🚀 go-pugleaf

Dean Schulze (schulze@lpl.arizona.edu) wrote:
:     First of all, thanks for taking the time to look into this problem.

You're welcome.  Like you I believe that the state of gcc is of huge
importance to Linux's status as a development environment.

:     Someone else responded via e-mail that he got newmat08 to compile
: with gcc 2.7.0 by turning off exception handling as you did and fixing
: some scoping problems.  gcc 2.7.2 generated a message for me to compile
: with -fhandle-exceptions (or something like that) and that solved the
: compile time exception problem for me. 

That's great.  The exception handling by gcc is rather new, I believe.
If it is not yet perfect then I would agree that it is a shame that there
is no 100% reliable gcc distributed with Linux distributuions.  But, then,
we can always not compile with -fhandle-exceptions.  At least in the case of
newmat08 the author provided a documented alternative.  The gcc development
people seem to be trying to implement many changes.  Some of these are
to enable new capabilities (exception handling) and some are to make gcc/g++
adhere to ANSI standards.  I think that discussion of problems arising from
these two areas could be discussed separately(?).

: ...so that someone can offer a reliable commercial alternative.  Breaking
: lots of existing code is a bad thing to do. 

My point before, about, say, LAPACK++, was that lots of existing code is
sometimes broken by changes to the ANSI standard(s).  That gcc development
attempts to keep up with this (rather than allow legacy code to keep on
being compilable) is in opposition to your blanket statement here.  The
gcc people clearly believe that adhering to ANSI standards is more important
than allowing all people's code to keep on working.

What we have not determined (and someone may) is whether newmat08 breaks
because of bad code or because of a bug in gcc-2.7.2's exception handling.
My point is that the fact that gcc-2.7.2 breaks newmat08 (which worked on
an earlier gcc, 2.6.3) doesn't strictly imply that the problem is gcc's.
It could be, but the mere act of breaking "old" code doesn't make it so.

:     This is a non-sequitur.  I recall nothing in the newmat08 documentation
: about problems optimizing under certain platforms.

Precisely, there is nothing in the newmat08 documentation hinting that his
(or gcc's internal) code will work with optimizing.  So
why ought one to complain if it doesn't?

:     I just want reliable tools, and so far it looks like gcc 2.7.2 is
: unreliable.

You could always revert back to gcc-2.6.3, or even gcc-2.5.8, if you want
to be able to compile all that older C/C++ non-(current-)ANSI code.  It's
probably available at the GNU repository at prep.ai.mit.edu, and you could
even run them all side-by-side with a little fiddling.

Dave Linder
dwl@apmaths.uwo.ca
http://pineapple.apmaths.uwo.ca/~dwl

% ./configure --enable-humour
checking for -lICE... no