Re: [boost] gcc: timer patch, and ios_base workarounds

From: Beman Dawes (bdawes_at_[hidden])
Date: 2001-04-15 19:46:55

At 10:51 AM 4/14/2001, Greg Chicares wrote:

 >status/cs-win32.html says gcc fails the test in libs/timer/timer_test.cpp 
.
 >
 >With these two patches, it works[1] with gcc-2.95.2-1[2]:
 >
 >Patch one:
 >
 >boost/timer.hpp
 >- #include <limits>
 >+ #include <boost/pending/limits.hpp>
 >
 >I believe John Maddock already did this on 2001-04-01
 >and replaced boost/pending/limits.hpp with boost/limits.hpp.

Yes, thanks.

 >Patch two:
 >
 >boost/progress.hpp
 >  #include <iostream>
 >+ #if defined(__GNUC__) && __GNUC__ == 2 && __GNUC_MINOR__ <= 95
 >+   namespace std { class ios_base : public ios{}; }
 >+ #endif // defined(__GNUC__) && __GNUC__ == 2 && __GNUC_MINOR__ <= 95
 >  namespace boost {
 >
 >Is this preferable to the workaround
 >  std::istream::fmtflags f_;      // old GNU c++ lib has no ios_base
 >in random.hpp?
 >
 >Either of these hacks could be put in a header /boost/iostream.hpp,
 >which could then be used in place of <iostream> throughout boost;
 >but I imagine that's objectionable for a single-compiler workaround.
 >
 >Another option is to use either hack wherever ios_base is used:
 >  progress.hpp
 >  rational.hpp [already uses std::istream::fmtflags]
 >  libs/regex/src/cpp_regex_traits.cpp
 >I think that's the best way.

See Jens' reply.

 >Leaving rational.hpp as it is and ignoring the other two files
 >that use ios_base is another option that doesn't seem good.
 >
 >[1] I get
 >  t1 elapsed: 1.05
 >  t2 elapsed: 1.05
 >  t1 and t2 should report the same times (very approximately 2 seconds).
 >Is 1.05 close enough to 2? With borland C++ 5.5 I get
 >  t1 elapsed: 1.339
 >  t2 elapsed: 1.339

Should be OK.  To tell the truth, I never trust any timer until I've run it 
many times against a stopwatch.

 >[2] The official gcc / win32 test platform seems to be cygwin;
 >I used the mingw32 port (www.mingw.org). Is there any interest
 >in reporting mingw32 test results as well in status/cs-win32.html?

What happens when gcc 3.0 ships?  Will we still have these variants?

Thanks,

--Beman