<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
  <head>
    <meta content="text/html; charset=windows-1252"
      http-equiv="content-type">
    <title>limitations.html</title>
    <link rel="stylesheet" type="text/css" href="../styles.css">
    <style>
    u { font-weight: normal; text-decoration: none; }
    </style>
  </head>
  <body>
    <h3>Preprocessor library limitations</h3>
    <blockquote>
      <p>The Boost preprocessor library has always had certain specific
        limitations, which are specifically explained in other areas of
        this documentation and are implemented as object-like macros in
        the header file config/limits.hpp. These limitations largely
        detail the maximum number for certain features in the library.
        These maximums have been chosen in order to accommodate the
        large number of compilers and their preprocessors which the
        library supports, in order to provide a lowest common set of
        limits with which the vast majority of these compilers can work
        successfully. This reflects the situation that the C/C++
        standard rarely specifies preprocessor limitations for compilers
        but rather allows each compiler to choose whatever limitations
        it deems necessary in its preprocessor. While this allows
        individual compilers to largely choose what preprocessor
        limitations it deems necessary, it makes it more difficult for a
        preprocessor library like this one to work with a large number
        of preprocessors across a number of operating system platforms.
        The difficulty of this situation means that this Boost
        preprocessor library has chosen fairly conservative limitations.<br>
      </p>
      <p>In fact the limitation macros, although indeed existing, have
        never previously been used internally in the preprocessor
        library code. Instead while the limitation macros do indeed
        exist, the library has internally used fixed numbers equivalent
        to what these macros represent. This historical situation most
        probably reflects the fact that the library did not want the
        end-user's possibly changing the limitations, by redefining the
        macros in config/limits.hpp, to change how the library works and
        therefore cause problems in using the library.<br>
      </p>
      <p>This situation has now changed and a number of the limitation
        macros specified in config/limits.hpp are now used internally in
        the library code when the preprocessor is C++ standard
        conforming. This topic is therefore about how an end-user can
        change the limitations of the library.<br>
      </p>
    </blockquote>
    <h4>Type of limitations</h4>
    <blockquote>
      <p>The type of limitations in the Boost preprocessor library can
        be divided as:<br>
      </p>
      <ul>
        <li>Numbers</li>
        <li>Sizes</li>
        <li>Loops</li>
        <li>Iterations</li>
      </ul>
      The library has support for numbers. This supports basic
      arithmetic, comparisons, and logical operations. This support is
      documented by the macro BOOST_PP_LIMIT_MAG, where 'MAG' is short
      for "MAGNITUDE'. The current fixed maximum value is 256. This
      means that the library supports numerical operations in the range
      between 0 and 256.<br>
      <br>
      The library has support for the maximum sizes of higher level
      types. These types are arrays, lists, seqs, tuples, and variadic
      data. By sizes for these higher level constructs it is meant the
      maximum number of elements of which a higher level construct can
      consist. For arrays and tuples this value is documented by the
      macro BOOST_PP_LIMIT_TUPLE while for variadic data this value is
      documented by the macro BOOST_PP_LIMIT_VARIADIC. In both these
      cases the current fixed maximum value is 64. For seqs the value is
      documented by the macro BOOST_PP_LIMIT_SEQ, whose current fixed
      value is 256. For lists there has never been a limitation macro
      but in actual fact the limitation on the maximum number of list
      elements is defined by the maximum number given above for
      BOOST_PP_LIMIT_MAG. So a list, like a seq, has a default maximum
      of 256 elements. The reason that the maximum number of elements
      for arrays, tuples, and variadic data is 64 while for seqs and
      lists it is the much larger value of 256 is because the
      implementation of the former constructs often involves passing the
      likewise number as parameters to internal macros and a number of
      compilers have placed limitations on the number of parameters a
      macro can have; so the much safer 64 was chosen so as to be able
      to work with these compilers.<br>
      <br>
      The looping constructs consist of the control structure
      BOOST_PP_WHILE and the repetition constructs BOOST_PP_FOR and
      BOOST_PP_REPEAT. Each of these have a limitation macro, all
      currently set to 256. This means that the loop can run at maximum
      256 times. The corresponding limitation macros are
      BOOST_PP_LIMIT_WHILE, BOOST_PP_LIMIT_FOR, and
      BOOST_PP_LIMIT_REPEAT.<br>
      <br>
      Iteration constructs consist of local iteration and file iteration
      constructs.The specifics are more fully discussed in their own
      topics. The corresponding macro is BOOST_PP_LIMIT_ITERATION whose
      value is 256.<br>
    </blockquote>
    <h4> Limitation dependencies</h4>
    <blockquote>A number of the limitations are actually dependent on
      other limitations. These are:<br>
      <ul>
        <li>BOOST_PP_LIMIT_MAG depends on BOOST_PP_LIMIT_WHILE. All the
          arithmetic macros except for BOOST_PP_INC and BOOST_PP_DEC
          depend on looping using BOOST_PP_WHILE. Similarly all the
          comparison macros except for BOOST_PP_EQUAL and
          BOOST_PP_NOT_EQUAL depend on BOOST_PP_WHILE. So whatever is
          the value for BOOST_PP_LIMIT_MAG is also the value for
          BOOST_PP_LIMIT_WHILE.</li>
        <li>BOOST_PP_LIMIT_TUPLE depends on BOOST_PP_LIMIT_VARIADIC,
          since array/tuple functionality often ends up calling variadic data
          functionality. So while BOOST_PP_LIMIT_TUPLE can be smaller
          than BOOST_PP_LIMIT_VARIADIC it can never be larger.<br>
        </li>
        <li>BOOST_PP_LIMIT_SEQ depends on BOOST_PP_LIMIT_MAG since the
          total size of a seq must be at most the largest number. So
          while BOOST_PP_LIMIT_SEQ could be smaller than
          BOOST_PP_LIMIT_MAG it can never be larger.</li>
        <li>BOOST_PP_LIMIT_FOR, BOOST_PP_LIMIT_REPEAT, and
          BOOST_PP_LIMIT_ITERATION all depend on BOOST_PP_LIMIT_MAG.
          While any of these values could be smaller than
          BOOST_PP_LIMIT_MAG none of them can be larger.</li>
      </ul>
    </blockquote>
    <h4>Changing the limitations</h4>
    <blockquote>
      <p>Limitations can be changed by the end-user defining a
        limitation macro prior to including preprocessor library header
        files. This can be done either in a source file or as part of a
        compiler command line.<br>
      </p>
    </blockquote>
    <blockquote>
      <p>Before specifying which limitations can be changed a number of
        items need to be emphasized:<br>
      </p>
      <ul>
        <li>The limitations which can be changed can only be done for
          C++ standard conforming preprocessors. You can use the macro
          BOOST_PP_IS_STANDARD, invoking it as BOOST_PP_IS_STANDARD(),
          to determine if your compiler has a C++ standard conforming
          preprocessor. Almost all major compilers, including gcc and
          clang, have C++ standard conforming preprocessors. Most
          notably the default Visual C++ compiler is not a C++ standard
          conforming preprocessor, but the new preprocessor for Visual
          C++, available in Visual Studio 2019 and above, is a C++
          standard conforming preprocessor. If you change a limitation
          for a compiler that is not a C++ standard conforming
          preprocessor the change is simply ignored and no preprocessor
          error will occur. This is because any non-conforming
          preprocessor does not actually include or use the macros in
          the config/limits.hpp file.<br>
        </li>
        <li>WARNING ! WARNING ! WARNING ! Increasing limitations may
          mean that preprocessors may fail because the compiler itself
          can not handle the increase in the limitations. This may occur
          because the compiler has an internal limit on the number of
          parameters that can be passed to a macro, the number of nested
          macro invocations, or even the amount of memory needed for the
          preprocessor to expand a macro.</li>
        <li>Except for possibly first including the preprocessor header
          boost/preprocessor/config/config.hpp, changing any limitation
          must be done before including Boost preprocessor headers in a
          TU ( translation unit ) to work properly. Attempting to change
          a limitation after including Boost preprocessor headers will
          at worst lead to preprocessor errors and at best lead to UB (
          undefined behavior ).</li>
        <li>Limitations can only be changed to specific values as given
          below. Attempting to change limitations to other than the
          specific values will lead to the change being ignored or, if
          incorrectly done after including Boost preprocessor headers, a
          preprocessor error.</li>
      </ul>
      All the limitations whose default value is 256 can be changed to
      either 512 or 1024, except for BOOST_PP_LIMIT_WHILE which can not
      be changed.<br>
      The two limitations whose default value is 64 can be changed to
      128 or 256.<br>
      <br>
      For the individual limitations:<br>
      <ul>
        <li>BOOST_PP_LIMIT_MAG : When you change this limitation the
          value of BOOST_PP_LIMIT_WHILE also changes so that it is equal
          to BOOST_PP_LIMIT_MAG. This is because numerical computations
          depend on the BOOST_PP_WHILE loop, and therefore
          BOOST_PP_LIMIT_WHILE itself can not be changed. Also when you
          change the BOOST_PP_LIMIT_MAG limitation the value of
          BOOST_PP_LIMIT_SEQ, if it has not been also changed to be less
          than BOOST_PP_LIMIT_MAG, is also changed to the value of
          BOOST_PP_LIMIT_MAG. This is so that seqs, like lists, will
          normally have the same maximum number of elements, as defined
          by BOOST_PP_LIMIT_MAG, unless you decide otherwise and change
          the BOOST_PP_LIMIT_SEQ&nbsp; yourself.<br>
        </li>
        <li>BOOST_PP_LIMIT_TUPLE: When you change this limitation the
          value of BOOST_PP_LIMIT_VARIADIC also changes so that it is at
          least equal to BOOST_PP_LIMIT_TUPLE if the value of
          BOOST_PP_LIMIT_VARIADIC has not also been changed to be
          greater than BOOST_PP_LIMIT_TUPLE. This is because
          tuples/arrays depend on variadic functionality.</li>
        <li>BOOST_PP_LIMIT_SEQ, BOOST_PP_LIMIT_FOR,
          BOOST_PP_LIMIT_REPEAT, BOOST_PP_LIMIT_ITERATION: If you try to
          set any of these values greater than the BOOST_PP_LIMIT_MAG
          value, the particular limitation is set to equal the value of
          BOOST_PP_LIMIT_MAG. This is because all of these limits depend
          on the range of numerical values.</li>
      </ul>
    </blockquote>
    <h4>Other considerations</h4>
    <blockquote>
      <p>There is no necessity to change any limitation if you are happy
        with its default value. However if you decide to increase a
        limitation you should be aware that doing so can affect both the
        amount of memory used by the preprocessor and the speed in which
        preprocessing is accomplished. In the first case some
        preprocessors have been known to run out of memory if a
        limitation value is increased, even though this should really
        never happen in modern computer systems. In the latter case
        preprocessing might become so slow that trying to use the
        library with some increased limitation, and have your
        preprocessing finish in some reasonable amount of time, becomes
        impossible. The latter can occur when you are using lists or
        seqs and while cycling through a large number of elements you
        are also doing time consuming operations on each element value
        which generates further macro looping. Nearly all arithmetic and
        comparison operations involve further macro looping.<br>
      </p>
      <p>For lists and seqs there is functionality which uses the
        BOOST_PP_FOR macro. If the maximum size for lists and seqs is
        increased, by using BOOST_PP_LIMIT_MAG macro, you may also need
        to define BOOST_PP_LIMIT_FOR to increase the maximum number of
        'for' loops in order to cycle through all lists or seqs using
        this functionality.<br>
      </p>
      <p>The BOOST_PP_WHILE construct is used internally for looping by
        nearly all arithmetic and comparison macros as well as
        internally by a number of other macros in the library which loop
        through elements. While the construct can be called from within
        an already running BOOST_PP_WHILE loop, it is possible to run
        out of BOOST_PP_WHILE loops when this happens. In order to solve
        this problem you can specify a BOOST_PP_LIMIT_MAG which is
        larger than the maximum number you will ordinarily use. This
        will give you twice the number of BOOST_PP_WHILE loops and will
        keep you from running out of loops if you have to do arithmetic
        computations while cycling through lists and seqs and the number
        of lists and/or seqs you cycle through is anywhere near your
        original&nbsp; maximum.<br>
      </p>
    </blockquote>
    <h4>Testing limitations</h4>
    <blockquote>
      <p>The default testing of the preprocessor library, using Boost
        Build's b2 command in the preprocessor test directory, only
        tests the default limitations. If you want to test, within the
        test directory, any of the non-default limitations which may be
        set, you can do so by invoking the b2 command with any of these
        explicit targets:<br>
      </p>
      <ul>
        <li>preprocessor_128 : arrays, tuples, and variadics with up to
          128 elements</li>
        <li>preprocessor_256 : arrays, tuples, and variadics with up to
          256 elements</li>
        <li>preprocessor_512 : numbers, lists, seqs, looping, and
          iteration constructs with 512 maximum</li>
        <li>preprocessor_1024 : numbers, lists, seqs, looping, and
          iteration constructs with 1024 maximum</li>
        <li>preprocessor_tup : both the first and second items in this
          list</li>
        <li>preprocessor_num : both the third and fourth items in this
          list</li>
        <li>preprocessor_limits : all preprocessor limitations</li>
      </ul>
      If you invoke b2 with the last preprocessor_limits target your
      testing may take awhile for your compiler. When you choose one of
      the above targets, as in <code>'b2 preprocessor_limits'</code>,
      the default tests are not run. If you want the default tests run,
      along with any of the targets such as preprocessor_limits, your
      command would be <code>'b2 . preprocessor_limits'</code>, where
      the '.' notation means all the non-explicit ( aka default )
      targets.<br>
    </blockquote>
    <blockquote> </blockquote>
    <blockquote>
      <ul>
      </ul>
    </blockquote>
    <blockquote> </blockquote>
    <b>See</b> <b>Also</b><br>
    <ul>
      <li><a href="../ref/is_standard.html">BOOST_PP_IS_STANDARD</a><br>
      </li>
      <li><a href="../ref/limit_mag.html">BOOST_PP_LIMIT_MAG</a></li>
      <li><a href="../ref/limit_tuple.html">BOOST_PP_LIMIT_TUPLE</a></li>
      <li><a href="../ref/limit_variadic.html">BOOST_PP_LIMIT_VARIADIC</a></li>
      <li><a href="../ref/limit_seq.html">BOOST_PP_LIMIT_SEQ</a></li>
      <li><a href="../ref/limit_while.html">BOOST_PP_LIMIT_WHILE</a></li>
      <li><a href="../ref/limit_for.html">BOOST_PP_LIMIT_FOR</a></li>
      <li><a href="../ref/limit_repeat.html">BOOST_PP_LIMIT_REPEAT</a></li>
      <li><a href="../ref/limit_iteration.html">BOOST_PP_LIMIT_ITERATION</a><br>
      </li>
    </ul>
    <hr size="1">
    <div style="margin-left: 0px;"> <i>� Copyright Edward Diener 2020</i>
    </div>
    <div style="margin-left: 0px;">
      <p><small>Distributed under the Boost Software License, Version
          1.0. (See accompanying file <a
            href="../../../../LICENSE_1_0.txt">LICENSE_1_0.txt</a> or
          copy at <a href="http://www.boost.org/LICENSE_1_0.txt">www.boost.org/LICENSE_1_0.txt</a>)</small></p>
    </div>
  </body>
</html>