[boost] Re: checked_delete / CW8
Daniel Frey <[EMAIL PROTECTED]> writes: > David Abrahams wrote: >> Daniel Frey <[EMAIL PROTECTED]> writes: >> >>>Peter Dimov wrote: >>> It is worth mentioning that this is a confirmed bug in CW with -iso_templates on (unreferenced typedefs are incorrectly optimized out at definition time). Masking compiler bugs in this way does not help compiler writers who use Boost as a test suite. >>> >>>The bug is already fixed for the CW9 and all other compilers weren't >>>affected anyway, so I thought it's pointless to add >>> >>>#if defined(__MWERKS__) && __MWERKS__ < 0x3200 >>> >>> to protect the new code. >> You should use BOOST_WORKAROUND(__MWERKS__, BOOST_TESTED_AT(0x3003)) >> in order to be friendly to compiler writers who want to use Boost for >> testing. > > I know, but this is exactly what Peter meant by "Dependencies. I hate > dependencies." IIUC and I tried to respect that. I think that's a bit extreme. If people refuse to use BOOST_WORKAROUND because it creates a dependency on a single header file, it will undermine what we are trying to achieve with it. -- Dave Abrahams Boost Consulting www.boost-consulting.com ___ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
[boost] Re: checked_delete / CW8
David Abrahams wrote: Daniel Frey <[EMAIL PROTECTED]> writes: Peter Dimov wrote: It is worth mentioning that this is a confirmed bug in CW with -iso_templates on (unreferenced typedefs are incorrectly optimized out at definition time). Masking compiler bugs in this way does not help compiler writers who use Boost as a test suite. The bug is already fixed for the CW9 and all other compilers weren't affected anyway, so I thought it's pointless to add #if defined(__MWERKS__) && __MWERKS__ < 0x3200 to protect the new code. You should use BOOST_WORKAROUND(__MWERKS__, BOOST_TESTED_AT(0x3003)) in order to be friendly to compiler writers who want to use Boost for testing. I know, but this is exactly what Peter meant by "Dependencies. I hate dependencies." IIUC and I tried to respect that. It's why I used a one-size-fits-all approach. Regards, Daniel -- Daniel Frey aixigo AG - financial training, research and technology Schloß-Rahe-Straße 15, 52072 Aachen, Germany fon: +49 (0)241 936737-42, fax: +49 (0)241 936737-99 eMail: [EMAIL PROTECTED], web: http://www.aixigo.de ___ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
Re: [boost] Re: checked_delete / CW8
Daniel Frey wrote: > Peter Dimov wrote: >> It is worth mentioning that this is a confirmed bug in CW >> with -iso_templates on (unreferenced typedefs are incorrectly >> optimized out at definition time). Masking compiler bugs in this way >> does not help compiler writers who use Boost as a test suite. > > The bug is already fixed for the CW9 and all other compilers weren't > affected anyway, so I thought it's pointless to add > > #if defined(__MWERKS__) && __MWERKS__ < 0x3200 > > to protect the new code. Also, you haven't done that for the fix for > the Intel-compiler yourself, so why do you mention it now? Because the two cases are different. In the Intel case, the compiler deliberately allows certain (not really uncommon) broken code in , presumably in order to compile some broken system header. In the Metrowerks case our tests exposed a compiler bug that wasn't deliberately coded in. Had our headers been "fixed" beforehand, the compiler bug wouldn't have been caught. Thanks for listening. ___ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
[boost] Re: checked_delete / CW8
Daniel Frey <[EMAIL PROTECTED]> writes: > Peter Dimov wrote: >> It is worth mentioning that this is a confirmed bug in CW >> with -iso_templates on (unreferenced typedefs are incorrectly optimized out >> at definition time). Masking compiler bugs in this way does not help >> compiler writers who use Boost as a test suite. > > The bug is already fixed for the CW9 and all other compilers weren't > affected anyway, so I thought it's pointless to add > > #if defined(__MWERKS__) && __MWERKS__ < 0x3200 > > to protect the new code. You should use BOOST_WORKAROUND(__MWERKS__, BOOST_TESTED_AT(0x3003)) in order to be friendly to compiler writers who want to use Boost for testing. -- Dave Abrahams Boost Consulting www.boost-consulting.com ___ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
[boost] Re: checked_delete / CW8
Peter Dimov wrote: It is worth mentioning that this is a confirmed bug in CW with -iso_templates on (unreferenced typedefs are incorrectly optimized out at definition time). Masking compiler bugs in this way does not help compiler writers who use Boost as a test suite. The bug is already fixed for the CW9 and all other compilers weren't affected anyway, so I thought it's pointless to add #if defined(__MWERKS__) && __MWERKS__ < 0x3200 to protect the new code. Also, you haven't done that for the fix for the Intel-compiler yourself, so why do you mention it now? And AFAICS if we would really be serious about the regression testing for compiler vendors, this needs to be pulled to the config-system anyway. Remember who said: "Dependencies. I hate dependencies." :) Regards, Daniel -- Daniel Frey aixigo AG - financial training, research and technology Schloß-Rahe-Straße 15, 52072 Aachen, Germany fon: +49 (0)241 936737-42, fax: +49 (0)241 936737-99 eMail: [EMAIL PROTECTED], web: http://www.aixigo.de ___ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
[boost] Re: checked_delete / CW8
Howard Hinnant <[EMAIL PROTECTED]> writes: > On Saturday, August 16, 2003, at 8:06 AM, David Abrahams wrote: > >>> Thanks. I'll keep my fingers crossed... >> >> Still no joy :( > > Could you elaborate? Perhaps I could help. Oh, just miscellaneous other regressions which are stopping the release. If you'd like to help, that'd be great ;-> i-really-hope-i've-nailed-them-now-though-ly y'rs, dave -- Dave Abrahams Boost Consulting www.boost-consulting.com ___ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
Re: [boost] Re: checked_delete / CW8
On Saturday, August 16, 2003, at 8:06 AM, David Abrahams wrote: Thanks. I'll keep my fingers crossed... Still no joy :( Could you elaborate? Perhaps I could help. -Howard ___ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
[boost] Re: checked_delete / CW8
Daniel Frey <[EMAIL PROTECTED]> writes: > On Sat, 16 Aug 2003 04:11:09 +0200, David Abrahams wrote: > >> David Abrahams <[EMAIL PROTECTED]> writes: >> >>> Daniel Frey <[EMAIL PROTECTED]> writes: >>> Hi Dave, I checked in a fix for checked_delete.hpp for the Metrowerks CW8 to CVS HEAD. It was created in cooperation with Howard and I'm positiv that it's a good one-size-fits-all solution. I don't know about your shedule for 1.30.2, but you might want to consider merging it to RC_1_30_0. I will not push this as I don't want to delay 1.30.2, having it in 1.31.0 is fine, too. >>> >>> Hmm, I use Pro8.3 all the time and have never seen a need for a patch >>> to checked_delete. Ah, the regressions were in expected-failure > tests? > > The regression depends on the compiler flags. It occurs with "-iso_templates > on" which - according to Howard - is a good idea to use. Personally, I > have no clue what it's good for... :) It enables conformance ;-) Of course all of my compiles and the Boost regression tests are run using -iso_templates on. >> OK, I like your patch and I've applied it in the branch. I'm going to >> release tomorrow morning after the meta-comm regressions have run again. > > Thanks. I'll keep my fingers crossed... Still no joy :( -- Dave Abrahams Boost Consulting www.boost-consulting.com ___ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
[boost] Re: checked_delete / CW8
On Sat, 16 Aug 2003 04:11:09 +0200, David Abrahams wrote: > David Abrahams <[EMAIL PROTECTED]> writes: > >> Daniel Frey <[EMAIL PROTECTED]> writes: >> >>> Hi Dave, >>> >>> I checked in a fix for checked_delete.hpp for the Metrowerks CW8 to >>> CVS HEAD. It was created in cooperation with Howard and I'm positiv >>> that it's a good one-size-fits-all solution. I don't know about your >>> shedule for 1.30.2, but you might want to consider merging it to >>> RC_1_30_0. I will not push this as I don't want to delay 1.30.2, >>> having it in 1.31.0 is fine, too. >> >> Hmm, I use Pro8.3 all the time and have never seen a need for a patch >> to checked_delete. Ah, the regressions were in expected-failure tests? The regression depends on the compiler flags. It occurs with "-iso_templates on" which - according to Howard - is a good idea to use. Personally, I have no clue what it's good for... :) > OK, I like your patch and I've applied it in the branch. I'm going to > release tomorrow morning after the meta-comm regressions have run again. Thanks. I'll keep my fingers crossed... Regards, Daniel ___ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
[boost] Re: checked_delete / CW8
David Abrahams <[EMAIL PROTECTED]> writes: > Daniel Frey <[EMAIL PROTECTED]> writes: > >> Hi Dave, >> >> I checked in a fix for checked_delete.hpp for the Metrowerks CW8 to CVS >> HEAD. It was created in cooperation with Howard and I'm positiv that it's >> a good one-size-fits-all solution. I don't know about your shedule for >> 1.30.2, but you might want to consider merging it to RC_1_30_0. I will not >> push this as I don't want to delay 1.30.2, having it in 1.31.0 is fine, >> too. > > Hmm, I use Pro8.3 all the time and have never seen a need for a patch > to checked_delete. Ah, the regressions were in expected-failure tests? OK, I like your patch and I've applied it in the branch. I'm going to release tomorrow morning after the meta-comm regressions have run again. -- Dave Abrahams Boost Consulting www.boost-consulting.com ___ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
[boost] Re: checked_delete / CW8
Daniel Frey <[EMAIL PROTECTED]> writes: > Hi Dave, > > I checked in a fix for checked_delete.hpp for the Metrowerks CW8 to CVS > HEAD. It was created in cooperation with Howard and I'm positiv that it's > a good one-size-fits-all solution. I don't know about your shedule for > 1.30.2, but you might want to consider merging it to RC_1_30_0. I will not > push this as I don't want to delay 1.30.2, having it in 1.31.0 is fine, > too. Hmm, I use Pro8.3 all the time and have never seen a need for a patch to checked_delete. Ah, the regressions were in expected-failure tests? -- Dave Abrahams Boost Consulting www.boost-consulting.com ___ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost