I agree that the purpose of --verbose is to provide additional
informational detail about what's being tested, not to increase
the strictness of the test. Adding a new option to enable
expensive memory checks sounds like the right approach to me.
Martin
Travis Vitek wrote:
I managed to prematurely send that last message. Here is the completed response
Farid Zaripov wrote:
From: Travis Vitek [mailto:[EMAIL PROTECTED]
To: dev@stdcxx.apache.org
Subject: RE: svn commit: r648128 - /stdcxx/trunk/tests/src/driver.cpp
* tests/src/driver.cpp [_MSC_VER] (_rw_opt_verbose):
Turn on checking the memory at every allocation and
deallocation in verbose mode.
Perhaps this should be put into its own routine option. I
believe that this memory checking can slow things down quite
a bit, and I'd like to avoid that overhead when doing verbose
checking.
And how often you're doing verbose checking? :)
I actually use it occasionally to verify that any rw_assert() calls I add will
display correctly in verbose mode, but that isn't really the point.
I see that
the only difference between the verbose and non-verbose modes
is printing the description (diag_msgs[].desc) in diag messages.
The definition of verbose is 'to use more words than are necessary'. Typically
verbose mode for a program enables additional output but it doesn't change the
behavior of the program in other ways.
So I thought that --verbose option is used very rarely at this
time and the name of this option is corresponding to added
actions - verbose checking the using of the heap memory :)
Just because we don't use the option frequently isn't a good reason misuse it.
At home we don't use our fine china dinnerware more than once a year, but I
wouldn't dare use the gravy boat as an oil pan.
(_rw_opt_compat): Disable MSVC debug popup's in compat mode.
According to some conversation we had with Martin last week,
this should have been enabled all the time. If this is the case,
maybe this code should be moved to rw_vtest() to make that happen?
Why? Personally I found the debug MSVC popups very useful in
debugging.You can't miss the debug message because you have to
press some button in message box to close it.
Oh, I agree. They make it difficult to not notice a failure, especially when
running the tests from the command line.
And you need press the only one button ("Retry") to launch and
attach the debugger and see the source file on failing line with
call stack. If they are will be disabled all the time, you will
need scan the program output for the debug messages and then
wast some time to find the failing line in source code...
Well, now I have two more problems with this change. First this change disables
the window for CRT errors only. It should probably do the same for the gpfault
error box via a call to SetErrorMode(). Second, it sends the messages only to
the debugger output window. I think we should be sending them stderr also so
that we can see the messages in the output spew. Otherwise it would be _very_
easy to overlook these types of errors.
_CrtSetReportMode( _CRT_ASSERT, _CRTDBG_MODE_FILE | _CRTDBG_MODE_DEBUG);
_CrtSetReportFile( _CRT_ASSERT, _CRTDBG_FILE_STDERR );
_CrtSetReportMode( _CRT_ASSERT, _CRTDBG_MODE_FILE | _CRTDBG_MODE_DEBUG);
_CrtSetReportFile( _CRT_ASSERT, _CRTDBG_FILE_STDERR );
_CrtSetReportMode( _CRT_ASSERT, _CRTDBG_MODE_FILE | _CRTDBG_MODE_DEBUG);
_CrtSetReportFile( _CRT_ASSERT, _CRTDBG_FILE_STDERR );
Of course we can enable these popups by adding a new program
option, but I launching the tests from IDE most of time and
it's not convinient to specify the program options manually
for every test (but I found that I can use the RWSTD_TESTOPTS
environment variable for this :) ).
If you're launching an individual test from the IDE with F5 or CTRL+F5 you
would have to specify the --compat command line option (or the hypothetical
--no-popups option) somewhere to disable the CRT error dialog, so this should
not be a problem. The CRT error dialogs should continue appear as it always has
in this situation.
If you are using the runtests project from within the IDE, I believe that it
sets the --compat flag automatically for each executable that is ran from
runall.wsh. So you would not see the CRT error messages and you would need to
find a way to disable the command line option then.
So what I'm saying here is that we should consider adding a new option
--no-popups that calls _CrtSetReportMode(), _CrtSetReportFile() and
SetDebugMode(). That option should probably be passed to the exec utility to
pass on to each test when running the nightly builds. Adding this option should
have no impact on the behavior you experience when running tests from the IDE
today.
Travis