From: Gennadiy Rozental (rogeeff_at_[hidden])
Date: 2007-10-26 03:22:54
Last couple weeks I spent incorporating new features into Boost.Test trunk.
Some of these were sitting on my local disk for quite some time, but it's
finally there. There are still number of features/request in pipeline to be
addressed, but I plan to concentrate primarily on docs update now. BTW they
are progressing and in much better shape now. I might need to work to
include all the fixes required to build it into boostbook. I plan to put
beta on my site soonish. Following are more or less detailed release notes
for all the changes. Any comments are welcome as usual.
I. New debug interfaces
In this update I made an effort to introduce portable interface for
communication with debugger and other debug facilities. The interface is
defined as set of free functions along with some helper structures. I left
out couple functions related to the portable stack dump/stack walk
interface, since it's not completely ready yet and I am do not have enough
time at the moment to finish it.
New interface consists of the following functions:
Detects if program is running under debugger
Causes program to break execution in debugger at call point
bool attach_debugger( bool break_or_continue = true );
Attaches debugger to the current process
void detect_memory_leaks( bool on_off );
Switches on/off memory leaks detection
void break_memory_alloc( long mem_alloc_order_num );
Causes program to break execution in debugger at specific allocation point
I've made an effort to implement these interface for number of available
configurations/debuggers for platforms I had an access to:NT, Linux,
Solaris. There are still a lot of work to make these interfaces available on
other systems. These is also interfaces to add support for additional
debuggers, setup default debuggers and configure some parameters of existing
>From usability standpoint there is new command line argument available for
UTF users: --auto_start_dbg=<yes|no|debugger_name>
Default value is no. If set to yes, in case of system error (sigsegv for
example) the program will try to attach default debugger defined for the
toolset and stop at fault location. If set to the different value this value
is treated as debugger id (one of the predefined set of supported debuggers)
and Boost.Test will try to use this debugger. For
example --auto_start_dbg=dbx-ddd causes ddd with dbx to be used,
while --auto_start_dbg=gdb-xterm will open new xterm window and attach gdb.
Some of the functions are implemented only for specific platforms for now.
But the main goal of this exercise is to define portable interfaces.
II. Execution monitor rework
The execution monitor implementation is rewritten almost completely. Some
new features introduces, some old done differently.
1. windows SEH handling
Windows SEH handling changed completely. Instead of seh translator execution
monitor relies on custom exception filters and double set of __try/__catch
constructs. There some additional tricks in place to make it work more
2. Signal handling on *nix
Signal handling is made more consistent and covers more signals. An
alternative stack is used by default signal handler by default, while there
is an option to avoid it (--use_alt_stack=no for UTF users)
3. New portable debug interfaces used in place of platform specific calls.
4. Error reporting is enhanced both on windows and *nix to provide more
information about error cause and location. For example these lines
int* p = (int*)0x01;
BOOST_CHECK( *p == 0 );
will now produce
"fatal error in "free_test_function": memory access violation occurred at
address 0x00000001, while attempting to read inaccessible data"
"fatal error in "free_test_function": memory access violation"
Similar enhancements made for different types of system errors.
5. Handling for invalid parameter error produced by VC 8.0 runtime
6. Floating point exceptions detection is introduced.
For windows based configurations you can tell the execution monitor to
detect floating point errors like underflow,overflow,divide by zero and
others. It's disabled by default. UTF users can enable it
III. Significant framework updates
1. run by name is finally there
UTF users can mow specify test unit(s) to run using command line argument.
<test-module> --run_test=<test to run>
Test to run specifies which test cases/test suites to execute. It specified
in a form a/b/c, where a,b and c are filters for corresponding levels of
test tree. '*' can be used at the beginning, at the end and as the level
filter itself. ',' is used to create lists. Here couple concrete examples:
--run_test=test1 - runs test case or test suite test1 that resides in master
--run_test=test1,test2 - runs test case or test suite test1 and test case or
test suite test2 that resides in master test suite.
--run_test=s1/test1 - runs test case or test suite test1 that resides in
test suite s1 that resides in master test suite.
--run_test=s1*/test1 - runs test case or test suite test1 that resides in
test suites with name stating with s1 and that resides in master test suite.
--run_test=*s1/test1 - runs test case or test suite test1 that resides in
test suites with name ending with s1 and that resides in master test
--run_test=*/test1* - runs any test unit with name starting with test1 that
resides in any test suite that resides in master test
--run_test=* - runs all test units
--run_test=a/* is equivalent to --runtest=a
2. New tools in toolbox.
Responding to the several requests I've implemented 5 new tools:
Floating point tools update unfortunately did not make a cut.
3. Test suite auto registration modified to allow c++ namespace like
It's now possible to restart the test suites and defines the same test
suites in different test file, effectively making test suites similar to the
C++ namespaces. for example
construct a test tree with single test suite s1 containing 2 test cases t1
4. Support for expected failures in test cases with automated registration
is reworked completely. It now allows to be used within auto-test-suites.
5. unit_test::framework API changed to return non const references to the
test units to allow post creation modifications. It now possible to modify
test tree at runtime. In addition test_suite::remove interface is introduced
to allow remove test units from the test suite if necessary.
6. init_unit_test_suite function invocation is now guarded, so any
exceptions produced by it doesn't cause UTF to report internal failure but
show correct cause of setup failure.
VI. Minor modifications and fixed
* Use portable message mechanism in execution monitor
* Interface to access results reporter stream provided
* Bug in output_test_stream constructor error generation fixed
* Test start and finish messages disabled if log level is set to nothing
* Test module initialization error message is redirected into result
* New message is added to report no assertion occurred in a test case
* Framework misuse guard is introduced to prevent usage of testing tools
before framework is initialized (for example from within init function)
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk