Ryan Hill <dirtyepic@xxxxxxxxxx> writes:
> On Mon, 4 Apr 2011 18:09:21 -0600
> Ryan Hill <dirtyepic@xxxxxxxxxx> wrote:
>
>> Before 4.6, you could write a spec rule for a preprocessor definition using
>> its joined non-canonical form. Eg.
>>
>> %{!D_FORTIFY_SOURCE:%{!D_FORTIFY_SOURCE=*:%{!U_FORTIFY_SOURCE:-D_FORTIFY_SOURCE=2}}}
>>
>> Several distros used this rule to enable -D_FORTIFY_SOURCE=2 by default.
>>
>> After 4.6, preprocessor definitions are passed in their separated, canonical,
>> form (see PR #47236) and the rule above stops working. After a couple hours
>> of experimenting, I can't find any way to write a rule that works with the
>> new form. Shouldn't the joined form still be usable, even if the new default
>> is separate? Am I missing something obvious? I know you can use whitespace
>> and * in a spec, but apparently not in the middle of a switch name.
>
> Could anyone tell me if it would be considered a bug that we can't write spec
> rules matching options having both separate and joined forms any more?
Yes, I think this is an omission in the spec language. Please file a
request at http://gcc.gnu.org/bugzilla/ to add this facility.
I think that your spec would fail even with earlier versions of gcc if
the user wrote -D _FORTIFY_SOURCE rather than -D_FORTIFY_SOURCE.
Indeed, Ubuntu gcc does in fact give an error if I write -D
_FORTIFY_SOURCE, although should be supported. So I think this change
to gcc is not clearly a bug; it is a clarification which should help
ensure that specs won't sometimes work and sometimes fail.
Ian
