[Graph] What to do about broken compilers?
The graph library has always been in a "barely usable" support on certain compilers (Borland <= 5.6.4 and MSVC 6.0, in particular). The regression tests have always been a mess---Boost 1.31.0 has very few passes for these compilers, and 1.32.0 isn't looking any better. However, the library is somewhat usable, if you stick to a very small subset and don't mind working around INTERNAL COMPILER ERRORs all of the time. What to do? I'm marking the library "unusable" on both compilers, so that we can concentrate our efforts on those compilers that might actually work. Complain loudly, quickly, and preferably with patches if you think this is the wrong approach. Doug
On Sun, 27 Jun 2004 14:03:15 -0500, Doug Gregor wrote
The graph library has always been in a "barely usable" support on certain compilers (Borland <= 5.6.4 and MSVC 6.0, in particular). The regression tests have always been a mess---Boost 1.31.0 has very few passes for these compilers, and 1.32.0 isn't looking any better. However, the library is somewhat usable, if you stick to a very small subset and don't mind working around INTERNAL COMPILER ERRORs all of the time. What to do?
I'm marking the library "unusable" on both compilers, so that we can concentrate our efforts on those compilers that might actually work. Complain loudly, quickly, and preferably with patches if you think this is the wrong approach.
From my perspective this is the correct approach. I'm starting to think that this is the last release I'm going to pay any attention to VC6 regressions with date-time. With Borland 5.6.4 it might be 1.32. It's just taking too much effort to support these old relics...
Jeff
Doug Gregor <dgregor@cs.indiana.edu> writes:
The graph library has always been in a "barely usable" support on certain compilers (Borland <= 5.6.4 and MSVC 6.0, in particular). The regression tests have always been a mess---Boost 1.31.0 has very few passes for these compilers, and 1.32.0 isn't looking any better. However, the library is somewhat usable, if you stick to a very small subset and don't mind working around INTERNAL COMPILER ERRORs all of the time. What to do?
I'm marking the library "unusable" on both compilers, so that we can concentrate our efforts on those compilers that might actually work. Complain loudly, quickly, and preferably with patches if you think this is the wrong approach.
Well, Boost.Python supports vc6 and uses Boost.Graph to do so. If you mark the library unusable, will all effort to keep the parts used by Boost.Python working with vc6 be stopped? -- Dave Abrahams Boost Consulting http://www.boost-consulting.com
On Jun 27, 2004, at 3:40 PM, David Abrahams wrote:
Doug Gregor <dgregor@cs.indiana.edu> writes:
The graph library has always been in a "barely usable" support on certain compilers (Borland <= 5.6.4 and MSVC 6.0, in particular). The regression tests have always been a mess---Boost 1.31.0 has very few passes for these compilers, and 1.32.0 isn't looking any better. However, the library is somewhat usable, if you stick to a very small subset and don't mind working around INTERNAL COMPILER ERRORs all of the time. What to do?
I'm marking the library "unusable" on both compilers, so that we can concentrate our efforts on those compilers that might actually work. Complain loudly, quickly, and preferably with patches if you think this is the wrong approach.
Well, Boost.Python supports vc6 and uses Boost.Graph to do so. If you mark the library unusable, will all effort to keep the parts used by Boost.Python working with vc6 be stopped?
No. As it stands, the Boost.Python tests that rely on Boost.Graph are the only tests of the BGL that actually tell us anything meaningful about VC6. Very few of the actual BGL tests pass on that compiler; even the "simple" things (BFS, DFS) aren't really tested because the compiler ICEs. So marking the lib "unusable" will keep us from expending a huge amount of effort to get these tests up running and will discourage users from trying something that will inevitably drive them insane; we'll still maintain enough compatibility for Boost.Python to work. I'd like to hear Jeremy's views on VC6 support as well... Doug
Hi All, I agree with Doug. We should still support Boost.Python for vc6 but otherwise mark vc6.0 as unusable. The BGL is just too fragile on that compiler. Perhaps someone could write down what Boost.Python uses so we know which parts of the BGL need to stay vc6 compatible. Cheers, Jeremy On Jun 27, 2004, at 4:18 PM, Doug Gregor wrote:
On Jun 27, 2004, at 3:40 PM, David Abrahams wrote:
Doug Gregor <dgregor@cs.indiana.edu> writes:
The graph library has always been in a "barely usable" support on certain compilers (Borland <= 5.6.4 and MSVC 6.0, in particular). The regression tests have always been a mess---Boost 1.31.0 has very few passes for these compilers, and 1.32.0 isn't looking any better. However, the library is somewhat usable, if you stick to a very small subset and don't mind working around INTERNAL COMPILER ERRORs all of the time. What to do?
I'm marking the library "unusable" on both compilers, so that we can concentrate our efforts on those compilers that might actually work. Complain loudly, quickly, and preferably with patches if you think this is the wrong approach.
Well, Boost.Python supports vc6 and uses Boost.Graph to do so. If you mark the library unusable, will all effort to keep the parts used by Boost.Python working with vc6 be stopped?
No.
As it stands, the Boost.Python tests that rely on Boost.Graph are the only tests of the BGL that actually tell us anything meaningful about VC6. Very few of the actual BGL tests pass on that compiler; even the "simple" things (BFS, DFS) aren't really tested because the compiler ICEs. So marking the lib "unusable" will keep us from expending a huge amount of effort to get these tests up running and will discourage users from trying something that will inevitably drive them insane; we'll still maintain enough compatibility for Boost.Python to work.
I'd like to hear Jeremy's views on VC6 support as well...
Doug
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
_______________________________________________ Jeremy Siek <jsiek@osl.iu.edu> http://www.osl.iu.edu/~jsiek Ph.D. Candidate, Indiana University Bloomington C++ Booster (http://www.boost.org) _______________________________________________
Jeremy Graham Siek <jsiek@osl.iu.edu> writes:
Hi All,
I agree with Doug. We should still support Boost.Python for vc6 but otherwise mark vc6.0 as unusable. The BGL is just too fragile on that compiler.
Perhaps someone could write down what Boost.Python uses so we know which parts of the BGL need to stay vc6 compatible.
See libs/python/src/object/inheritance.cpp. -- Dave Abrahams Boost Consulting http://www.boost-consulting.com
participants (4)
-
David Abrahams -
Doug Gregor -
Jeff Garland -
Jeremy Graham Siek