bump version to 12.2.2-pve1

[ceph.git] / ceph / src / boost / libs / test / doc / testing_tools / boost_test_technical_details.qbk

[/
 / Copyright (c) 2015 Boost.Test contributors 
 /
 / Distributed under the Boost Software License, Version 1.0. (See accompanying
 / file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
 /]


[section:internal_details `BOOST_TEST`: details on expressions]
Let's consider the following example:

[bt_example boost_test_macro3..BOOST_TEST reporting..run-fail]

It was already mentioned that the reporting is not symmetrical (see [link boost_test.testing_tools.reports here]).
An expression is constructed from the `statement` appearing in the `BOOST_TEST` macro. This expression allows evaluation and reporting such
as `"13 - 1 >= 12" failed` along with a copy of the `statement`, which contains more details than `"a - 1 < b" failed`.
In details, what happens is the following:

# a special object, the `seed` of the expression, is composed from the left side of `statement`.
  This initial composition has highest precedence over the supported operations. The expression below:

    a op1 b op2 c op3 d

  is actually seen as

    ( seed a ) op1 b op2 c op3 d

# The "`seed a`" returns an `expression` object that keep tracks of the type of `a`. This expression
  has overloads for left-to-right associativity, and the
  operations `op1`, `op2` ... are /chained/ to the right of this expression object:

    a op1 b

  yields to the pseudo-code

    expression1 = create-expression(a)
    expression2 = create-expression(expression1, op1, b)

  `expression1` and `expression2` keep track of their left and right operands, and the operation on those operands. The
  expressions keep also track of the result type of the associated sub-expression. In the above example, `expression1` and `expression2`
  have result type `decltype(a)` and `decltype(a op1 b)` respectively. The result type allows for chaining
  sub-expressions.

# The C++ operators precedence rules apply in any case. What is seen by the expression is what is reachable with left-to-right
  composition. Any other operation that happens before it reaches the expression's right operand is not parsed as a sub-expression
  and is seen as a single operand: the right operand is not developed further by the framework.
  Let's suppose `op2` below has higher precedence than `op1`, then

    a op1 b op2 c

  is equivalent to:

    create-expression(create-expression(a), op1, (b op2 c))

  In the above statement, the final expression can only see the result of `(b op2 c)` to its right, for which no further detail
  can be provided in the logs. This is also the case for /right-to-left/ associative operators, such as `!`, `~`, `-` (unary negation)
  etc.

  [caution Since the `expression` object is composed from left-to-right, it actually observes a chain of operations and
    not the full expression tree.]

# Once the full expression chain is built, it is evaluated as a chain of sub-expressions from left-to-right, exactly as the
  composition rule above. The evaluated elements are the ones of the expression itself. The expression

    a op1 b

  yields to the following evaluation chain:

    expression2.result = expression1.result op1 b
    expression1.result = a

  The final expression of the statement is cast to a boolean, which is in turn evaluated by the __UTF__.


The example below illustrates the construction of the left-to-right /chained/ expression.

[bt_example boost_test_macro2..BOOST_TEST compound statements..run-fail]
[endsect]