy-Regression-Test.

44) Privoxy-Log-Parser: Consider highlighting "Connection" in:
    23:13:03.506 283b6100 Header: Replaced: 'Connection: Keep-Alive' with 'Connection: close'

50) Investigate possible PCRS template speedup when searching
    macros with strstr() before compiling pcrs commands.
    Investigated, needs some restructuring but is probably worth it.

51) Make user-manual directive more generic to allow serving
    the FAQ and other stuff, too. Consider changing the port
    for "same origin policy" issues.

53) Find a more reliable hoster. Involves finding out what our
    requirements are and which SF alternatives fulfil them.
    It would probably also make sense to look into what other
    projects did when migrating away from SF.

    2014-05-13: Work in progress. Hosting wish list at the end
    of this file.

54) Move away from CVS to a more modern revision control system.
    Find out if there are any objection against going with Git.
    Using Git would also have the advantage that SF now pretends
    to support it, so we could do it independently from 53).

55) Apply for Coverity scans: http://scan.coverity.com/

56) Apply for the "free online access for qualified open-source
    software projects" for the Co-Advisor HTTP compliance tests:
    http://coad.measurement-factory.com/details.html#pricing
    (Application sent at 2014-05-12)

57) Allow piping into external programs to allow more powerful
    filters and policy decisions. Incomplete support available
    in Fabian's popen branch.

58) Move more template strings from the code into the actual templates.

59) Import the German template translation.

60) Ask the Russian translators for input on how to make their
    life easier.

61) Consider (optionally?) skipping the hostname comparison when
    checking if a connections that goes to a HTTP proxy can be reused.
    Do all HTTP proxy support that? Is it worth it?

63) Reject clearly too large requests earlier?

64) Use proper copyright attribution. "Privoxy Developers"
    is no legal entity.

65) Polish Website. Probably involves ditching the Docbook
    mess in favour of wml. There are already several threads
    in the mailinglist archives about this. See also #41.

66) Stop hard-coding the number of action and filter files.

67) Clean up source code directory layout. Depends on 54 so
    we don't lose the revision history.

68) Use standard make syntax so we don't depend on GNU make.

69) Update autoconf setup (or move away from it).

70) If the server connection is reset but the headers are
    received, consider passing the mess to the client instead
    of showing the connect-failed template. Relates to #2698674.

74) Let Privoxy-Regression-Test optionally check that action
    sections which disable actions actually are preceded by
    sections that enable said actions.

75) Create a tool that creates Privoxy action (and filter?) files
    out of adblock files. Could be implemented as option for
    url-pattern-translator.pl.

76) Cache DNS responses. Note that this has been requested
    several times by users, but is not a developer priority.
    If you care about this, feel free to submit patches.

77) Allow to configure the IP address used in outgoing connections.

78) Allow to optionally use pcre's DFA algorithm.

79) Evaluate pcre alternatives.

80) Change FEATURE_EXTENDED_HOST_PATTERNS to support both
    extended and vanilla host patterns at the same time.

    Note that the requirement is to allow the user to decide
    if the domain pattern should be interpreted as regex or
    traditional host pattern and if it's not obvious that the
    user made any decision, default to the latter.

    Possible solutions would be:

    1. An always-use-regex-domain-patterns config option
    2. An enable-regex-domain-patterns-for-this-action-file option
    3. An enable-regex-domain-patterns-for-this-action-file-until-the-user-says-otherwise option
    4. A treat-the-domain-pattern-in-this-line-as-regex(-or-not) option
    5. Combinations of the options above

    With 2+4, 3+4 or 2+3+4 being the preferences until
    further discussion.

82) Detect if the system time goes back in time let the user
    know if it caused any connections to get closed.

84) Flesh out the user-manual delivery to serve pages from
    other directories, too.

85) Once #84 is done, write a script that populates a directory with
    various common third-party icons (stumbleupon.png, facebook.png ...)
    and redirect requests for them to Privoxy.

86) Add a server-body-tagger action. This is trivial as as all the
    functionality required to do it already exists.

87) Add a client-body-tagger action. This is less trivial as we currently
    don't buffer client bodies. After 14) is implemented it would be
    trivial, though.

88) Investigate if there's a Perl module that Privoxy-Regression-Test
    could optionally use to keep connections alive, preferably while
    requiring less forks at the same time.

89) When multiple block actions apply, consider showing all the block
    reasons on the blocked page that haven't been overruled, not just
    the last one.

91) Add an optional limit for internal redirects. It would probably
    be reasonable to default to a limit of one and showing an error
    message if the request for the redirect URL would be redirected
    again.

92) The statistics currently aren't calculated correctly by Privoxy
    as each thread is only counted as one request which is no longer
    correct. This should be fixed, or the statistic code removed.
    Privoxy-Log-Parser's provides more detailed statistics, anyway.

93) Add a config directive to let Privoxy explicitly request either
    IPv4 (or IPv6) addresses, even if the system supports both.
    Could be useful as a workaround for misconfigured setups where the
    libc returns IPv6 addresses even if there's no IPv6 connectivity.

94) Add a config directive to let Privoxy prefer either IPv4 (or IPv6)
    addresses, instead of trusting the libc to return them in an order
    that makes sense. Like #93, this could be useful as a workaround
    for misconfigured setups.

95) Support a non-standard client header in CONNECT requests that
    contains the URL of the requested resource, which is then treated
    like the request URL.

    This way the client could opt-in for path-based blocking of https
    requests. Given that the headers from the CONNECT request aren't
    forwarded to the destination server, an unencrypted URL should be
    acceptable if the client and Privoxy are running on the same system
    or in a trusted environment.

96) Filters should be easier to look up. Currently get_filter() has to
    go through all filters and skip the filter types the caller isn't
    interested in.

98) When showing action section on the CGI pages, properly escape
    line breaks so they can be copy&pasted into action files without
    adjustments.

99) Figure out a mechanism through which a user can easily enable
    site-specific action sections that are too aggressive to be
    enabled by default. This could be similar to the presettings
    in default.action, but could also be just another action file
    that isn't used by default.

100) Create a cross-platform Privoxy control program and retire
     the win32 GUI. Integrate support for Privoxy-Regression-Test,
     Privoxy-Log-Parser, Privoxy-Filter-Test, uagen and similar tools.

102) Add an include directive to split the config file into several parts.

103) Potential performance improvement for large action files:
     when figuring out which actions apply, check the action bit mask
     before pattern matching and skip section that wouldn't modify the
     actions already set. To increase the impact the sections would have
     to be applied in reverse.

104) The code to modify global_toggle_state should be factored out into
     a separate function. Currently we mess with it in three different
     files, but only in w32log.c the tray icon is explicitly set.
     The logging is inconsistent as well. For details see #3525694.

105) Add support for socks authentication.

106) actionlist.h should be embedded in a way that causes less text
     segment bloat.

107) Support more pcrs variables, for example $destination-ip-address
     and $source-ip-address.

108) Allow to use a somewhat random string instead of PRIVOXY-FORCE.

109) Let log_error() support the format specifier %S which should
     work like %s but escape new lines like %N. This would be useful
     to log the result of header filters which may inject new lines.

110) Add a global-buffer-limit directive that roughly limits how
     much malloc'ed memory Privoxy will use and can potentially
     be smaller than (buffer-limit * max-client-connections).

111) Reject requests if hosts and ports in request line and Host
     header don't match (before filters have been applied).

112) If a header filter is used to inject another header by inserting
     a \r\n (undocumented feature), detect it and split the headers so
     following header actions do not treat them as a single string.
     Alternatively add another header injection mechanism.

113) Log statistics upon receiving a certain signal (SIGINFO or SIGUSR1).

114) Properly deal with status code 100. The current "Continue hack"
     can cause problems for gpg when uploading keys through Privoxy.

115) Add ICAP (RFC 3507) support. FR #3615158.

116) Due to the use of sscanf(), Privoxy currently will fail to properly
     parse chunks whose size can't be represented with 32 bit. This is
     unlikely to cause problems in the real world, but should eventually
     be fixed anyway. See also:
     https://bugzilla.mozilla.org/show_bug.cgi?id=959100

118) There should be "escaped" dynamic variables that are guaranteed
     not to break filters.

119) Evaluate using pcre's jit mode.

120) Add an option to limit pcre's recursion limit below the default.
     On some platforms the recursion limit doesn't prevent pcre from
     running out of stack space, causing the kernel to kill Privoxy
     ungracefully.

121) Add HTTP/2 support. As a first step, incomming HTTP/1.x requests
     should be translated to outgoing HTTP/2 requests where possible
     (and if desired by the user).

122) Allow customized log messages.

123) Evaluate if the voluntarily-disclose-session-keys option in Firefox
     (and other browsers) can be leveraged. Probably depends on #16.

124) Add support for the "lightweight OS capability and sandbox framework"
     Capsicum. http://www.cl.cam.ac.uk/research/security/capsicum/

125) Allow clients to HTTPS-encrypt the proxy connection.

##########################################################################

Hosting wish list (relevant for #53)

What we need:

- Bug tracker
- Mailinglists (Mailman with public archives preferred)
- Webspace (on a Unix-like OS that works with the webserver targets
  in GNUMakefile)
- Source code repositories (currently CVS, but migrating away
  from it is TODO #54 anyway and shouldn't be too much trouble)
- Commit mails (preferably with unified diffs)

(Unsorted) details to look at when evaluating hosters:

1. Preferably no third-party ads and trackers.
   External images, CSS and JavaScript may count as trackers
   but texts like "supported by company XYZ" may be acceptable.

2. JavaScript should be optional or not used at all.

3. Services we don't need shouldn't be enabled anyway.
   (We currently don't use Web forums, wikis, surveys etc.)

4. It would be preferable if the hoster didn't have a bad track
   record as far as user experience, security and privacy are
   concerned and if the terms of services are "reasonable" and
   haven't changed too often in the past. Updates in the past
   should have been improvements and not regressions.

5. It would be preferable if most of the server administration
   is done by a trusted third-party (or at least not a lot of work
   for us).

6. The server(s) should be located in a country with laws we can
   understand and follow (or at least not unintentionally violate).

7. A server location in a country with some kind of due process
   and strong data protection laws (at least on paper) would be
   preferable.

8. Given that Privoxy is a free software project it would be
   preferable if the hoster would use free software where possible.

9. Migrating away from the hoster in the future without losing
   any important data should be possible without writing web
   scrapers first.