NEWS

                    User-Visible C TAP Harness Changes

C TAP Harness 3.1 (2014-07-02)

    ok, okv, and all is_* functions now return true if the test succeeds
    and false if it fails, matching the return status of the corresponding
    Perl Test::More functions.  This allows more succinct code when the
    actions of a test program should vary based on the success or failure
    of previous tests.  Based on a patch by Peter Pöschl.

    diag and sysdiag now always return 1, making it easier to insert calls
    into compound statements when debugging tests.  Based on a patch by
    Peter Pöschl.

    Add new breallocarray API that does the same as realloc but takes
    calloc-style arguments to specify the size and checks internally for
    integer overflow.  Inspired by the OpenBSD reallocarray function.

    Check for integer overflows on memory allocation.  All the possible
    issues for this code are rather theoretical, but one may as well
    strive for correctness.

    Replace all uses of sprintf with a simpler string concatenation
    function that checks for allocation overflow.  (The standards
    assumptions for this package don't permit assuming asprintf or a
    sufficiently non-broken snprintf to simulate asprintf.)

C TAP Harness 3.0 (2014-01-28)

    The test_cleanup_register API has changed in this release.  Cleanup
    functions must now take two parameters, not one, and are called from
    all test processes, not just the primary one.  The new second argument
    indicates whether the cleanup function was called in the primary
    process (the one in which plan or plan_lazy was called).  External
    resources, such as files, should generally only be freed when the
    cleanup function is called in the primary process, but tests may want
    to free internal resources, like memory, in all processes to ease
    analysis with tools like valgrind.

    When running test programs from a list, reopen standard input for each
    program to /dev/null, and be more careful about closing all duplicates
    of file descriptors left behind after setting up standard output and
    standard error so that extraneous file descriptors aren't leaked to
    the child process.

C TAP Harness 2.4 (2013-12-25)

    Add new diag_file_add and diag_file_remove APIs to the basic C TAP
    library.  These functions manage a list of registered file that
    contains supplemental diagnostic information.  Each registered file is
    checked before each output function for any new lines, and any lines
    are displayed as if they'd been passed to diag().  This can be useful
    if, for example, the test involves a background daemon whose output
    can be logged to a disk file.

C TAP Harness 2.3 (2013-11-13)

    Add new test_cleanup_register API to the basic C TAP library.  This
    registers a C callback function that's called during exit from the
    test and passed in a boolean argument indicating whether the test
    succeeded or failed.

    Suppress lazy plans and the summary of tests at the end of a test
    program if the program aborted with bail or sysbail.

    Add warn_unused_result gcc attributes to the C TAP library functions
    where ignoring the return value is almost certainly a bug (such as all
    the malloc functions).

    Add portability for warn_unsed_result attributes to tap/macros.h.

C TAP Harness 2.2 (2013-08-14)

    bail and sysbail now exit with status 255 to match the behavior of
    BAIL_OUT in Perl's Test::More.

    Document that runtests handling of test cases that bail out differs
    from the documented behavior of BAIL_OUT in Perl's Test::More and the
    behavior of prove, and document why.

C TAP Harness 2.1 (2013-03-15)

    When locating test programs, try a suffix (-t, .t, or no suffix) with
    all bases before moving on to the next suffix.  The behavior in the
    previous release was not backward-compatible: it would find the
    unsuffixed helper program in the build directory instead of the actual
    test in the source directory for some rra-c-util tests when the build
    directory and the source directory weren't the same.

    Document that TAP version directives in the TAP output are ignored.

C TAP Harness 2.0 (2013-03-14)

    The default behavior of tests/runtests has changed to make it act more
    like other test drivers.  Now, to provide a file containing a list of
    tests, use the -l option.  Existing users should add -l to the command
    line in front of the test list.  Otherwise, the command-line arguments
    are taken as tests to run, as if they were all listed in a test list
    file.

    runtests now tries the test name without any extension if the test
    name with either -t or .t was not found.  It also does not consider a
    directory to be a valid test when searching for the executable, even
    if the directory is executable.

    Flush the harness output after each test.  This effectively implements
    line buffering even when standard output is not a terminal and
    therefore output isn't flushed after each line of test output.

    When displaying partial status of tests with a deferred plan, show the
    total number of tests as ? rather than the number of tests seen so
    far.  This matches the behavior of Perl's prove utility.

    More correctly handle backspacing over the test count in several abort
    cases when reporting status to a terminal.

    Add GCC annotations to some internal functions in runtests to help
    clang --analyze better understand code correctness.  Remove a dead
    store caught by clang --analyze.

C TAP Harness 1.12 (2012-05-11)

    Fix additional uses of local in the shell TAP library for portability
    to Solaris /bin/sh and document a bad interaction between backquotes
    and double quotes for the puts, test_file_path, and test_tmpdir
    functions.

    In the C TAP library, if bail is called after plan_lazy but before any
    tests are run, don't output the plan or the summary of test results.

    Add the UNUSED macro (to mark possibly-usused variables for gcc) to
    tests/tap/macros.h for use in the C TAP library.

    Make the __attribute__ handling in tests/tap/macros.h more aware of
    different versions and compilers.  All attributes are now suppressed
    for GCC older than 2.96 (instead of 2.7), since __malloc__ became
    available then.  __alloc_size__ is suppressed for GCC older than 4.3.
    Warnings about unknown diagnostics are suppressed for LLVM and Clang,
    which pretend to be GCC but don't support all the same attributes.

C TAP Harness 1.11 (2012-04-25)

    Reorganize the source package so that runtests.c is in the tests
    directory and the C and shell TAP libraries are in tests/tap.  Change
    the include directives in the C TAP library source files accordingly.
    This is somewhat confusing for the standalone package, but means that
    the files are located in the same locations here as they would be in
    packages into which they're copied, which simplifies instructions and
    the build system.  This also removes the need to build the C TAP
    library with -I flags at the cost of requiring that it be located in a
    tests subdirectory, allowing easier use of additional supporting
    headers without requiring that all tests be built with a -I directive.

    Do not attempt to use feature-test macros unless compiled with gcc's
    -ansi flag or PEDANTIC is defined.  Defining _XOPEN_SOURCE 600 on
    Solaris 10, which is required for isinf() if feature-test macros are
    in use, causes compilation to fail unless the compiler is run in C99
    mode due to an ABI change.  The basic TAP library doesn't care about
    this distinction and doesn't want to force specific compiler flags, so
    avoiding feature-test macros is a better fix.  Thanks to Bob
    Friesenhahn for the bug report.

    Do not use local in the shell TAP library.  Solaris /bin/sh doesn't
    support it, and the shell library is otherwise portable to that shell.
    Instead, use a prefix of tap_ on all non-public variables.

    Move the __attribute__ and BEGIN_DECLS/END_DECLS macros from
    tests/tap/basic.h to a new tests/tap/macros.h header file so that they
    can be more easily reused by any other TAP add-on.  Be sure to copy
    tests/tap/macros.h into your source package and add it to the source
    files for the libtap library when upgrading.

    Move the is_double C TAP test function into a separate source file.
    Including this function may require linking with libm on some
    platforms, which is undesireable if the package otherwise doesn't use
    math functions.  The new tests/tap/float.c and tests/tap/float.h files
    need only be included in a package that wants to do floating point
    tests.  Users of is_double will now need to include tests/tap/float.h
    in addition to tests/tap/basic.h (and link all clients of the libtap
    library with -lm).

C TAP Harness 1.10 (2012-02-29)

    Add bstrndup to the basic C TAP library.  This is akin to strndup
    (copying at most n characters of the string), but calls sysbail if
    memory allocation fails.

C TAP Harness 1.9 (2011-12-24)

    Add test_tmpdir and test_tmpdir_free functions to the basic C TAP
    library and a test_tmpdir function to the shell TAP library.  These
    functions create a temporary directory under the build tests directory
    for tests to use for transient files.

    runtests now frees all allocated resources on exit so that it can be
    used to run test cases with valgrind memory allocation checking
    without adding false positives.

C TAP Harness 1.8 (2011-09-02)

    Add bmalloc, bcalloc, brealloc, and bstrdup functions to the basic C
    TAP library.  These functions call sysbail if memory allocation fails.

    If BUILD were not set when building runtests and the -b option was not
    present, SOURCE and the -s option would be ignored when searching for
    tests.  Fix runtests to still honor SOURCE and -s when searching for
    tests even in that case.

C TAP Harness 1.7 (2011-04-28)

    Add a more complete usage message to runtests and add support for a -h
    command-line flag to display the usage message.

C TAP Harness 1.6 (2010-12-29)

    is_double() now takes a third argument, an epsilon.  Two numbers are
    considered equal if their absolute difference is less than epsilon.
    is_double() also now treats wanted and seen values of NaN (not a
    number) as equal.  Thanks to PICCA Frédéric-Emmanuel for the proposed
    changes.

    The ok_program function in the shell libtap.sh library no longer
    strips text after a colon and a space from the program output if the
    expected status is non-zero.  Instead, if program output may contain
    system-specific error messages after a colon and a space, put the new
    function strip_colon_error before the program to do this stripping.
    Thanks to Carsten Hey for the idea.

    strip_colon_error is now smarter about preserving an initial word
    ending in a colon (which is generally the program name) while still
    stripping error messages later in the line.

    The test_file_path function in the shell libtap.sh library now always
    returns the empty string, rather than possible absolute paths starting
    at /, if $BUILD and $SOURCE are not set.

    Flush standard error in the C TAP library before printing results for
    more deterministic output.  Thanks to Carsten Hey for the idea.

    All of C TAP Harness now compiles with gcc -ansi -pedantic and should
    be fully C89-compatible.  Note that either C99 or SUSv3 is required to
    build C TAP Harness.  (This should not be a problem on any modern
    platform.)  Based on work by Carsten Hey.

    Simplify and improve output formatting in the summary of failing tests
    in some edge cases.

    Add explicit license statements to the files meant to be copied into
    other packages rather than referring to LICENSE.

C TAP Harness 1.5 (2010-08-25)

    In ok_program in the shell TAP library, strip error output after the
    second colon, not after the first.  This code helps comparing error
    output that includes system error messages, but needs to skip past the
    first colon, since it separates the program name from the error
    message.

    runtests.c now builds without warnings with gcc -Wswitch-enum.

C TAP Harness 1.4 (2010-07-07)

    Add a test_file_path() function to the basic C and shell TAP
    libraries, which searches the build and source directories for a
    particular file and returns the full path.  This is a utility function
    that can be used to find test data files.

C TAP Harness 1.3 (2010-06-09)

    Add API documentation for the functions provided by the basic TAP C
    library.  The documentation is in POD and *roff format in docs/api and
    installed as man pages in the (generally unused) case of make install.

    Add docs/writing-tests, which provides some basic instructions for how
    to write TAP tests in Perl, C, and shell using the facilities provided
    by C TAP Harness.  This document is suitable for inclusion in packages
    that use C TAP Harness as instructions for contributors.

    Add an okv() function to the C TAP library.  This is the same as ok()
    but takes the test description as a va_list to make it easier to reuse
    the reporting infrastructure in other tests.

C TAP Harness 1.2 (2010-04-13)

    Support giving the plan at any point in the test output rather than
    requiring it be first.  The TAP protocol permits giving the plan last
    instead of first, but since it's just as easy to support providing the
    plan in the middle of the test output, allow that as well.

    Add support for lazy planning to the basic TAP library.  Instead of
    calling plan at the start of the test, call plan_lazy instead, which
    doesn't require specifying a test count.  Then, the plan will
    automatically be printed based on the last test number at the end of
    the program.  This is not the recommended approach for writing tests
    since, should the test program exit in the middle for some reason,
    everything will appear to succeed.  But it can save time in writing
    tests quickly and allows for dynamic numbers of tests.

    Add diag and sysdiag functions to the basic TAP library, which output
    diagnostics to standard error preceded by the protocol-required #
    character.

    Clean up data types in the basic C TAP library.  Test numbers and
    counts are now unsigned long, and ok_int compares longs.

    Add the GCC nonnull attribute to the bail functions in the TAP
    library.

C TAP Harness 1.1 (2009-08-15)

    Set an exit handler in libtap that prints a summary of the test
    results at the end of a test script, similar to Perl's Test::More.

    When looking for 0..1 # skip plan lines indicating the whole test case
    should be skipped, correctly compare "skip" case-insensitively.  This
    fixes interoperability with Perl's Test::More.

    Consume all output from the test case before closing its file
    descriptor, even if the case is aborted due to bailing out or sending
    an invalid test number.  This prevents tests from spuriously dying
    with SIGPIPE.

    Enable Automake silent rules.  For a quieter build, pass the
    --enable-silent-rules option to configure or build with make V=0.

C TAP Harness 1.0 (2009-05-21)

    Initial public release based on the version of runtests.c and the
    libtest library that shipped with remctl 2.13.  The changes below are
    relative to that version.

    Set the SOURCE and BUILD environment variables when running tests to
    the roots of the source and build test directories if set at compile
    time or via the -s and -b command-line options to runtests.

    Search for test programs first in the current directory and then in
    the build and source directories, in that order, as specified by
    SOURCE and BUILD or by the -s and -b options.  This allows some tests
    to be compiled and some shipped with the package and merged together
    at run time by runtests, even if the build directory is not the source
    directory.

    Support running a single test program with the -o command-line option.
    This is mostly interesting to take advantage of the environment
    variables and test searching algorithm described above.

    Add support for 0..1 # skip <reason> plan lines, indicating that the
    entire test case should be skipped.

    Add support for aborting a test case with "Bail out!" and reporting
    the reason given on that line.

    Add support for omitting the test number from an "ok" or "not ok"
    line, in which case the number is assumed to be one more than the
    previous number.

    Add support for todo tests, which are recorded as skipped if failing
    and failed if they succeed unexpectedly.

    Omitting the space after "ok" or "not ok" line is now supported, and a
    test number of 0 now results in an error.

    Recognize skip directives regardless of case.

    Change the reporting of reasons to abort the test case to use ABORTED
    and the abort reason in parentheses for more consistent output.

    Rename test_init() to plan() and ok_int(), ok_double(), and
    ok_string() to is_int(), is_double(), and is_string() in the C TAP
    library to more closely match the function names in Perl's test
    modules.

    Add is_hex() C TAP library function to compare two unsigned longs and
    print mismatches in hex.

    Change all C library functions to take a test description
    (printf-style) and to no longer take the test number.  Instead, keep a
    static count of the current test number.

    Add bail() and sysbail() C TAP library functions to bail out with the
    right magic output string.

    Add skip_all() C TAP library function to skip all tests and exit.

    Stop exporting the printcount interface in the shell version of the
    TAP library.  Instead, provide plan, skip_all, ok, skip, ok_block,
    skip_block, ok_program, and bail shell functions, similar to the C
    library interface, with the ok* functions taking a test description as
    the first argument.

    Show mismatch output from the is_*() functions before the "not ok"
    line instead of after it.

    Add the initial "1.." to plan() output for better TAP compliance.

    Support either "-t" or ".t" as the ending of test cases.

    Remove the dependency on both config.h and the general portability
    framework and instead include the necessary C headers directly.  This
    requires ANSI C.

    Remove the functions from libtap that assume libutil from rra-c-util.
    Those library functions will be provided by that package.