This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: RFA: fix for mudflap 19319, tree_addressable aggregate returns

On Mon, 10 Jan 2005, Frank Ch. Eigler wrote:

Hi -

dberlin wrote:

Uh, if only mudflap needs this, shouldn't we only do this when
flag_mudflap is true?

That's possible, but I wanted to minimize the code path differences, and wondered what other worms peek out from under the rocks, if the somewhat counterintuitive historical setting of the flag was reexamined.

Well, as i learned the hard way in a few weeks, the front ends set this flag if and only if the address is actually taken in the code, regardless of if the address of the thing could be taken (for example, if it's a global) somewhere we can't see.

There are also many places where the address is implicitly taken during rtl generation, but we don't set TREE_ADDRESSABLE.

If mudflap needs all of those changed, i'd much rather see that handled under flag_mudflap, or the mudflap pass to add them.

mudflap isn't the common case, so i'd also rather see us not change a common code path (set addressable then wait for alias1 to discover and clear them) so that things are slightly faster on the uncommon one, especially if it is likely you are going to have to add more of these types of changes to catch all the other implicit address taking cases.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]