GCC infrastructure and non-GPLed optimizers

GCC infrastructure and non-GPLed optimizers

[S. Bosscher]

Hi,

Yesterday in the discussion about Geoff's "intermodule optimisation patch"
one of the reasons mentioned against the intermodule approach proposed by
Chris Lattner was that "(it) permits people to write non-GPLed optimisers
and code generators and still use GCC's parsers."
(http://gcc.gnu.org/ml/gcc-patches/2003-05/msg01679.html)

The optimization framework proposed by Chris is very similar to the
intermodule framework proposed by many papers, and in fact similar to what
existing compilers, like ORC/Open64, do with some success.  So if this
argument holds, then GCC could never have such a framework for intermodule
optimizers.
In a way, the argument blocks the development of GCC itself.  Maybe it
should be reconsidered.

So the question is: How relevant do you all think this argument still is
today, considering the facts that:
- Like Chris said, there already are backends
  available that write out enough information
  (XML, C backends) for a separate non-GPLed
  compiler with its own optimizers and backends.
- It does not look very difficult to modify the
  tree dumpers in the tree-ssa branch to dump
  tree-optimized, compilable C/C++ source code.

So anyone who would want to use GCC's parsers in combination with non-GPLed
optimizers/backends can already do so with only a few modifications.  Then
the argument seems to be quite irrelevant.

Thoughts?

Greetz
Steven


PS. Wasn't this argument also used in the past to block making the backend
library a shared library?