addresses of labels don't work with today's CVS source

[Mark Mitchell]

>>>>> "Geoff" == Geoff Keating <geoffk@cygnus.com> writes:

    g++.sum g++.oliva/thunk1.C gcc.sum
    gcc.c-torture/compile/920301-1.c, gcc.sum
    gcc.c-torture/compile/920428-3.c, gcc.sum
    gcc.c-torture/compile/941014-4.c, gcc.sum
    gcc.c-torture/compile/950613-1.c, gcc.sum
    gcc.c-torture/execute/920302-1.c gcc.sum
    gcc.c-torture/execute/920501-4.c gcc.sum
    gcc.c-torture/execute/comp-goto-1.c

    were broken by:

    +Wed May  3 09:29:17 2000  Richard Kenner  <kenner@vlsi1.ultra.nyu.edu>
    +
    +	* expr.c (expand_expr, case COMPONENT_REF): Don't check for checking
    +	memory usage if not in a function.
    +	* varasm.c (initializer_constant_valid_p, case ADDR_EXPR): Only
    +	return address if static.
    +

That would indicate that this patch likely caused failures on targets
other than thunk-based targets.  I had thought perhaps the only
failures were in C++ on thunk-using targets -- which means that a
bootstrap/test on a non-thunks target might have passed cleanly.

Kenner, please test your work more carefully.  I think the rules for
checkins have been made abundantly clear at this point.  Although we
all occassionally forget, and we certainly all check in buggy changes
with some regularity, the number of problems that originate from your
checkins seems to me to be disproportionately high.  (In all fairness,
your changes are often of a fundamental nature which increases both
their risk factor and their benefit!)

Please indicate with each check-in which platform you used for
bootstrapping and testing.  (Jeff instituted this requirement a while
back, and I think it is a very good one.)  Doing that will give us all
confidence that you are indeed following the procedures.  (Perhaps you
*did* test this check-in, and perhaps it did pass.  That seems
unlikely -- but it could well be true!  In that case, I'm being
unfairly critical, and I apologize -- but it would help if I *knew* I
was wrong.)  The effort to add the lines to patch email should be
trivial compared with the effort used to generate the average patch.

By no means does this rule apply only to Kenner!  This rule should be
applied uniformly to all contributors to language-independent code.  I
will attempt to begin enforcing this rule a bit more stringently,
including with respect to my own checkins.

Thanks in advance,

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com