Re: Multiple declaration support in HP-FORTRAN compiler

On 2007-01-05 10:07:00 -0400, "Steve Lionel" <steve.lionel@xxxxxxxxx> said:

Tim Prince wrote:

If a customer with enough clout demands that a compiler accept source
code as is, it becomes so.

Yes indeed, but in this case, despite repeated requests over the years,
we have said "NO" on this one. I don't understand how such code comes
to be written in the first place.

Steve

If it were program generated it is easy to imagine various components of
the generation process independently adding declarations.

A more common symptom of program generated code is unused identifiers.

Not having to do multiple passes, or otherwise buffer the output, makes the
program generators much simpler. Although writing a final cleanup to toss
the duplicates is only a bit more complicated that a fancy pretty printer.

Having been there my observation is that if you fail to take the bit of
extra trouble on the way it becomes a bother to add it later. (That applies
to a lot of things like quality, portability, ...). Adding the final cleanup
postprocessor sounds too much like admitting it was done wrong in he beginning.

I expect that compiler writers must "love" program generated code as it tends
to require giant symbol tables, have huge subroutines and other strangenesses
that do not arise on hand crafted stuff.










.

Relevant Pages