plan.html

<?xml version="1.0" encoding="utf-8"?>
<!DOCTYPE html 
     PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN"
     "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd"> 
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
    <link rel="stylesheet" title="Default"
      type="text/css" href="todo.css" media="all"/>
    <meta http-equiv="Content-Type"
      content="text/html; charset=utf-8"/>
<title>TODO for coming releases</title>
</head>
<body>
  <h1> Nettle release plans </h1>
  <p> This is an attempt at defining a development target for
  Nettle-2.7, inspired by similar pages for recent GMP releases. [Last
  updated 2013-04-02]</p>
  <p class='should'>
    This really ought to be done before release
  </p>
  <p class='wish'>
    Try to get this done before release
  </p>
  <p class='done'>
    Done!
  </p>
  <p class='postponed'>
    Leave for some later release!
  </p>

  <h1> Plan for the Nettle-2.7 release </h1>
  <h2> Code </h2>
  <p class='should'>
    Support older GMP versions, as far as practical.
  </p>
  <p class='should'>
    Make use of GMP's mpn_cnd_add_n and mpn_sub_n (which will be
    available in the next GMP release).
  </p>
  <p class='should'>
    Rewrite x86_64/sha3-permute.asm. Moves between xmm registers and
    regular registers kill performance at least on some AMD
    processors.
  </p>
  <p class='wish'>
    Include UMAC. Needs a serious effort to produce test vectors.
  </p>
  
  <h2> Documentation </h2>
  <p class='should'>
    Document new ECC functions.
  </p>
  <p class='should'>
    Document salsa20r12.
  </p>
  <p class='should'>
    Update NEWS file.
  </p>

  <h2> Build system </h2>
  <p class='done'>
    Ensure that <tt>make install</tt> without a preceding
    <tt>make</tt> works.
  </p>
  <p class='wish'>
    Make the assembly ALIGN-macro non-logarithmic, reusing the m4 log2
    macro in GMP. Current macros have the problem that in m4 eval,
    <tt>&lt;&lt;</tt> is not supported by Solaris, while <tt>**</tt>
    is not supported by OpenBSD.
  </p>

  <h2> Testing </h2>
  <p> Since xenofarm isn't up and running, do some manual testing:
  </p>
  <ul>
    <li class='should'> x86_64-linux-gnu</li>
    <li class='should'> x86_64-freebsd</li>
    <li class='should'> x86-linux-gnu</li>
    <li class='should'> x86-freebsd</li>
    <li class='should'> x86-w*ndows (using cross compiler and wine)</li>
    <li class='wish'> x86_64-w*ndows (seems a bit harder since 64-bit
    wine is not yet available in debian)</li>
    <li class='should'> x86-darwin (needs help from Nettle users)</li>
    <li class='should'> x86_64-darwin (needs help from Nettle users)</li>
    <li class='should'> armv7-linux-gnu (pandaboard test platform)</li>
    <li class='wish'> armv7-android (possible test platform: N10 tablet)</li>
    <li class='should'> sparc32-solaris10</li>
    <li class='should'> sparc64-solaris10</li>
  </ul>

  <h1> Plans for nettle-3.0 </h1>

  <h2> Interface changes </h2>
  <p class='should'>
    Change the type of all lengths from <tt>unsigned</tt>
    to <tt>size_t</tt>. An ABI change on most 64-bit platforms.
  </p>
  <p class='should'>
    Change argument type of memxor and memxor3 from uint8_t * to void
    * (like modern memcpy). Consider them unconditionally part of the
    nettle library, with nettle_ prefix on the symbols, and no
    AC_REPLACE_FUNCS.
  </p>
  <p class='should'>
    For Merkle-Damgaard hash functions, separate the state and the
    buffering. E.g., when using them for HMAC keyed "inner" and
    "outer" states, we now get three buffers but we only need one.
  </p>
  <p class='should'>
    Use the nettle_cipher abstraction only for block ciphers (in
    particular, exclude arcfour). Use a const for the ctx argument to
    nettle_crypt_func.
  </p>
  <p class='wish'>
    Consider making a public interface similar to nettle_aead? With
    the above change, it can't use nettle_crypt_func.
  </p>
  <p class='wish'>
    Make it possible to build nettle and hogweed using mini-gmp.
  </p>
  <p class='should'>
    Reorganize private key operations. Need to support RSA with and
    without blinding, and DSA according to spec and some deterministic
    variant (like putty), and possibly also smartcard versions where
    the private key is not available to the library. And without an
    explosion of the number of functions.
  </p>
</body>
</html>