This clarifies a lot. I agree with Jan and his patch is ok IMO.
Shall I update the manual with this information before we go release?
PS: I've also looked at this new bug 1723128 which is (partly) about a
return-use-only (ruonly) bit. I don't have a solution yet. But it is a
regression from 2.6.0 and kind of critical IMO. Please advise.
> -------- Original-Nachricht --------
> Betreff: Re: sdccman.lyx (checking in later today)
> Datum: Mon, 16 Apr 2007 15:38:12 +0200
> Von: Jan Waclawek <wek@...>
> Organisation: EFTON sro.
> An: frieder.ferlemann@...
> I thought more about this topic.
>>Note, I didn't commit the part 3.12.3 about the inline labels.
>>I think it's not the $ that makes the label local, instead I
>>think it's the use of a single ':' as opposed to '::' or a .globl
> There are 3 levels of "locality" of labels in this - rather strange -
> - global labels (with ::, == or declared via .globl) (see chapter 220.127.116.11
> in /doc/as/asmlnk.doc) - these can be referenced from
> outside the given file via the linker; this is equivalent to declaring a
> symbol PUBLIC in other assemblers
> - labels (just this simple, plain :-) ) - not "global" nor "local", with
> the scope within a .asm file
> - local labels - of the form NNNN$: - (see chapter 1.3.3 in
> /doc/as/asmlnk.doc - that needs a correction too, the number NNNN
> can be any integer) - their scope is limited to in between two "normal"
> labels - note that it's not '$', but the leading number
> AND '$' which makes these labels "local"
>>As can be seen here:
>>labels like "codeptr$" are no problems within inline assembler.
> So, I still believe everything I wrote in that chapter (3.12.3) holds
> Can you please review it once again.